Machine learning in ovarian cancer: a bibliometric and visual analysis from 2004 to 2024.

After removing duplicates, 777 distinct records were retrieved from the WoSCC database. From 2004 to 2024, the volume of publications related to ML and OC has consistently increased. Notably, there was a significant rise in publications following 2020, as illustrated in Fig.  1 A. This increasing trend indicates that there is a growing academic and clinical interest in the application of ML to OC research. Fig. 1 Trends in annual publications in the field of ML in OC, 2004–2024. A Trends in publishing outputs, by year. B Distribution of corresponding authors countries and cooperation Trends in annual publications in the field of ML in OC, 2004–2024. A Trends in publishing outputs, by year. B Distribution of corresponding authors countries and cooperation An evaluation of the geographical distribution of corresponding authors revealed that China (n = 254) was the leading contributor, followed by the USA (n = 189), the United Kingdom (n = 45), Italy (n = 37), and India (n = 33). Furthermore, 62.2% of the publications from the United Kingdom involved multi-country collaborations (MCPs), as illustrated in Fig.  1 B and detailed in Table  2 . While China leads in publication volume, the United Kingdom demonstrates a more expansive and diverse international collaboration network, as shown in Fig.  2 A. Additionally, the collaboration network identifies Fudan University (n = 18) and the Chinese Academy of Sciences (n = 14) as key hubs of collaborative activity (see Fig.  2 B and Table  3 ). These findings indicate that Chinese scholars have a strong interest in applying ML to the treatment of OC. Table 2 Most relevant countries by corresponding authors of ML in OC Country Articles SCP MCP Freq MCP_Ratio China 254 222 32 0.327 0.126 USA 189 143 46 0.243 0.243 United Kingdom 45 17 28 0.058 0.622 Italy 37 26 11 0.048 0.297 India 33 23 10 0.042 0.303 Korea 27 20 7 0.035 0.259 Japan 25 20 5 0.032 0.2 Canada 16 12 4 0.021 0.25 Australia 14 4 10 0.018 0.714 Poland 14 12 2 0.018 0.143 Germany 11 7 4 0.014 0.364 Singapore 10 8 2 0.013 0.2 France 7 5 2 0.009 0.286 Iran 7 3 4 0.009 0.571 Belgium 6 1 5 0.008 0.833 Egypt 6 1 5 0.008 0.833 Finland 6 3 3 0.008 0.5 Greece 6 6 0 0.008 0 Netherlands 5 3 2 0.006 0.4 Spain 5 4 1 0.006 0.2 Note: MCP: Multiple country publication; SCP: Single country publication Fig. 2 Map of ML in OC Countries/Regions and Institutions, 2004–2024. A Map of cooperation between different countries. B Map of cooperation between different institutions Table 3 Top 10 most relevant affiliations of ML in OC Rank Affliation Articles (n) 1 Fudan Univ 18 2 Shanghai Jiao Tong Univ 16 3 Harvard Med Sch 15 3 Chinese Acad Sci 14 3 Zhejiang Univ 14 4 Nanjing Med Univ 13 4 Massachusetts Gen Hosp 12 5 Sun Yat Sen Univ 12 5 Univ Texas Md Anderson Canc Ctr 12 5 Southern Med Univ 11 Most relevant countries by corresponding authors of ML in OC Note: MCP: Multiple country publication; SCP: Single country publication Map of ML in OC Countries/Regions and Institutions, 2004–2024. A Map of cooperation between different countries. B Map of cooperation between different institutions Top 10 most relevant affiliations of ML in OC To investigate the journals that have made the most substantial contributions in terms of publication and citation within the domains of ML and OC, we utilized the Bibliometrix package in R software. Graphical representations were created using the ggplot2 package. Additionally, a co-citation analysis of the journals was conducted using VOSviewer. Our analysis identified a total of 777 documents distributed across 357 academic journals (see Annex 1 for comprehensive details). As shown in Table  4 and illustrated in Fig.  3 A, Gynecologic Oncology (n = 44, IF = 4.5) emerged as the leading publisher, followed by Cancers (n = 35, IF = 4.5), Scientific Reports (n = 31, IF = 3.8), Frontiers in Oncology (n = 23, IF = 3.5), and International Journal of Molecular Sciences (n = 10, IF = 4.9). Table 5 and Fig.  3 B highlight the most frequently cited journals (see Annex 2 for comprehensive details), with Gynecologic Oncology (n = 885, IF = 4.5), Nature (n = 658, IF = 50.5), Clinical Cancer Research (n = 612, IF = 10), Cancer Research (n = 551, IF = 12.5), and Journal of Clinical Oncology (n = 529, IF = 42.4) leading the list. Importantly, the co-citation map depicted in Fig.  4 illustrates that Gynecologic Oncology , Nature , and Clinical Cancer Research serve as pivotal collaboration hubs. These results collectively emphasize the significant influence of Gynecologic Oncology and Bioinformatics within the context of ML in OC research. Table 4 Top 10 journals with the most published Sources Documents IF(2023) Cites Gynecologic Oncology 44 4.5 885 Cancers 35 4.5 407 Scientific Reports 31 3.8 498 Frontiers In Oncology 23 3.5 369 International Journal Of Molecular Sciences 10 4.9 226 Bioinformatics 9 4.4 485 Frontiers In Immunology 9 5.7 168 Journal of Ovarian Research 9 3.8 167 Nature Communications 9 14.7 434 PLoS One 9 2.9 429 Fig. 3 Journal with the largest number of articles published and the journal with the largest number of citations. A Journal with the largest number of articles published. B Journals with the largest number of citations Table 5 Top 10 journals with the most cited Sources Cites IF(2023) Documents Gynecologic Oncology 885 4.5 44 Nature 658 50.5 0 Clinical Cancer Research 612 10 6 Cancer Research 551 12.5 3 Journal of Clinical Oncology 529 42.4 1 Scientific Reports 498 3.8 31 Bioinformatics 485 4.4 9 New England Journal of Medicine 437 94.3 0 Nature Communications 434