Review
Abstract
Background: Emerging evidence underscores the potential application of artificial intelligence (AI) in discovering noninvasive blood biomarkers. However, the diagnostic value of AI-derived blood biomarkers for ovarian cancer (OC) remains inconsistent.
Objective: We aimed to evaluate the research quality and the validity of AI-based blood biomarkers in OC diagnosis.
Methods: A systematic search was performed in the MEDLINE, Embase, IEEE Xplore, PubMed, Web of Science, and the Cochrane Library databases. Studies examining the diagnostic accuracy of AI in discovering OC blood biomarkers were identified. The risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies–AI tool. Pooled sensitivity, specificity, and area under the curve (AUC) were estimated using a bivariate model for the diagnostic meta-analysis.
Results: A total of 40 studies were ultimately included. Most (n=31, 78%) included studies were evaluated as low risk of bias. Overall, the pooled sensitivity, specificity, and AUC were 85% (95% CI 83%-87%), 91% (95% CI 90%-92%), and 0.95 (95% CI 0.92-0.96), respectively. For contingency tables with the highest accuracy, the pooled sensitivity, specificity, and AUC were 95% (95% CI 90%-97%), 97% (95% CI 95%-98%), and 0.99 (95% CI 0.98-1.00), respectively. Stratification by AI algorithms revealed higher sensitivity and specificity in studies using machine learning (sensitivity=85% and specificity=92%) compared to those using deep learning (sensitivity=77% and specificity=85%). In addition, studies using serum reported substantially higher sensitivity (94%) and specificity (96%) than those using plasma (sensitivity=83% and specificity=91%). Stratification by external validation demonstrated significantly higher specificity in studies with external validation (specificity=94%) compared to those without external validation (specificity=89%), while the reverse was observed for sensitivity (74% vs 90%). No publication bias was detected in this meta-analysis.
Conclusions: AI algorithms demonstrate satisfactory performance in the diagnosis of OC using blood biomarkers and are anticipated to become an effective diagnostic modality in the future, potentially avoiding unnecessary surgeries. Future research is warranted to incorporate external validation into AI diagnostic models, as well as to prioritize the adoption of deep learning methodologies.
Trial Registration: PROSPERO CRD42023481232; https://www.crd.york.ac.uk/PROSPERO/view/CRD42023481232
doi:10.2196/67922
Keywords
Introduction
Ovarian Cancer Diagnosis: Status and Demands
Ovarian cancer (OC) is the deadliest gynecologic malignancy, characterized by nonspecific symptoms that often remain undetected until the disease has progressed [,]. The conventional diagnosis of OC principally depends on imaging techniques (encompassing ultrasound, computed tomography, and magnetic resonance imaging); serum biomarkers (such as cancer antigen 125, carcinoembryonic antigen, and human epididymis protein 4); along with the invasive procedure (histological biopsy) [,]. However, the sensitivity and specificity of imaging techniques and biomarkers are restricted []. Furthermore, the histopathological test is inherently invasive []. Therefore, there is an urgent need for more accurate, noninvasive, and reliable diagnostic methods.
Potential of Noninvasive Blood Markers for OC
Minimally invasive diagnostic procedures, particularly the use of blood samples, are among the foremost common methods of detection []. Moreover, patients are generally more willing to undergo blood tests, leading to higher compliance rates []. Blood contains a rich repertoire of biomolecules, including proteins, nucleic acids, and metabolites, which can potentially serve as indicators of OC diagnosis [-]. The development of omics has opened new doors for biomarker discovery. The genomics, proteomics, and metabolomics of blood samples can provide a wealth of information about the molecular changes in cancer []. For instance, Ke et al [] systematically investigated OC metabolism through the metabolic profiling of 448 plasma samples. The analysis of dysregulated metabolic pathways extends our current understanding of OC metabolism. Similarly, Dhar et al [] applied glycoproteomics to serum of women with OC or benign pelvic masses and healthy controls and analyzed glycosylation patterns in serum markers and supported the hypothesis that blood glycoproteomic profiling can be used for OC diagnosis and staging. Notably, the exponential growth of multiomics data has presented a major challenge that surpasses traditional analytic capabilities [,]. Fortunately, artificial intelligence (AI) algorithms offer a solution [,]. AI can manage complex datasets and spot hidden patterns and potential biomarkers, enabling more accurate OC diagnosis and personalized treatment.
Application of AI in OC Blood Markers
AI, particularly machine learning (ML) and deep learning (DL), has attracted increasing attention in medical research due to its capability to analyze large biomedical datasets [-]. AI-driven models have emerged as a promising tool for developing predictive models for OC by analyzing complex and multidimensional datasets to uncover biomarkers. For instance, a multicenter retrospective study screened 52 features from laboratory tests in blood samples to build an ML model. Integrating 20 base AI models, it performed well internally and externally, outperforming the CA125 and HE4 biomarkers in identifying OC []. In addition, a blood-based metabolite panel demonstrated independent predictive ability and complemented the risk of ovarian malignancy algorithm for distinguishing early-stage OC from benign disease to better inform clinical decision-making []. Despite such encouraging results, the related results are scattered. Whether the application of AI in OC blood biomarker research can significantly improve diagnostic accuracy remains controversial.
Purpose of This Study (AI and OC Blood Markers)
Therefore, there is a crucial need for a high-quality synthesis of the available evidence. The purpose of this study is to provide a systematic overview of the diagnostic accuracy of AI techniques in identifying OC blood markers as well as to elucidate their applicability, potential, and limitations.
Methods
Protocol Registration and Study Design
This meta-analysis was conducted in adherence to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist () and MOOSE (Meta-Analysis of Observational Studies in Epidemiology) reporting guidelines [,]. The protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO; CRD42023481232).
Literature Search and Eligibility Criteria
A comprehensive search of the MEDLINE, Embase, IEEE Xplore, PubMed, Web of Science, and the Cochrane Library databases was carried out from inception to January 16, 2024. Detailed search strategies are summarized in . Two independent investigators (MQJ and HLX) assessed the records after removing the duplicates at the title and abstract level, and finally at the full-text level, according to the inclusion and exclusion criteria (
Textbox 1). Two investigators (MQJ and XYL) independently appraised the articles for eligibility. An inconsistency in selection was reconciled through discussion with a third independent investigator (HLX).
Inclusion criteria
- Population: adults (aged ≥18 years) with potential ovarian cancer lesions
- Intervention: artificial intelligence–assisted blood test
- Comparison: histopathology or other reliable clinical diagnosis
- Outcomes: diagnostic performance (ie, sensitivity and specificity or detailed information that could extract or construct 2×2 contingency tables)
- Studies: original articles (ie, observational studies or randomized controlled trials)
- Language: English
Exclusion criteria
- Population: nonhuman samples or other diseases
- Intervention: nonblood samples or no artificial intelligence algorithms
- Comparison: no control group
- Outcomes: no diagnostic performance data (ie, 2×2 contingency tables cannot be extracted or constructed from the provided data)
- Studies: records, such as letters, conference abstracts, case reports, or review articles
- Language: non-English
Data Extraction
Data were independently extracted by 2 investigators (QPM and XYL) using a predefined data extraction sheet, with any discrepancies resolved through the adjudication of a third investigator (HLX). Necessary data of 2×2 contingency tables that included true positives (TP), false positives, true negatives (TN), and false negatives were extracted. The details of the data are presented in [,,,-]. In cases where these values were not explicitly reported, values were calculated using descriptive statistics available in the study. For studies presenting multiple contingency tables, either for identical or disparate AI algorithms, each table was treated as an independent result [].
Study Quality Assessment
The risk of bias and concerns about the applicability of all included studies were assessed by 2 independent investigators (MQJ and HLX), using the Quality Assessment of Diagnostic Accuracy Studies–Artificial Intelligence (QUADAS-AI) criteria []. Conflicts were discussed and solved with a third investigator (XYL). The risk of bias assessment included 4 domains: patient selection, index test, reference standard, and flow and timing. For assessing clinical applicability, only the first 3 domains were evaluated [,]. In addition, the median was used as the threshold for determining the risk level of bias, with studies classified as low risk if ≥2 domains were deemed as low risk and high risk if <2 domains were considered low risk [].
Data Analysis
We used the bivariate diagnostic random effects model to compute the summary receiver operating characteristics to determine summary estimates of the sensitivity, specificity, and area under the curve (AUC) with their respective 95% CIs []. Sensitivity was defined as the probability of a person with OC having a positive test result, indicating the capacity of the index test to identify patients, considered by the equation: sensitivity = TP / (TP + false negatives). Specificity was the probability of a woman without OC having a negative test result, reflecting the test's ability to correctly identify OC-free individuals calculated by the equation: specificity = TN / (false positives + TN) [,]. The performance of the test can also be assessed using the AUC. This area may be interpreted as the probability that a random woman with OC has a higher value of the measurement than a random person without OC. In general, an AUC of >0.80 is considered good []. A perfect test would have an AUC of 1 and a useless test would have an AUC of 0.5 [].
Study heterogeneity was determined using the I2 statistic and the Q test []. When the same or different AI models were tested within the same article, the proposed model with the best accuracy was used for further meta-analysis []. Subgroup and regression analyses were performed to explore potential sources of heterogeneity. Sensitivity analysis was implemented via a qualitative systematic review of studies from which concatenated tables could not be extracted or constructed. Publication bias was assessed using funnel plot asymmetry test by Deeks et al [].
Subgroup analyses were performed according to the following: (1) AI algorithms (ML or DL), (2) external validation (yes or no), (3) levels of risk of bias (low or high), (4) year of publication (after or before 2022), (5) geographical distribution (Asia, North America, or Europe), (6) sample size (>300 or ≤300 as median), (7) blood sample type (serum or plasma), (8) biomarkers type (protein or mixed), and (9) number of modeling biomarkers (>8 or ≤8 as median).
The variability in sensitivity and specificity estimates was graphically represented through a cross-hairs plot, generated using R software (version 4.2.1; R Foundation for Statistical Computing) []. All other statistical analyses were conducted in Stata (version 17.0; StataCorp). The statistical significance was defined as P<.05.
Results
Study Selection and Characteristics of Eligible Studies
The database search identified 1566 records from which 604 duplicates were removed. We then performed the title and abstract screening of 962 records and subsequently a full-text evaluation of 55 records. Following the exclusion of 15 articles, as detailed in [-], a total of 40 studies were included in this meta-analysis (
Figure 1).
Most of the studies were performed with retrospectively (21/40, 52%) and prospectively (18/40, 45%) collected data, and one study collected both retrospective and prospective data (
Table 1). In total, 12% (5/40) of studies sourced their data from public databases. In terms of AI algorithm types, a total of 36 (90%) studies were classified as ML, whereas 10% (4/40) of studies were classified as DL. Most (36/40, 90%) of the studies validated their algorithms, while only some (7/40, 18%) studies carried out an external validation (Table 2). The blood samples were mainly serum (27/40, 68%), and the type of blood biomarkers was mainly protein (25/40, 62%; Table 3).
| Study | Study design | Data source | Selection criteria | Time frame | Age (y), mean or median | Sample size, n |
| Cai et al [], 2024 | Retrospective | Data from Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, central China; Women’s Hospital, School of Medicine, Zhejiang University, eastern China; and Sun Yat-Sen University Cancer Center, southern China | Patients with a history of other malignant cancers or precancers, pregnancy in the last 6 mo, or affected by HIV; not newly diagnosed in any of the 3 hospitals; individuals without any available laboratory tests were excluded | From January 2012 to April 2021 | Cancer: (53/51/56)a; control: (34/34/48)a | 10,992 (3007/5641/2344)a |
| Abuzinadah et al [], 2023 | Retrospective | Data from the Third Affiliated Hospital of Soochow University | All patients underwent postoperative case diagnosis, and none of them had received preoperative radiotherapy or chemotherapy | From July 2011 to July 2018 | NRb | 349 (244/105)c |
| Bifarin and Fernandez [], 2024 | Retrospective | Data from a serum lipidomic analysis of ovarian cancer patients of Korean descent | NR | NR | NR | 325 (227/98)c |
| Cameron et al [], 2023 | Retrospective | Data from the Welcome Trust Clinical Research Facility at the Western General Hospital, Edinburgh, the Emergency Medicine Research Group at the Edinburgh Royal Infirmary, the Beatson West of Scotland Cancer Centre in Glasgow, the University of Swansea, Royal Preston Hospital, and Manchester Cancer Research Centre | NR | NR | Ovary: 61; NCSd female individuals only: 56 | 385 (NR) |
| Chen et al [], 2023 | Retrospective | Data from the Department of Gynecology of Harbin Medical University Cancer Hospital, Gene Expression Omnibus database, UCSC Xena | Patients with a primary radiological diagnosis of ovarian tumor; newly diagnosed patients without any significant comorbidities or history of previous malignancies; willingness to participate in the study and provision of written informed consent | From December 2020 to July 2021 | NR | 44 (44)e |
| Dhar et al [], 2023 | Retrospective | Data from Indivumed (Hamburg, Germany) | NR | NR | NR | 351 (237/114)c |
| Hamidi et al [], 2023 | Retrospective | Data from the Gene Expression Omnibus database, a Japanese nationwide research project, and patients with cancer who were referred or admitted to the National Cancer Center Hospital | The serum samples for noncancer controls who had no history of cancer and no hospitalization during the previous 3 months were collected; patients with cancer who were treated with preoperative chemotherapy and radiotherapy before serum collection were excluded | NR | Internal set: 52 | 3411 (2156/3079/92/240)f |
| Lai et al [], 2023 | Retrospective | Data from clinical laboratory examination | NR | From January 2013 to October 2022 | 46.4 | 778 (545/233)g |
| Li et al [], 2024 | Retrospective | Data from the Cancer Hospital of the University of Chinese Academy of Sciences (Zhejiang Cancer Hospital) | No patients with ovarian cancer received chemotherapy, radiotherapy, or surgery, no healthy donors had a history of cancer before sample collection | NR | Stage 1: 53.6; stage 2: 55.8; stage 3 and 4: 54.9; control: 34.1 | 69 (NR) |
| Reilly et al [], 2023 | Retrospective and prospective | Data from multiple studies spanning multiple centers | Patient age ≥18 years; informed consent provided by the patient to participate in research; patient agreeable to phlebotomy; patient had a documented adnexal mass | NR | 47.5 | 2186 (NR) |
| Zhang et al [], 2023 | Retrospective | Data from patients who underwent physical examination at Chinese People’s Armed Police Force and First Medicine Center, People\'s Liberation Army General Hospital | Discharge diagnosis confirmed by clinical signs, imaging, and pathology for patients with gynecologic tumors and benign gynecologic diseases; if no histopathological examination was available, it was consistently confirmed by ≥2 types of imaging evidence, availability of laboratory test data at the time of first diagnosis, and blood collection before treatment | From January 2010 to June 2019 | Ovarian cancer: (52.11/52.69/55.81)h; NOMGTi: (54.66/52.66/53.95)h; BGDj: (44.30/46.33/39.21)h; Healthy control: (49.96/47.26/49.67)h | 1633(600/301)g |
| Ahamad et al [], 2022 | Retrospective | Data from Third Affiliated Hospital of Schow University | NR | From July 2017 to July 2018 | NR | 349 (85/21)c |
| Bahado-Singh et al [], 2022 | Prospective | Data from Oakland University William Beaumont School of Medicine | NR | NR | Cases: 66.2; control: 67.8 | 17 (NR) |
| Gupta et al [], 2022 | Retrospective | Data from 3 separate commercial biobanks: Dx Biosamples (San Diego, CA), Reprocell USA Inc (Beltsville, MD), and Fidelis Research AD (Sofa, Bulgaria) | NR | NR | NR | 1243 (681)k |
| Hinestrosa et al [], 2022 | Retrospective | Data from a commercial biorepository (ProteoGenex, Inglewood, CA, United States) | The control group has no known history of cancer, autoimmune diseases, or neurodegenerative disorders, nor any presence of diabetes mellitus | From January 2014 to September 2020 | 60 | 323 (216/107)c |
| Irajizad et al [], 2022 | Prospective | Data from Anderson Cancer Center and at the Fred Hutchinson Cancer Research Center | NR | NR | NR | 409 (108/118)c |
| Kim et al [], 2022 | Retrospective | Data from Oakland University William Beaumont School of Medicine | NR | NR | NR | 269 (215)e |
| Li et al [], 2022 | Prospective | Data from 3 institutions | Women diagnosed with benign, borderline, and malignant ovarian tumors | From December 2018 to January 2020 | NR | 362 (178/184)c |
| Pais et al [], 2022 | Retrospective | Data from a commercially stored biological sample biobank (Invent diagnostica, Berlin, Germany) | NR | NR | NR | 181 (NR) |
| Jeong et al [], 2021 | Retrospective | Data from Kangnam Sacred Heart Hospital | NR | From June 2014 to December 2020 | Cancer: 54; control: 49 | 730 (511/219)c |
| Lu et al [], 2020 | Prospective | Data from the Third Affiliated Hospital of Soochow University | None of the patients with ovarian cancer received preoperative chemotherapy or radiotherapy | From July 2011 to July 2018 | NR | 349(235/114)c |
| Banaei et al [], 2019 | Prospective | Data from the UMass Memorial Medical Center Chemotherapy Infusion Center and Gastroenterology Clinics and Innovative Research | NR | NR | NR | 20 (40/160)c |
| Whitwell et al [], 2018 | Retrospective | Data from a synthetic dataset modeled from the United Kingdom Collaborative Trial of Ovarian Cancer Screening | Trial participants at enrollment were postmenopausal women aged 50-74 y who had no family history of ovarian cancer | NR | NR | 89 (NR) |
| Ivanova et al [], 2016 | Retrospective | Data from the clinical diagnostic laboratory of the LLC LYTECH, the Blokhin Cancer Research Center of Russian Academy of Medical Sciences, the National Research Center of Coloproctology, the Moscow Dermatovenerologic Dispensary, clinical hospitals of Peoples’ Friendship University of Russia | NR | NR | Cancer: 52; control: 49 | 67 (NR) |
| Jiang et al [], 2013 | Prospective | Data from the affiliated hospital, Sun Yat-Sen University | NR | NR | 61.7 | 87 (51/36)l |
| Yang et al [], 2013 | Prospective | Data from the Peking University Third Hospital | NR | From January 2003 to December 2009 | Stage I/II: 54.8; stage III: 57.3; stage IV: 58.2; normal: 52.8; carcinoid: 51.6 | 246 (NR) |
| Shan et al [], 2012 | Prospective | Data from the Tampa, Florida metropolitan area | Women with a prior unilateral or bilateral oophorectomy were ineligible, as were women with a previous history of cancer. All patients underwent preoperative radiologic imaging, either by pelvic ultrasound, CT, or MRI. Only patients who underwent surgery based on clinical suspicion of ovarian cancer were eligible | NR | NR | 423 (NR) |
| Thakur et al [], 2011 | Prospective | Data from the FDA-NCI clinical proteomics program databank | NR | NR | NR | 216 (173/43)c |
| Donach et al [], 2010 | Prospective | Data from Padua Hospital (now the Veneto Oncology Institute) | NR | From 1999 to 2005 | 48 | 201 (NR) |
| Ziganshin et al [], 2008 | Prospective | Data from Byelorussian Oncology Center with patients with ovarian cancer and the Clinical Diagnostic Laboratory with clinically healthy women | NR | NR | Cancer: 51;control: 49 | 118 (NR) |
| Liu et al [], 2007 | Prospective | Data from Northwestern University, Johns Hopkins University in Baltimore, MD, and the University of Innsbruck, Austria | Normal samples were from patients who had 4-year follow-up examinations to ensure that they did not have cancer at the time the samples were taken | From 1999 to 2002 | NR | 563 (315/78/170)m |
| Zhang et al [], 2007 | Prospective | Data from the Duke University Medical Center, Durham, NC, St Bartholomew’s Hospital, London, United Kingdom, and the Groningen University Hospital, Groningen, Netherlands | NR | NR | NR | 468 (200/150)c |
| Chatterjee et al [], 2006 | Retrospective | Data from the Barbara Ann Karmanos Cancer Institute, the MD Anderson Cancer Center, Weill Medical College of Cornell University, Northwestern University Robert H Lurie Comprehensive Cancer Center, and the Gynecologic Oncology Group Tissue Bank | NR | NR | NR | 129 (85/44)c |
| Lin et al [], 2006 | Prospective | Data from the Tri-Service General Hospital, Taiwan, and Republic of China | Patients with any history of cancer, operations that had removed body organ, or current chronic or acute diseases were excluded | NR | NR | 65 (65)e |
| Liu [], 2006 | Prospective | Data from Clinical Proteomic Program Databank | NR | NR | NR | 253 (NR) |
| Wu et al [], 2006 | Prospective | Data from the Tri-Service General Hospital, Taiwan | No history of gynecologic tumors and had a normal pelvic examination and pelvic sonography | NR | NR | 65 (NR) |
| Li and Ramamohanrao [],2004 | Prospective | Data from a public website | NR | From November 2003 | NR | 253 (215/112)c |
| Li et al [], 2004 | Retrospective | Data from a public website | NR | From February 2002 | NR | 469 (100/116)c |
| Zhang et al [], 1999 | Retrospective | Data from an existing data set of clinically diagnosed with pelvic masses and University of Texas MD Anderson Cancer Center | NR | NR | NR | 625 (174/255)c |
| Wilding et al [], 1994 | Prospective | Data from the Hospital of the University of Pennsylvania | Patients with carcinoma in situ were excluded | NR | NR | 98 (NR) |
aTraining/external validation 1/external validation 2.
bNR: not reported.
cTraining/testing.
dNCS: noncancer symptomatic.
eTraining.
fTraining/internal validation/external validation 1/external validation 2.
gTraining/internal validation.
hTraining/internal validation/external validation.
iNOMGT: nonovarian malignant gynecologic tumor.
jBGD: benign gynecologic disease.
kTesting.
lTraining/prediction.
mTraining/testing/internal validation.
| Study | Reference standard | Algorithms | MLa or DLb | Type of internal validation | External validation |
| Cai et al [], 2024 | Histopathology | MCFc, XGBd, LGBMe, CatBoost, GBMf, RFg, NBh, LRi | ML | 5-fold cross-validation | Yes |
| Abuzinadah et al [], 2023 | Histopathology | RF, KNNj, SGDk, ETCl, XGB, GBM | ML | K-fold cross-validation | No |
| Bifarin and Fernandez [], 2024 | Histopathology | AutoMLm, RF, SVMn, KNN | ML | 5-fold cross-validation | No |
| Cameron et al [], 2023 | Histopathology | NR | ML | A nested cross-validation | No |
| Chen et al [], 2023 | Histopathology | CBSo, GISTICp | ML | 3-fold cross-validation | Yes |
| Dhar et al [], 2023 | Histopathology | LR | ML | 10-fold cross-validation | No |
| Hamidi et al [], 2023 | NR | LR, DTq, RF, ANNr, XGB | ML | 5-fold cross-validation | Yes |
| Lai et al [], 2023 | Histopathology | SVM | ML | A validation (unclear) | No |
| Li et al [], 2024 | Histopathology | LDAs, RF, NNt, SVM | ML | A validation (unclear) | No |
| Reilly et al [], 2023 | Histopathology | MIA3Gu | DL | NRv | No |
| Zhang et al [], 2023 | Histopathology | LR, FLDw, SVM, RF, ANN | ML | Cross-validation | Yes |
| Ahamad et al [], 2022 | Histopathology | RF, SVM, DT, XGBM, LR, GBM, LGBM | ML | 5-fold cross-validation | No |
| Bahado-Singh et al [], 2022 | NR | RF, SVM, LDA, PAMx, GLMy, DL | ML | 10-fold cross-validation | No |
| Gupta et al [], 2022 | NR | OVRz | ML | NR | No |
| Hinestrosa et al [], 2022 | Histopathology | RFEaa | ML | 5-fold cross-validation | No |
| Irajizad et al [], 2022 | NR | DL, RF, ELab, GBM | DL | 5-fold cross-validation | Yes |
| Kim et al [], 2022 | Histopathology | DT, LR, ANN, RF, SVM | ML | 10-fold cross-validation | Yes |
| Li et al [], 2022 | Histopathology | SVM | ML | 5-fold cross-validation | No |
| Pais et al [], 2022 | Histopathology | EvA-3ac, OSCad | ML | A validation (unclear) | No |
| Jeong et al [], 2021 | NR | ROMAae | ML | 3-fold cross-validation | No |
| Lu et al [], 2020 | Histopathology | ROMA, DT, LR | ML | 10-fold cross-validation | No |
| Banaei et al [], 2019 | NR | CTaf, KNN | ML | 5-fold cross-validation | No |
| Whitwell et al [], 2018 | Histopathology | Parenclitic networks, LR, RDLGag | ML | Monte Carlo cross-validation | No |
| Ivanova et al [], 2016 | Histopathology | GAah, SNNai | ML | Leave one out cross-validations | No |
| Jiang et al [], 2013 | NR | ANN, CT, Split-point score analysis | ML | One cross-validation | No |
| Yang et al [], 2013 | Histopathology | ANN | ML | Blind test validation | No |
| Shan et al [], 2012 | Histopathology | HH-SVMaj | ML | 5-fold cross-validation | No |
| Thakur et al [], 2011 | NR | ANNs, LDA | ML | Cross-validation | No |
| Donach et al [], 2010 | NR | ANN | ML | NR | No |
| Ziganshin et al [], 2008 | NR | GA, SNN | ML | Cross-validation | No |
| Liu et al [], 2007 | NR | PLSak, SVM, DT C5.0 | ML | 10-fold cross-validation | No |
| Zhang et al [], 2007 | Histopathology | ANN | ML | Cross-validation | Yes |
| Chatterjee et al [], 2006 | NR | Feed-forward NN | DL | NR | No |
| Lin et al [], 2006 | NR | DT | ML | Cross-validation | No |
| Liu [], 2006 | NR | SVM | ML | 10-fold cross-validation | No |
| Wu et al [], 2006 | Histopathology | CT | ML | Cross-validation | No |
| Li and Ramamohanrao [], 2004 | NR | SVM, NB, KNN, DT, CS4al | ML | 10-fold cross-validation | No |
| Li et al [], 2004 | NR | SVM | ML | One out cross-validation | No |
| Zhang et al [], 1999 | Histopathology | ANN | ML | Cross-validation | No |
| Wilding et al [], 1994 | Histopathology | Backpropagation NN | DL | A validation (unclear) | No |
aML: machine learning.
bDL: deep learning.
cMCF: multi-criteria decision making-based classification fusion.
dXGB: extreme gradient boosting.
eLGBM: light gradient boosting machine.
fGBM: gradient boosting machine.
gRF: random forest.
hNB: naive Bayes.
iLR: logistic regression.
jKNN: k-nearest neighbor.
kSGD: stochastic gradient descent.
lETC: extra-trees classifier.
mAutoML: automated machine learning.
nSVM: support vector machine.
oCBS: circular binary segmentation algorithm.
pGISTIC: the genomic identification of significant targets in cancer 2.0 algorithm.
qDT: decision tree.
rANN: artificial neural network.
sLDA: Linear Discriminant Analysis.
tNN: neural network.
uMIA3G: multivariate index assay 3G.
vNR: not reported.
wFLD: Fisher linear discriminant.
xPAM: prediction analysis for microarrays.
yGLM: generalized linear model.
zOVR: One Versus Rest classifier multiclass classification model.
aaRFE: Recursive Feature Elimination.
abEL: ensemble learning.
acEva-3: evolutionary algorithm 3
adOSC: compose the classification algorithm.
aeROMA: risk of ovarian malignancy algorithm.
afCT: classification tree.
agRDLG: raw data logistic regression.
ahGA: genetic algorithm.
aiSNN: supervised neural network.
ajHH-SVM: hybrid huberized support vector machine.
akPLS: partial least-square regression.
alCS4: cascading-and-sharing for ensembles of decision trees.
| Study | Blood sample type | Device or method | Biomarker type | Number of modeling biomarkers | Number of detection marker |
| Cai et al [], 2024 | Blood | NRa | Protein, mixed | 52 | NR |
| Abuzinadah et al [], 2023 | Blood | General chemical tests, blood routine tests | Mixed | 49 | 49 |
| Bifarin and Fernandez [], 2024 | Serum | NR | Mixed | 17 | 17 |
| Cameron et al [], 2023 | Serum | Spectrum | Mixed | 5 | 5 |
| Chen et al [], 2023 | Plasma | LC-WGSb | DNA | NR | NR |
| Dhar et al [], 2023 | Serum | LCMSc | Protein | 27 | 571 |
| Hamidi et al [], 2023 | Serum | miRNA labeling kit and miRNA Oligo Chip | RNA | 10 | 2568 |
| Lai et al [], 2023 | Serum | MSd | Protein | 7 | 95 |
| Li et al [], 2024 | Plasma | Nanoflow cytometry, SECe, CA125 ELISAf kit, and HE4 ELISA kit | Protein | 7 | 7 |
| Reilly et al [], 2023 | Serum | Roche cobas 6000 clinical analyzer | Protein | 7 | 7 |
| Zhang et al [], 2023 | Blood | NR | Mixed | 25 | 25 |
| Ahamad et al [], 2022 | Blood and serum | NR | Mixed | 47 | 47 |
| Bahado-Singh et al [], 2022 | Plasma | Illumina Infnium MethylationEPIC BeadChip arrays or methylation analysis | DNA | 25 | 179,238 |
| Gupta et al [], 2022 | Serum | UHPLC‑MSg | Mixed | 25 | 6336 |
| Hinestrosa et al [], 2022 | Plasma | ACEh | Protein | 34 | 42 |
| Irajizad et al [], 2022 | Plasma | LCMS analysis | Mixed | 7 | 475 |
| Kim et al [], 2022 | Serum | Immunoassay, C8000 analyzer, diazo reagent | Mixed | NR | NR |
| Li et al [], 2022 | Plasma | QIAamp Circulating Nucleic Acid kit | DNA | 5 | 1272 |
| Pais et al [], 2022 | Serum | MALDINR-TOF MSi | Protein | CHCAj:26-57; SAk:12-113 | CHCA:8500; SA:8500 |
| Jeong et al [], 2021 | Serum | 2-step chemiluminescent microparticle immunoassay | Protein | 16 | 3 |
| Lu et al [], 2020 | Serum | Roche Cobas 8000 modular analyzer series | Protein | 2 | 2 |
| Banaei et al [], 2019 | Serum | A microfluidic SERSl-based immunoassay method | Protein | 5 | 5 |
| Whitwell et al [], 2018 | Serum | Olink’s multiplex immunoassay Oncology II panel | Protein | 92 | 92 |
| Ivanova et al [], 2016 | Serum | MALDI-TOF MS | Protein | 7 | 200-400 |
| Jiang et al [], 2013 | Serum | ELISA | Protein | 5 | 174 |
| Yang et al [], 2013 | Serum | SELDI-TOF MSm | Protein | 184 | 184 |
| Shan et al [], 2012 | Serum | Liquid chromatography electrospray tandem mass spectrometry | Mixed | 18 | 18 |
| Thakur et al [], 2011 | Serum | SELDI-TOF MS | Protein | NR | NR |
| Donach et al [], 2010 | Serum | Radioimmunoassay kits | Protein | 4 | 6 |
| Ziganshin et al [], 2008 | Serum | MALDI-TOF MS or Ultraflex TOF mass spectrometer | Protein | MB-IMAC Cun:13; MB-WCX:11 | MB-HIC8o:135;MB-HIC18p:137; MB-IMAC Cu:115; MB-WCXq:96 |
| Liu et al [], 2007 | Serum | prOTOF MS | Protein | NR | 96 |
| Zhang et al [], 2007 | Serum | Radioimmunoassay kits | Protein | 4 | 4 |
| Chatterjee et al [], 2006 | Serum | ELISA | Protein | 65 | 65 |
| Lin et al [], 2006 | Plasma | SELDI analysis,WCX2 chip analysis,SAX2 chip analysis | Protein | 3 | 4 |
| Liu [], 2006 | Serum | MS | Protein | NR | NR |
| Wu et al [], 2006 | Plasma | SELDI-TOF MS | Protein | 5 | NR |
| Li and Ramamohanrao [], 2004 | Serum | MS | Protein | 72 | 15,154 |
| Li et al [], 2004 | Serum | SELDI-TOF MS | Protein | 10 | 15,155 |
| Zhang et al [], 1999 | Serum | Radioimmunoassay kits, a spectrophotometric method with a kit | Mixed | 4 | 4 |
| Wilding et al [], 1994 | Serum and plasma | Radioimmunoassay | Mixed | 8 | 8 |
aNR: not reported.
bLC-WGS: low-coverage whole genome sequencing.
cLC-MS: liquid chromatography-mass spectrometry.
dMS: mass spectrometry.
eSEC: size exclusion chromatography.
fELISA: enzyme-linked immunosorbent assay.
gUHPLC-MS: ultra-high performance liquid chromatography-mass spectrometry
hACE: alternating current electrokinetics.
iMALDINR-TOF MS: matrix-assisted laser desorption/Ionization neutral reflector time-of-flight mass spectrometry.
jCHCA: α-Cyano-4-hydroxycinnamic acid matrix.
kSA: sinapinic acid matrix.
lSERS: surface-enhanced Raman spectroscopy.
mSELDI -TOF MS: surface-enhanced laser desorption and ionization mass spectrometry.
nMB-IMAC Cu: magnetic beads MB-IMAC Cu.
oMB-HIC8: magnetic beads MB-HIC 8.
pMB-HIC18: magnetic beads MB-HIC 18.
qMB-WCX: magnetic beads MB-WCX.
Quality Assessment
The quality of the included studies was appraised using the QUADAS-AI (Figures S1 and S2 in ). In detail, most of the studies were rated as having a high or unclear risk of bias based on patient selection (22/40, 55%) and index test (33/40, 82%) domains. These assessments might be attributed to the absence of explicit delineation of included patients, such as previous testing history and clinical setting, as well as to deficiencies in rigorous external validation of the AI models.
Pooled Performance of AI Algorithms
The summary receiver operating characteristics curves for the 40 included studies with 342 contingency tables are shown in 
Figure 2. The pooled sensitivity and specificity were 85% (95% CI 83%-87%) and 91% (95% CI 90%-92%), respectively, with an AUC of 0.95 (95 % CI 0.92-0.96) for all AI algorithms. Notably, when contingency tables with the highest accuracy were extracted from each study, the pooled sensitivity and specificity were 95% (95% CI 90%-97%) and 97% (95% CI 95%-98%), respectively, with an AUC of 0.99 (95% CI 0.98-1.00). Reported point estimates and CIs of all included studies are shown in a cross-hairs plot (
Figure 3).
Subgroup Analyses and Meta-Regression Analysis
Results of the subgroup analyses revealed that acceptable diagnostic performance was observed in all subgroups, ranging from 74% to 98% for sensitivity and 85% to 96% for specificity. Detailed results are shown in Table 4, and the corresponding plots are presented in Figures S3-S21 in . We divided the studies into subgroups according to the modalities of algorithms (Figures S3 and S13 in ), existence of external validation (Figures S4 and S14 in ), levels of risk of bias (Figures S5 and S15 in ), year of publication (Figures S6 and S16 in ), geographical distribution (Figures S7 and S17 in ), sample size (Figures S8 and S18 in ), blood sample type (Figures S9 and S19 in ), biomarkers type (Figures S10 and S20 in ), and number of modeling biomarkers (Figures S11 and S21 in ).
| Number of studies, n (%) | Sensitivity | P valuea | Specificity | P valuea | |||||||||
| Sensitivity (95% CI) | I2（95% CI） | P valueb | Specificity (95% CI) | I2（95% CI) | P valueb | ||||||||
| Overall | 40 (100) | 0.85 (0.83-0.87) | 97.54 (97.41-97.66) | <.001 | —c | 0.91 (0.90-0.92) | 99.15 (99.12-99.18) | <.001 | — | ||||
| Artificial intelligence algorithmsd | .07 | .10 | |||||||||||
| Machine learning | 36 (90) | 0.86 (0.83-0.88) | 97.79 (97.67-97.90) | <.001 | — | 0.92 (0.90-0.93) | 99.27 (99.24-99.30) | <.001 | — | ||||
| Deep learning | 4 (10) | 0.77 (0.70-0.82) | 85.81 (81.19-90.42) | <.001 | — | 0.85 (0.83-0.87) | 68.55 (55.84-81.26) | <.001 | — | ||||
| External validation | <.001 | <.001 | |||||||||||
| Yes | 7 (18) | 0.74 (0.69-0.79) | 98.35 (98.23-98.46) | <.001 | — | 0.94 (0.93-0.95) | 99.54 (99.52-99.57) | <.001 | — | ||||
| No | 33 (82) | 0.90 (0.88-0.92) | 94.91 (94.48-95.34) | <.001 | — | 0.89 (0.86-0.91) | 94.17 (93.65-94.68) | <.001 | — | ||||
| Levels of risk of biase | <.001 | .28 | |||||||||||
| Low | 31 (78) | 0.81 (0.79-0.84) | 97.40 (97.25-97.55) | <.001 | — | 0.91 (0.89-0.92) | 99.07 (99.03-99.11) | <.001 | — | ||||
| High | 9 (22) | 0.98 (0.96-0.99) | 95.51 (94.74-96.27) | <.001 | — | 0.94 (0.89-0.96) | 96.85 (96.37-97.33) | <.001 | — | ||||
| Year of publication | <.001 | <.001 | |||||||||||
| After 2022 | 19 (48) | 0.81 (0.78-0.83) | 97.44 (97.28-97.59) | <.001 | — | 0.90 (0.88-0.91) | 99.04 (98.99-99.08) | <.001 | — | ||||
| Before 2022 | 21 (52) | 0.95 (0.91-0.97) | 96.83 (96.47-97.19) | <.001 | — | 0.96 (0.93-0.97) | 97.64 (97.39-97.88) | <.001 | — | ||||
| Geographical distribution | 0.01 | .92 | |||||||||||
| Asia | 18 (45) | 0.82 (0.78-0.84) | 97.78 (97.65-97.91) | <.001 | — | 0.91 (0.89-0.92) | 99.23 (99.20-99.26) | <.001 | — | ||||
| North America | 15 (38) | 0.92 (0.88-0.95) | 94.47 (93.70-95.24) | <.001 | — | 0.91 (0.87-0.93) | 95.88 (95.35-96.40) | <.001 | — | ||||
| Europe | 7 (18) | 0.91 (0.81-0.96) | 96.54 (95.69-97.38) | <.001 | — | 0.96 (0.90-0.98) | 97.02 (96.33-97.71) | <.001 | — | ||||
| Sample size | <.001 | .06 | |||||||||||
| >300 | 21 (52) | 0.83 (0.80-0.85) | 97.78 (97.65-97.91) | <.001 | — | 0.91 (0.89-0.92) | 99.24 (99.21-99.27) | <.001 | — | ||||
| ≤300 | 19 (48) | 0.92 (0.88-0.95) | 93.93 (93.07-94.79) | <.001 | — | 0.94 (0.91-0.96) | 92.88 (91.82-93.94) | <.001 | — | ||||
| Blood sample typef | <.02 | .12 | |||||||||||
| Serum | 27 (68) | 0.94 (0.92-0.96) | 98.13 (97.98-98.29) | <.001 | — | 0.96 (0.95-0.98) | 98.57 (98.46-98.68) | <.001 | — | ||||
| Plasma | 8 (20) | 0.83 (0.78-0.87) | 83.32 (79.03-87.61) | <.001 | — | 0.91 (0.88-0.94) | 70.15 (61.13-79.17) | <.001 | — | ||||
| Marker typeg | .12 | <.001 | |||||||||||
| Protein | 25 (62) | 0.87 (0.82-0.91) | 98.65 (98.56-98.75) | <.001 | — | 0.95 (0.93-0.96) | 99.65 (99.63-99.66) | <.001 | — | ||||
| Mixed | 12 (30) | 0.79 (0.77-0.82) | 94.74 (94.27-95.21) | <.001 | — | 0.86 (0.84-0.88) | 98.56 (98.47-98.64) | <.001 | — | ||||
| Number of modeling markerh | .54 | .31 | |||||||||||
| >8 | 15 (38) | 0.82 (0.79-0.85) | 97.73 (97.59-97.87) | <.001 | — | 0.90 (0.88-0.92) | 99.20 (99.16-99.23) | <.001 | — | ||||
| ≤8 | 14 (35) | 0.79 (0.74-0.83) | 88.11 (85.74-90.49) | <.001 | — | 0.90 (0.88-0.92) | 79.83 (75.13-84.52) | <.001 | — | ||||
aP value for heterogeneity between subgroups with meta-regression analysis.
bP value for heterogeneity within each subgroup.
cNot applicable.
dArtificial intelligence algorithms include machine learning and deep learning.
eLow: ≥2 domain low risk; high: <2 domain low risk.
fFive articles that used incomplete information on blood sample type were excluded from this subgroup analysis.
gThree DNA studies that used DNA and 1 study that used RNA were excluded for this subgroup analyses.
hIn total, 11 articles used incomplete information on the number of model markers were excluded for this subgroup analysis.
The meta-analysis uncovered substantial heterogeneity among studies, as evidenced by an I2 of 97.54% (P<.05) for sensitivity and 99.15% (P<.05) for specificity (Figure S12 in ). To further explore the causes of study heterogeneity, a meta-regression analysis was conducted (Table 4). The results showed that both external validation and year of publication were significant factors that influenced study heterogeneity with regard to sensitivity and specificity. Subgroups based on levels of risk of bias, geographical distribution, sample size, and blood sample type showed intergroup heterogeneity in the sensitivity of prediction (P<.05). In terms of marker type, specificity presented significant heterogeneity between groups (P<.05).
Sensitivity Analyses and Publication Bias
A qualitative systematic review was performed for studies lacking directly or indirectly available contingency tables. The findings from this review were in alignment with the main analysis (Tables S1-S3 in [-]). In addition, the analysis did not reveal any publication bias in this meta-analysis (P=.72; ).
Discussion
Principal Findings
The burgeoning evolution of AI within the medical field has captured the attention of an increasing cadre of researchers, particularly in its applicability to disease diagnosis [,]. To the best of our knowledge, this meta-analysis is a pioneering effort specifically exploring the efficacy of AI in OC diagnosis via blood biomarkers. AI algorithms exhibited exceptional diagnostic capabilities for OC, boasting a pooled sensitivity of 85% (95% CI 83%-87%) and specificity of 91% (95% CI 90%-92%). Moreover, we identified substantial heterogeneity among the selected studies and determined the potential contributing factors through subgroup and meta-regression analyses. Overall, these results should be interpreted with caution as described by the constraints mentioned in subsequent sections.
Heterogeneity
Heterogeneity is an inevitable problem in meta-analyses []. Significant interstudy heterogeneity was noted in terms of sensitivity (I2=97.54%) and specificity (I2=99.15%) in this study. External validation emerged as a crucial variable influencing study heterogeneity. Studies without external validation might yield results that were hard to generalize owing to factors such as sample selection bias and the characteristics of the research setting []. To address this, future research should focus on standardizing and applying the validation procedures, thus getting closer to the goal of providing more accurate and reliable diagnostic tools for clinical practice []. Besides, several factors contributed to the heterogeneity observed in sensitivity in this study. Studies with a pronounced risk of bias were predisposed to introduce uncertainties. Such biases could stem from flaws in study design, improper data collection, or inappropriate statistical analyses, all of which might distort the true relationship between the biomarker and the condition under investigation [,]. On the other hand, geographic disparities might be attributed to a complex interplay of genetic polymorphisms and environmental factors that differentially modulate biomarker expression levels, which could result in significant variability in biomarker sensitivity [,]. In addition, larger sample sizes generally offer enhanced statistical power and precision, enabling more reliable estimations of biomarker performance []. Understanding and accounting for these factors comprehensively will help to further reduce heterogeneity and enhance the validity and clinical relevance of meta-analysis results, ultimately leading to more precise and useful diagnostic tools for clinical application.
Implication of Blood Sample Types
Blood specimens are relatively stable and can be easily accessed. Therefore, blood-based biomarkers have been regarded as a minimally invasive method with great value for disease diagnosis [,]. Plasma and serum are rich sources of information regarding an individual’s health state and are the focus of this study’s investigation []. Particularly noteworthy is that serum samples showed higher sensitivity than plasma ones in this study, which can potentially be ascribed to multiple factors. First, the distinct sample preparation procedures for serum and plasma may lead to variations in the concentration and availability of biomarkers. Serum is obtained after blood clotting, during which certain intracellular components can be released, potentially augmenting the repertoire of biomarkers []. In contrast, the anticoagulants used in plasma collection may impede the integrity or accessibility of biomarkers []. Second, the microenvironment within serum and plasma varies. Serum contains a more intricate milieu of proteins, enzymes, and other biomolecules that can interact with cancer biomarkers in a manner that heightens their detectability []. For example, the presence of specific binding proteins or proteases in serum may modify the conformation of biomarkers, rendering them more amenable to detection by the analytic methods used []. Furthermore, the centrifugation processes involved in separating serum and plasma can differentially partition the biomarkers. The speed, duration, and temperature of centrifugation may cause certain biomarkers to be preferentially retained in the serum fraction, contributing to the observed higher sensitivity []. In addition, the limited number of included studies for plasma may contribute to this phenomenon to some extent. Therefore, it is of paramount importance to judiciously select the sample type in the context of developing and implementing blood-based biomarker assays for OC.
Implication of Algorithm Types
In subgroup analysis, ML surpassed DL in both sensitivity and specificity. This phenomenon warrants an in-depth exploration of the underlying reasons and improvement directions. The edge of ML likely stems from its algorithmic traits. For structured and well-defined data, traditional algorithms (eg, logistic regression and support vector machine) can adeptly capture biomarker-disease associations via mathematical and statistical tenets, yielding high diagnostic accuracy [,]. DL, empowered by its strong automatic feature extraction and complex architecture, can theoretically handle large data and extract deep patterns []. However, in this study, the number of studies included for DL was only 4, compared with 36 for ML, presumably constraining the exertion of DL’s advantages. The scant number of DL studies gives rise to data that are circumscribed in both sample variety and the expanse of feature distribution. To break through this dilemma, several optimization strategies can be considered. First, data augmentation, such as random rotations, scaling, flipping, and noise addition to the data can enhance dataset diversity, facilitating the DL model to learn more extensive features and patterns for better generalization [-]. Second, transfer learning is applicable. Using pretrained models from related medical or bioanalysis fields and fine-tuning with OC data, the DL model can draw on prior knowledge, accelerating training convergence and potentially improving performance [,]. In addition, model compression techniques, including pruning (ie, eliminating less important connections or neurons to maintain performance while reducing complexity) and quantization (ie, lowering parameter precision for faster inference and less memory use) can be used []. While these strategies hold promise, their implementation and efficacy in the context of OC diagnosis warrant further investigation and optimization.
Implication of External Validation
At present, the data amassed for AI applications in OC diagnosis is circumscribed by the paucity of diverse external validation. Many studies rely on a single dataset for discovery, with cross-validation to estimate algorithm performance. Given the generalizable issues to unseen data, accuracy drops when tested on other research datasets, and substantially when tested on clinical data [,]. To address these challenges, several strategies can be implemented. First, multicenter collaborations should be actively pursued. Combining data from different medical institutions and regions to build a heterogeneous and comprehensive dataset and exposing algorithms to wider patient characteristics and biomarker profiles will enhance generalizability [-]. Second, standardized data collection and annotation protocols are crucial. They ensure data consistency and comparability among studies, minimizing variability from inconsistent methods []. This allows algorithms to be trained on more reliable and reproducible data, strengthening the foundation for AI application. Moreover, continuous evaluation and improvement of algorithms in clinical settings are essential. Prospective studies integrating the algorithms into routine practice and monitoring their performance can offer valuable feedback []. This iterative testing and refinement process helps algorithms adapt to clinical complexity, leading to more accurate OC diagnostic tools. Despite persisting challenges, we anticipate that these efforts will incrementally enhance diagnostic accuracy. Sustained refinement and collaboration are essential to exploiting the full potential of AI in OC diagnosis.
Future Directions
Beyond the previous mentions, the existing literature in the field exhibits certain areas of improvement to reduce the gap between research and deployment. First, one of the richest data sources of patient health and clinical history is embedded in the electronic health records of a patient but remains hugely underutilized. AI’s ability to integrate blood biomarkers with other clinically relevant nonblood biomarkers, such as age, cancer history, and family history of cancer, could potentially outpace current practices if trained on sufficiently extensive datasets []. Future research could explore the synergistic integration of AI tools with clinical expertise, echoing a more realistic clinical scenario. Second, the problem of explainability is the subject of intensive research and various initiatives. Although symbolic AI or simple ML models, such as decision trees or linear regression, are still fully understood by people, understanding becomes increasingly difficult with more advanced techniques and is now impossible with many DL models; this situation can lead to unexpected results and nondeterministic behavior [,]. Third, data privacy and patient consent are critical concerns that need to be addressed before adopting the use of AI in clinical practice []. The integration of AI into clinical workflows requires careful consideration of ethical, legal, and regulatory aspects [,]. Transparent guidelines and regulations should be established to govern the use of AI in health care and ensure its responsible and ethical implementation.
Strengths and Limitations
Our meta-analysis has several strengths. First, to the best of our knowledge, it represents a novel effort as the first systematic review and meta-analysis dedicated to evaluating the diagnostic performance of blood biomarker-based AI for OC. Our findings illuminate the considerable potential AI holds in this domain, while also highlighting the advantages of blood tests, such as their noninvasive nature, better patient compliance, and cost-effectiveness. Second, our comprehensive investigation included multiple subgroup analyses, all of which yielded acceptable diagnostic performance for the AI model. Third, the stringent quality assessment of all included studies was conducted using QUADAS-AI tool. In addition, the robustness of our meta-analysis was reinforced through a sensitivity analysis underpinned by a qualitative systematic review.
The results of our meta-analysis are likely to be overestimated or underestimated for some reasons. One limitation lies in the high heterogeneity of the studies included. Nonetheless, we thoroughly explored potential sources of between-study heterogeneity through meta-regression and subgroup analyses. Another limitation is that the contingency tables of 5 studies included in our systematic review were not directly or indirectly available. These studies provided only indicators, such as AUC, accuracy, and F1-score, which did not allow for the construction of contingency tables [-]. Nevertheless, we conducted a qualitative systematic review of these 5 studies and discovered that the findings aligned with the main analysis. Moreover, although no publication bias was noticed, it is still highly likely that there is unpublished material for this topic from the ever-growing nature of the framework and the likelihood of undisclosed research for commercial development []. Moreover, available AI research tends to be skewed toward the publication of positive results, indicating a potential publication bias.
Conclusions
The findings of this study indicated that the use of AI for the analysis of noninvasive blood biomarkers in OC diagnostics holds substantial potential for achieving satisfactory predictive outcomes. Among the analyzed studies, those that used DL were notably fewer in number than those that used ML. This underscores a critical need for future research to prioritize the incorporation of DL methodologies. Furthermore, pursuing external validation datasets was a necessary avenue to optimize the performance and applicability of AI in this field.
Acknowledgments
This work was supported by the National Key Research and Development Program of China (QJW: 2022YFC2704205), the Natural Science Foundation of China (QJW: 82073647 and 82373674; TTG: 82103914), Outstanding Scientific Fund of Shengjing Hospital (QJW), Liaoning Province Science and Technology Plan (QJW: 2023JH2/20200019), and Scientific Research Project of Education Department of Liaoning Province (TTG: LJKMZ20221137).
Data Availability
The datasets generated or analyzed during this study are available from the corresponding author on reasonable request.
Authors' Contributions
HLX, TTG, and QJW contributed to the study design. HLX, XYL, MQJ, QPM, YHZ, FHL, YQ, YHC, YL, XYC, YLX, DRL, and DDW contributed to the collection and analysis of data. HLX, XYL, MQJ, QPM, DHH, QX, YHZ, SG, XQ, TT, TTG, and QJW wrote the first draft of the manuscript and edited the manuscript. All authors read and approved the final manuscript. HLX, XYL, MQJ, and QPM contributed equally to this work.
Conflicts of Interest
None declared.
Multimedia Appendix 4
The list of the excluded records during the process of full-text review.
DOCX File , 16 KBMultimedia Appendix 5
The results of Quality Assessment of Diagnostic Accuracy Studies–Artificial Intelligence and subgroup analyses.
DOCX File , 39633 KBMultimedia Appendix 6
The characteristics for the 5 studies without sufficient data.
DOCX File , 19 KBReferences
- Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin. 2024;74(1):12-49. [FREE Full text] [CrossRef] [Medline]
- Torre LA, Trabert B, DeSantis CE, Miller KD, Samimi G, Runowicz CD, et al. Ovarian cancer statistics, 2018. CA Cancer J Clin. Jul 2018;68(4):284-296. [FREE Full text] [CrossRef] [Medline]
- Li G, Zhang Y, Li K, Liu X, Lu Y, Zhang Z, et al. Transformer-based AI technology improves early ovarian cancer diagnosis using cfDNA methylation markers. Cell Rep Med. Aug 20, 2024;5(8):101666. [FREE Full text] [CrossRef] [Medline]
- Zeng S, Wang XL, Yang H. Radiomics and radiogenomics: extracting more information from medical images for the diagnosis and prognostic prediction of ovarian cancer. Mil Med Res. Dec 14, 2024;11(1):77. [FREE Full text] [CrossRef] [Medline]
- Dochez V, Caillon H, Vaucel E, Dimet J, Winer N, Ducarme G. Biomarkers and algorithms for diagnosis of ovarian cancer: CA125, HE4, RMI and ROMA, a review. J Ovarian Res. Mar 27, 2019;12(1):28. [FREE Full text] [CrossRef] [Medline]
- Perakis S, Speicher MR. Emerging concepts in liquid biopsies. BMC Med. Apr 06, 2017;15(1):75. [FREE Full text] [CrossRef] [Medline]
- Qian L, Sun R, Xue Z, Guo T. Mass spectrometry-based proteomics of epithelial ovarian cancers: a clinical perspective. Mol Cell Proteomics. Jul 2023;22(7):100578. [FREE Full text] [CrossRef] [Medline]
- Dhar C, Ramachandran P, Xu G, Pickering C, Čaval T, Wong M, et al. Diagnosing and staging epithelial ovarian cancer by serum glycoproteomic profiling. Br J Cancer. Jun 2024;130(10):1716-1724. [CrossRef] [Medline]
- Terp SK, Stoico MP, Dybkær K, Pedersen IS. Early diagnosis of ovarian cancer based on methylation profiles in peripheral blood cell-free DNA: a systematic review. Clin Epigenetics. Feb 14, 2023;15(1):24. [FREE Full text] [CrossRef] [Medline]
- Singh A, Shannon CP, Kim YW, Yang CX, Balshaw R, Cohen Freue GV, et al. Novel blood-based transcriptional biomarker panels predict the late-phase asthmatic response. Am J Respir Crit Care Med. Feb 15, 2018;197(4):450-462. [CrossRef] [Medline]
- O'Neill S, Bohl M, Gregersen S, Hermansen K, O'Driscoll L. Blood-based biomarkers for metabolic syndrome. Trends Endocrinol Metab. Jun 2016;27(6):363-374. [CrossRef] [Medline]
- Gao Y, Cao D, Li M, Zhao F, Wang P, Mei S, et al. Integration of multiomics features for blood-based early detection of colorectal cancer. Mol Cancer. Aug 22, 2024;23(1):173. [CrossRef] [Medline]
- Ke C, Hou Y, Zhang H, Fan L, Ge T, Guo B, et al. Large-scale profiling of metabolic dysregulation in ovarian cancer. Int J Cancer. Feb 01, 2015;136(3):516-526. [CrossRef] [Medline]
- Xie J, Xu P, Lin Y, Zheng M, Jia J, Tan X, et al. LncRNA-miRNA interactions prediction based on meta-path similarity and Gaussian kernel similarity. J Cell Mol Med. Oct 2024;28(19):e18590. [FREE Full text] [CrossRef] [Medline]
- Yin S, Xu P, Jiang Y, Yang X, Lin Y, Zheng M, et al. Predicting the potential associations between circRNA and drug sensitivity using a multisource feature-based approach. J Cell Mol Med. Oct 2024;28(19):e18591. [FREE Full text] [CrossRef] [Medline]
- Wang T, Sun J, Zhao Q. Investigating cardiotoxicity related with hERG channel blockers using molecular fingerprints and graph attention mechanism. Comput Biol Med. Feb 2023;153:106464. [CrossRef] [Medline]
- Wang J, Zhang L, Sun J, Yang X, Wu W, Chen W, et al. Predicting drug-induced liver injury using graph attention mechanism and molecular fingerprints. Methods. Jan 2024;221:18-26. [CrossRef] [Medline]
- Mohsen F, Al-Absi HR, Yousri NA, El Hajj N, Shah Z. A scoping review of artificial intelligence-based methods for diabetes risk prediction. NPJ Digit Med. Oct 25, 2023;6(1):197. [FREE Full text] [CrossRef] [Medline]
- Chen Z, Zhang L, Sun J, Meng R, Yin S, Zhao Q. DCAMCP: a deep learning model based on capsule network and attention mechanism for molecular carcinogenicity prediction. J Cell Mol Med. Oct 2023;27(20):3117-3126. [FREE Full text] [CrossRef] [Medline]
- Wang W, Zhang L, Sun J, Zhao Q, Shuai J. Predicting the potential human lncRNA-miRNA interactions based on graph convolution network with conditional random field. Brief Bioinform. Nov 19, 2022;23(6):bbac463. [CrossRef] [Medline]
- Zhu F, Niu Q, Li X, Zhao Q, Su H, Shuai J. FM-FCN: a neural network with filtering modules for accurate vital signs extraction. Research (Wash D C). 2024;7:0361. [FREE Full text] [CrossRef] [Medline]
- Liu L, Wei Y, Zhang Q, Zhao Q. SSCRB: predicting circRNA-RBP interaction sites using a sequence and structural feature-based attention model. IEEE J Biomed Health Inform. Mar 2024;28(3):1762-1772. [CrossRef] [Medline]
- Yang X, Sun J, Jin B, Lu Y, Cheng J, Jiang J, et al. Multi-task aquatic toxicity prediction model based on multi-level features fusion. J Adv Res. Feb 2025;68:477-489. [FREE Full text] [CrossRef] [Medline]
- Cai G, Huang F, Gao Y, Li X, Chi J, Xie J, et al. Artificial intelligence-based models enabling accurate diagnosis of ovarian cancer using laboratory tests in China: a multicentre, retrospective cohort study. Lancet Digit Health. Mar 2024;6(3):e176-e186. [FREE Full text] [CrossRef] [Medline]
- Irajizad E, Han CY, Celestino J, Wu R, Murage E, Spencer R, et al. A blood-based metabolite panel for distinguishing ovarian cancer from benign pelvic masses. Clin Cancer Res. Nov 01, 2022;28(21):4669-4676. [FREE Full text] [CrossRef] [Medline]
- Liberati A, Altman DG, Tetzlaff J, Mulrow C, Gøtzsche PC, Ioannidis JP, et al. The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration. PLoS Med. Jul 21, 2009;6(7):e1000100. [FREE Full text] [CrossRef] [Medline]
- Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis of Observational Studies in Epidemiology (MOOSE) group. JAMA. Apr 19, 2000;283(15):2008-2012. [CrossRef] [Medline]
- Abuzinadah N, Kumar Posa S, Alarfaj AA, Alabdulqader EA, Umer M, Kim TH, et al. Improved prediction of ovarian cancer using ensemble classifier and shaply explainable AI. Cancers (Basel). Dec 11, 2023;15(24):5793. [FREE Full text] [CrossRef] [Medline]
- Bifarin OO, Fernández FM. Automated machine learning and explainable AI (AutoML-XAI) for metabolomics: improving cancer diagnostics. J Am Soc Mass Spectrom. Jun 05, 2024;35(6):1089-1100. [FREE Full text] [CrossRef] [Medline]
- Cameron JM, Sala A, Antoniou G, Brennan PM, Butler HJ, Conn JJ, et al. A spectroscopic liquid biopsy for the earlier detection of multiple cancer types. Br J Cancer. Nov 2023;129(10):1658-1666. [FREE Full text] [CrossRef] [Medline]
- Chen L, Ma R, Luo C, Xie Q, Ning X, Sun K, et al. Noninvasive early differential diagnosis and progression monitoring of ovarian cancer using the copy number alterations of plasma cell-free DNA. Transl Res. Dec 2023;262:12-24. [CrossRef] [Medline]
- Hamidi F, Gilani N, Arabi Belaghi R, Yaghoobi H, Babaei E, Sarbakhsh P, et al. Identifying potential circulating miRNA biomarkers for the diagnosis and prediction of ovarian cancer using machine-learning approach: application of Boruta. Front Digit Health. 2023;5:1187578. [FREE Full text] [CrossRef] [Medline]
- Lai Z, Wang Z, Yuan Z, Zhang J, Zhou J, Li D, et al. Disease-specific haptoglobin N-glycosylation in inflammatory disorders between cancers and benign diseases of 3 types of female internal genital organs. Clin Chim Acta. Jul 01, 2023;547:117420. [CrossRef] [Medline]
- Li J, Li Y, Li Q, Sun L, Tan Q, Zheng L, et al. An aptamer-based nanoflow cytometry method for the molecular detection and classification of ovarian cancers through profiling of tumor markers on small extracellular vesicles. Angew Chem Int Ed Engl. Jan 22, 2024;63(4):e202314262. [CrossRef] [Medline]
- Reilly GP, Dunton CJ, Bullock RG, Ure DR, Fritsche H, Ghosh S, et al. Validation of a deep neural network-based algorithm supporting clinical management of adnexal mass. Front Med (Lausanne). 2023;10:1102437. [FREE Full text] [CrossRef] [Medline]
- Zhang T, Pang A, Lyu J, Ren H, Song J, Zhu F, et al. Application of nonlinear models combined with conventional laboratory indicators for the diagnosis and differential diagnosis of ovarian cancer. J Clin Med. Jan 20, 2023;12(3):844. [FREE Full text] [CrossRef] [Medline]
- Ahamad MM, Aktar S, Uddin MJ, Rahman T, Alyami SA, Al-Ashhab S, et al. Early-stage detection of ovarian cancer based on clinical data using machine learning approaches. J Pers Med. Jul 25, 2022;12(8):1211. [FREE Full text] [CrossRef] [Medline]
- Bahado-Singh RO, Ibrahim A, Al-Wahab Z, Aydas B, Radhakrishna U, Yilmaz A, et al. Precision gynecologic oncology: circulating cell free DNA epigenomic analysis, artificial intelligence and the accurate detection of ovarian cancer. Sci Rep. Nov 03, 2022;12(1):18625. [FREE Full text] [CrossRef] [Medline]
- Gupta A, Sagar G, Siddiqui Z, Rao KV, Nayak S, Saquib N, et al. A non-invasive method for concurrent detection of early-stage women-specific cancers. Sci Rep. Feb 10, 2022;12(1):2301. [FREE Full text] [CrossRef] [Medline]
- Hinestrosa JP, Kurzrock R, Lewis JM, Schork NJ, Schroeder G, Kamat AM, et al. Early-stage multi-cancer detection using an extracellular vesicle protein-based blood test. Commun Med (Lond). 2022;2:29. [FREE Full text] [CrossRef] [Medline]
- Kim M, Chen C, Wang P, Mulvey JJ, Yang Y, Wun C, et al. Detection of ovarian cancer via the spectral fingerprinting of quantum-defect-modified carbon nanotubes in serum by machine learning. Nat Biomed Eng. Mar 2022;6(3):267-275. [FREE Full text] [CrossRef] [Medline]
- Li N, Zhu X, Nian W, Li Y, Sun Y, Yuan G, et al. Blood-based DNA methylation profiling for the detection of ovarian cancer. Gynecol Oncol. Nov 2022;167(2):295-305. [CrossRef] [Medline]
- Pais RJ, Zmuidinaite R, Lacey JC, Jardine CS, Iles RK. A rapid and affordable screening tool for early-stage ovarian cancer detection based on MALDI-ToF MS of blood serum. Appl Sci. Mar 16, 2022;12(6):3030. [CrossRef]
- Jeong S, Son DS, Cho M, Lee N, Song W, Shin S, et al. Evaluation of combined cancer markers with lactate dehydrogenase and application of machine learning algorithms for differentiating benign disease from malignant ovarian cancer. Cancer Control. 2021;28:10732748211033401. [FREE Full text] [CrossRef] [Medline]
- Lu M, Fan Z, Xu B, Chen L, Zheng X, Li J, et al. Using machine learning to predict ovarian cancer. Int J Med Inform. Sep 2020;141:104195. [CrossRef] [Medline]
- Banaei N, Moshfegh J, Mohseni-Kabir A, Houghton JM, Sun Y, Kim B. Machine learning algorithms enhance the specificity of cancer biomarker detection using SERS-based immunoassays in microfluidic chips. RSC Adv. Jan 14, 2019;9(4):1859-1868. [FREE Full text] [CrossRef] [Medline]
- Whitwell HJ, Blyuss O, Menon U, Timms JF, Zaikin A. Parenclitic networks for predicting ovarian cancer. Oncotarget. Apr 27, 2018;9(32):22717-22726. [FREE Full text] [CrossRef] [Medline]
- Ivanova OM, Ziganshin RH, Arapidi GP, Kovalchuk SI, Azarkin IV, Sorokina AV, et al. Scope and limitations of MALDI-TOF MS blood serum peptide profiling in cancer diagnostics. Russ J Bioorg Chem. Sep 27, 2016;42(5):497-505. [CrossRef]
- Jiang W, Huang R, Duan C, Fu L, Xi Y, Yang Y, et al. Identification of five serum protein markers for detection of ovarian cancer by antibody arrays. PLoS One. 2013;8(10):e76795. [FREE Full text] [CrossRef] [Medline]
- Yang J, Zhu Y, Guo H, Wang X, Gao R, Zhang L, et al. Identifying serum biomarkers for ovarian cancer by screening with surface-enhanced laser desorption/ionization mass spectrometry and the artificial neural network. Int J Gynecol Cancer. May 2013;23(4):667-672. [CrossRef] [Medline]
- Shan L, Chen YA, Davis L, Han G, Zhu W, Molina AD, et al. Measurement of phospholipids may improve diagnostic accuracy in ovarian cancer. PLoS One. 2012;7(10):e46846. [FREE Full text] [CrossRef] [Medline]
- Thakur A, Mishra V, Jain SK. Feed forward artificial neural network: tool for early detection of ovarian cancer. Sci Pharm. 2011;79(3):493-505. [FREE Full text] [CrossRef] [Medline]
- Donach M, Yu Y, Artioli G, Banna G, Feng W, Bast RC, et al. Combined use of biomarkers for detection of ovarian cancer in high-risk women. Tumour Biol. Jun 2010;31(3):209-215. [FREE Full text] [CrossRef] [Medline]
- Ziganshin RK, Alekseev DG, Arapidi GP, Ivanov VT, Moshkovskiĭ SA, Govorun VM. Serum proteome profiling for ovarion cancer diagnosis using ClinProt magnetic bead technique and MALDI-TOF-mass-spectrometry [Article in Russian]. Biomed Khim. 2008;54(4):408-419. [Medline]
- Liu C, Shea N, Rucker S, Harvey L, Russo P, Saul R, et al. Proteomic patterns for classification of ovarian cancer and CTCL serum samples utilizing peak pairs indicative of post-translational modifications. Proteomics. Nov 2007;7(22):4045-4052. [CrossRef] [Medline]
- Zhang Z, Yu Y, Xu F, Berchuck A, van Haaften-Day C, Havrilesky LJ, et al. Combining multiple serum tumor markers improves detection of stage I epithelial ovarian cancer. Gynecol Oncol. Dec 2007;107(3):526-531. [FREE Full text] [CrossRef] [Medline]
- Chatterjee M, Mohapatra S, Ionan A, Bawa G, Ali-Fehmi R, Wang X, et al. Diagnostic markers of ovarian cancer by high-throughput antigen cloning and detection on arrays. Cancer Res. Jan 15, 2006;66(2):1181-1190. [FREE Full text] [CrossRef] [Medline]
- Lin YW, Lin CY, Lai HC, Chiou JY, Chang CC, Yu MH, et al. Plasma proteomic pattern as biomarkers for ovarian cancer. Int J Gynecol Cancer. 2006;16 Suppl 1:139-146. [CrossRef] [Medline]
- Liu Y. Serum proteomic pattern analysis for early cancer detection. Technol Cancer Res Treat. Feb 2006;5(1):61-66. [FREE Full text] [CrossRef] [Medline]
- Wu SP, Lin YW, Lai HC, Chu TY, Kuo YL, Liu HS. SELDI-TOF MS profiling of plasma proteins in ovarian cancer. Taiwan J Obstet Gynecol. Mar 2006;45(1):26-32. [FREE Full text] [CrossRef] [Medline]
- Li J, Ramamohanarao K. A tree-based approach to the discovery of diagnostic biomarkers for ovarian cancer. In: Proceedings of the 8th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining. 2004. Presented at: PAKDD '04; May 26-28, 2004:682-691; Sydney, Australia. URL: https://link.springer.com/chapter/10.1007/978-3-540-24775-3_80 [CrossRef]
- Li L, Tang H, Wu Z, Gong J, Gruidl M, Zou J, et al. Data mining techniques for cancer detection using serum proteomic profiling. Artif Intell Med. Oct 2004;32(2):71-83. [CrossRef] [Medline]
- Zhang Z, Barnhill SD, Zhang H, Xu F, Yu Y, Jacobs I, et al. Combination of multiple serum markers using an artificial neural network to improve specificity in discriminating malignant from benign pelvic masses. Gynecol Oncol. Apr 1999;73(1):56-61. [CrossRef] [Medline]
- Wilding P, Morgan MA, Grygotis AE, Shoffner MA, Rosato EF. Application of backpropagation neural networks to diagnosis of breast and ovarian cancer. Cancer Lett. Mar 15, 1994;77(2-3):145-153. [CrossRef] [Medline]
- Xue P, Wang J, Qin D, Yan H, Qu Y, Seery S, et al. Deep learning in image-based breast and cervical cancer detection: a systematic review and meta-analysis. NPJ Digit Med. Feb 15, 2022;5(1):19. [FREE Full text] [CrossRef] [Medline]
- Sounderajah V, Ashrafian H, Rose S, Shah NH, Ghassemi M, Golub R, et al. A quality assessment tool for artificial intelligence-centered diagnostic test accuracy studies: QUADAS-AI. Nat Med. Oct 2021;27(10):1663-1665. [CrossRef] [Medline]
- Whiting PF, Rutjes AW, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. Oct 18, 2011;155(8):529-536. [FREE Full text] [CrossRef] [Medline]
- Xu HL, Gong TT, Liu FH, Chen HY, Xiao Q, Hou Y, et al. Artificial intelligence performance in image-based ovarian cancer identification: a systematic review and meta-analysis. EClinicalMedicine. Nov 2022;53:101662. [FREE Full text] [CrossRef] [Medline]
- Jones CM, Athanasiou T. Summary receiver operating characteristic curve analysis techniques in the evaluation of diagnostic tests. Ann Thorac Surg. Jan 2005;79(1):16-20. [CrossRef] [Medline]
- Khatami F, Saatchi M, Zadeh SS, Aghamir ZS, Shabestari AN, Reis LO, et al. A meta-analysis of accuracy and sensitivity of chest CT and RT-PCR in COVID-19 diagnosis. Sci Rep. Dec 28, 2020;10(1):22402. [FREE Full text] [CrossRef] [Medline]
- Jones CM, Ashrafian H, Skapinakis P, Arora S, Darzi A, Dimopoulos K, et al. Diagnostic accuracy meta-analysis: a review of the basic principles of interpretation and application. Int J Cardiol. Apr 15, 2010;140(2):138-144. [CrossRef] [Medline]
- Bhinder B, Gilvary C, Madhukar NS, Elemento O. Artificial intelligence in cancer research and precision medicine. Cancer Discov. Apr 2021;11(4):900-915. [FREE Full text] [CrossRef] [Medline]
- Schlattmann P. Tutorial: statistical methods for the meta-analysis of diagnostic test accuracy studies. Clin Chem Lab Med. Apr 25, 2023;61(5):777-794. [FREE Full text] [CrossRef] [Medline]
- Huedo-Medina TB, Sánchez-Meca J, Marín-Martínez F, Botella J. Assessing heterogeneity in meta-analysis: Q statistic or I2 index? Psychol Methods. Jun 2006;11(2):193-206. [CrossRef] [Medline]
- Deeks JJ, Macaskill P, Irwig L. The performance of tests of publication bias and other sample size effects in systematic reviews of diagnostic test accuracy was assessed. J Clin Epidemiol. Sep 2005;58(9):882-893. [CrossRef] [Medline]
- Phillips B, Stewart LA, Sutton AJ. 'Cross hairs' plots for diagnostic meta-analysis. Res Synth Methods. Jul 2010;1(3-4):308-315. [CrossRef] [Medline]
- Jing B, Chen G, Yang M, Zhang Z, Zhang Y, Zhang J, et al. Development of prediction model to estimate future risk of ovarian lesions: a multi-center retrospective study. Prev Med Rep. Oct 2023;35:102296. [FREE Full text] [CrossRef] [Medline]
- Lu H, Liu Y, Wang J, Fu S, Wang L, Huang C, et al. Detection of ovarian cancer using plasma cell-free DNA methylomes. Clin Epigenetics. Jun 09, 2022;14(1):74. [FREE Full text] [CrossRef] [Medline]
- Kawakami E, Tabata J, Yanaihara N, Ishikawa T, Koseki K, Iida Y, et al. Application of artificial intelligence for preoperative diagnostic and prognostic prediction in epithelial ovarian cancer based on blood biomarkers. Clin Cancer Res. May 15, 2019;25(10):3006-3015. [CrossRef] [Medline]
- Song HJ, Yang ES, Kim JD, Park CY, Kyung MS, Kim YS. Best serum biomarker combination for ovarian cancer classification. Biomed Eng Online. Nov 06, 2018;17(Suppl 2):152. [FREE Full text] [CrossRef] [Medline]
- Elias KM, Fendler W, Stawiski K, Fiascone SJ, Vitonis AF, Berkowitz RS, et al. Diagnostic potential for a serum miRNA neural network for detection of ovarian cancer. Elife. Oct 31, 2017;6:e28932. [FREE Full text] [CrossRef] [Medline]
- González G, Lakatos K, Hoballah J, Fritz-Klaus R, Al-Johani L, Brooker J, et al. Characterization of cell-bound CA125 on immune cell subtypes of ovarian cancer patients using a novel imaging platform. Cancers (Basel). Apr 25, 2021;13(9):2072. [FREE Full text] [CrossRef] [Medline]
- Gao Q, Lin YP, Li BS, Wang GQ, Dong LQ, Shen BY, et al. Unintrusive multi-cancer detection by circulating cell-free DNA methylation sequencing (THUNDER): development and independent validation studies. Ann Oncol. May 2023;34(5):486-495. [FREE Full text] [CrossRef] [Medline]
- Ramamoorthy S, Sundaramoorthy S, Gupta A, Siddiqui Z, Sagar G, Rao KV, et al. Beta-validation of a non-invasive method for simultaneous detection of early-stage female-specific cancers. medRxiv. Preprint posted online October 6, 2023. [FREE Full text] [CrossRef]
- Luan Y, Zhong G, Li S, Wu W, Liu X, Zhu D, et al. A panel of seven protein tumour markers for effective and affordable multi-cancer early detection by artificial intelligence: a large-scale and multicentre case-control study. EClinicalMedicine. Jul 2023;61:102041. [FREE Full text] [CrossRef] [Medline]
- Marcišauskas S, Ulfenborg B, Kristjansdottir B, Waldemarson S, Sundfeldt K. Univariate and classification analysis reveals potential diagnostic biomarkers for early stage ovarian cancer Type 1 and Type 2. J Proteomics. Mar 30, 2019;196:57-68. [CrossRef] [Medline]
- Chiappa V, Interlenghi M, Bogani G, Salvatore C, Bertolina F, Sarpietro G, et al. A decision support system based on radiomics and machine learning to predict the risk of malignancy of ovarian masses from transvaginal ultrasonography and serum CA-125. Eur Radiol Exp. Jul 26, 2021;5(1):28. [FREE Full text] [CrossRef] [Medline]
- Hwangbo S, Kim SI, Kim JH, Eoh KJ, Lee C, Kim YT, et al. Development of machine learning models to predict platinum sensitivity of high-grade serous ovarian carcinoma. Cancers (Basel). Apr 14, 2021;13(8):1875. [FREE Full text] [CrossRef] [Medline]
- Akazawa M, Hashimoto K. Artificial intelligence in ovarian cancer diagnosis. Anticancer Res. Aug 2020;40(8):4795-4800. [CrossRef] [Medline]
- Wu M, Zhao Y, Dong X, Jin Y, Cheng S, Zhang N, et al. Artificial intelligence-based preoperative prediction system for diagnosis and prognosis in epithelial ovarian cancer: a multicenter study. Front Oncol. 2022;12:975703. [FREE Full text] [CrossRef] [Medline]
- Yue Z, Sun C, Chen F, Zhang Y, Xu W, Shabbir S, et al. Machine learning-based LIBS spectrum analysis of human blood plasma allows ovarian cancer diagnosis. Biomed Opt Express. May 01, 2021;12(5):2559-2574. [FREE Full text] [CrossRef] [Medline]
- Deo RC. Machine learning in medicine. Circulation. Nov 17, 2015;132(20):1920-1930. [FREE Full text] [CrossRef] [Medline]
- Handelman GS, Kok HK, Chandra RV, Razavi AH, Lee MJ, Asadi H. eDoctor: machine learning and the future of medicine. J Intern Med. Dec 2018;284(6):603-619. [FREE Full text] [CrossRef] [Medline]
- Leeflang MM. Systematic reviews and meta-analyses of diagnostic test accuracy. Clin Microbiol Infect. Feb 2014;20(2):105-113. [FREE Full text] [CrossRef] [Medline]
- Wang SC, Nickel G, Venkatesh KP, Raza MM, Kvedar JC. AI-based diabetes care: risk prediction models and implementation concerns. NPJ Digit Med. Feb 15, 2024;7(1):36. [FREE Full text] [CrossRef] [Medline]
- Hsu W, Hippe DS, Nakhaei N, Wang PC, Zhu B, Siu N, et al. External validation of an ensemble model for automated mammography interpretation by artificial intelligence. JAMA Netw Open. Nov 01, 2022;5(11):e2242343. [FREE Full text] [CrossRef] [Medline]
- Tomlinson E, Cooper C, Davenport C, Rutjes AW, Leeflang M, Mallett S, et al. Common challenges and suggestions for risk of bias tool development: a systematic review of methodological studies. J Clin Epidemiol. Jul 2024;171:111370. [FREE Full text] [CrossRef] [Medline]
- Wu D, Dou J, Chai X, Bellis C, Wilm A, Shih CC, et al. Large-scale whole-genome sequencing of three diverse Asian populations in Singapore. Cell. Oct 17, 2019;179(3):736-49.e15. [FREE Full text] [CrossRef] [Medline]
- Padilla CM, Kihal-Talantikit W, Perez S, Deguen S. Use of geographic indicators of healthcare, environment and socioeconomic factors to characterize environmental health disparities. Environ Health. Jul 22, 2016;15(1):79. [FREE Full text] [CrossRef] [Medline]
- Gibson E, Hu Y, Huisman HJ, Barratt DC. Designing image segmentation studies: statistical power, sample size and reference standard quality. Med Image Anal. Dec 2017;42:44-59. [FREE Full text] [CrossRef] [Medline]
- Gao F, Dai L, Wang Q, Liu C, Deng K, Cheng Z, et al. Blood-based biomarkers for Alzheimer's disease: a multicenter-based cross-sectional and longitudinal study in China. Sci Bull (Beijing). Aug 30, 2023;68(16):1800-1808. [FREE Full text] [CrossRef] [Medline]
- Zhu Y. Plasma/serum proteomics based on mass spectrometry. Protein Pept Lett. 2024;31(3):192-208. [FREE Full text] [CrossRef] [Medline]
- Geyer PE, Voytik E, Treit PV, Doll S, Kleinhempel A, Niu L, et al. Plasma proteome profiling to detect and avoid sample-related biases in biomarker studies. EMBO Mol Med. Nov 07, 2019;11(11):e10427. [FREE Full text] [CrossRef] [Medline]
- Zhang X, Takeuchi T, Takeda A, Mochizuki H, Nagai Y. Comparison of serum and plasma as a source of blood extracellular vesicles: increased levels of platelet-derived particles in serum extracellular vesicle fractions alter content profiles from plasma extracellular vesicle fractions. PLoS One. 2022;17(6):e0270634. [FREE Full text] [CrossRef] [Medline]
- Lan J, Núñez Galindo A, Doecke J, Fowler C, Martins RN, Rainey-Smith SR, et al. Systematic evaluation of the use of human plasma and serum for mass-spectrometry-based shotgun proteomics. J Proteome Res. Apr 06, 2018;17(4):1426-1435. [CrossRef] [Medline]
- Yu Z, Kastenmüller G, He Y, Belcredi P, Möller G, Prehn C, et al. Differences between human plasma and serum metabolite profiles. PLoS One. 2011;6(7):e21230. [FREE Full text] [CrossRef] [Medline]
- Malm L, Tybring G, Moritz T, Landin B, Galli J. Metabolomic quality assessment of EDTA plasma and serum samples. Biopreserv Biobank. Oct 2016;14(5):416-423. [CrossRef] [Medline]
- Ngiam KY, Khor IW. Big data and machine learning algorithms for health-care delivery. Lancet Oncol. May 2019;20(5):e262-e273. [CrossRef] [Medline]
- Parvatikar PP, Patil S, Khaparkhuntikar K, Patil S, Singh PK, Sahana R, et al. Artificial intelligence: machine learning approach for screening large database and drug discovery. Antiviral Res. Dec 2023;220:105740. [CrossRef] [Medline]
- Jiang Y, Yang M, Wang S, Li X, Sun Y. Emerging role of deep learning-based artificial intelligence in tumor pathology. Cancer Commun (Lond). Apr 2020;40(4):154-166. [FREE Full text] [CrossRef] [Medline]
- Wu M, Yan C, Liu H, Liu Q. Automatic classification of ovarian cancer types from cytological images using deep convolutional neural networks. Biosci Rep. Jun 29, 2018;38(3):BSR20180289. [FREE Full text] [CrossRef] [Medline]
- Teramoto A, Tsukamoto T, Kiriyama Y, Fujita H. Automated classification of lung cancer types from cytological images using deep convolutional neural networks. Biomed Res Int. 2017;2017:4067832. [FREE Full text] [CrossRef] [Medline]
- Mathivanan SK, Sonaimuthu S, Murugesan S, Rajadurai H, Shivahare BD, Shah MA. Employing deep learning and transfer learning for accurate brain tumor detection. Sci Rep. Mar 27, 2024;14(1):7232. [FREE Full text] [CrossRef] [Medline]
- Shamshirband S, Fathi M, Dehzangi A, Chronopoulos AT, Alinejad-Rokny H. A review on deep learning approaches in healthcare systems: taxonomies, challenges, and open issues. J Biomed Inform. Jan 2021;113:103627. [FREE Full text] [CrossRef] [Medline]
- Tung F, Mori G. Deep neural network compression by in-parallel pruning-quantization. IEEE Trans Pattern Anal Mach Intell. Mar 2020;42(3):568-579. [CrossRef] [Medline]
- Winchester LM, Harshfield EL, Shi L, Badhwar A, Khleifat AA, Clarke N, et al. Artificial intelligence for biomarker discovery in Alzheimer's disease and dementia. Alzheimers Dement. Dec 2023;19(12):5860-5871. [CrossRef] [Medline]
- Mårtensson G, Ferreira D, Granberg T, Cavallin L, Oppedal K, Padovani A, et al. The reliability of a deep learning model in clinical out-of-distribution MRI data: a multicohort study. Med Image Anal. Dec 2020;66:101714. [FREE Full text] [CrossRef] [Medline]
- Wang X, Chou K, Zhang G, Zuo Z, Zhang T, Zhou Y, et al. Breast cancer pre-clinical screening using infrared thermography and artificial intelligence: a prospective, multicentre, diagnostic accuracy cohort study. Int J Surg. Oct 01, 2023;109(10):3021-3031. [FREE Full text] [CrossRef] [Medline]
- Zeng Q, Klein C, Caruso S, Maille P, Allende DS, Mínguez B, et al. Artificial intelligence-based pathology as a biomarker of sensitivity to atezolizumab-bevacizumab in patients with hepatocellular carcinoma: a multicentre retrospective study. Lancet Oncol. Dec 2023;24(12):1411-1422. [CrossRef] [Medline]
- Akazawa M, Hashimoto K. Artificial intelligence in gynecologic cancers: current status and future challenges - a systematic review. Artif Intell Med. Oct 2021;120:102164. [CrossRef] [Medline]
- Parimbelli E, Wilk S, Cornet R, Sniatala P, Sniatala K, Glaser SL, et al. A review of AI and data science support for cancer management. Artif Intell Med. Jul 2021;117:102111. [FREE Full text] [CrossRef] [Medline]
- Narang A, Bae R, Hong H, Thomas Y, Surette S, Cadieu C, et al. Utility of a deep-learning algorithm to guide novices to acquire echocardiograms for limited diagnostic use. JAMA Cardiol. Jun 01, 2021;6(6):624-632. [FREE Full text] [CrossRef] [Medline]
- Xiang H, Xiao Y, Li F, Li C, Liu L, Deng T, et al. Development and validation of an interpretable model integrating multimodal information for improving ovarian cancer diagnosis. Nat Commun. Mar 27, 2024;15(1):2681. [FREE Full text] [CrossRef] [Medline]
- Nensa F, Demircioglu A, Rischpler C. Artificial intelligence in nuclear medicine. J Nucl Med. Sep 2019;60(Suppl 2):29S-37S. [FREE Full text] [CrossRef] [Medline]
- Nussberger AM, Luo L, Celis LE, Crockett MJ. Public attitudes value interpretability but prioritize accuracy in artificial intelligence. Nat Commun. Oct 03, 2022;13(1):5821. [FREE Full text] [CrossRef] [Medline]
- Aggarwal R, Farag S, Martin G, Ashrafian H, Darzi A. Patient perceptions on data sharing and applying artificial intelligence to health care data: cross-sectional survey. J Med Internet Res. Aug 26, 2021;23(8):e26162. [FREE Full text] [CrossRef] [Medline]
- Murdoch B. Privacy and artificial intelligence: challenges for protecting health information in a new era. BMC Med Ethics. Sep 15, 2021;22(1):122. [FREE Full text] [CrossRef] [Medline]
- Larson DB, Magnus DC, Lungren MP, Shah NH, Langlotz CP. Ethics of using and sharing clinical imaging data for artificial intelligence: a proposed framework. Radiology. Jun 2020;295(3):675-682. [CrossRef] [Medline]
- Sarayar R, Lestari YD, Setio AA, Sitompul R. Accuracy of artificial intelligence model for infectious keratitis classification: a systematic review and meta-analysis. Front Public Health. 2023;11:1239231. [FREE Full text] [CrossRef] [Medline]
Abbreviations
| AI: artificial intelligence |
| AUC: area under the curve |
| DL: deep learning |
| ML: machine learning |
| MOOSE: Meta-Analysis of Observational Studies in Epidemiology |
| OC: ovarian cancer |
| PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses |
| QUADAS-AI: Quality Assessment of Diagnostic Accuracy Studies–Artificial Intelligence |
| TN: true negatives |
| TP: true positives |
Edited by A Mavragani; submitted 24.10.24; peer-reviewed by Q Zhao, Z Jiang; comments to author 18.12.24; revised version received 06.01.25; accepted 22.01.25; published 24.03.25.
Copyright©He-Li Xu, Xiao-Ying Li, Ming-Qian Jia, Qi-Peng Ma, Ying-Hua Zhang, Fang-Hua Liu, Ying Qin, Yu-Han Chen, Yu Li, Xi-Yang Chen, Yi-Lin Xu, Dong-Run Li, Dong-Dong Wang, Dong-Hui Huang, Qian Xiao, Yu-Hong Zhao, Song Gao, Xue Qin, Tao Tao, Ting-Ting Gong, Qi-Jun Wu. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 24.03.2025.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research (ISSN 1438-8871), is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.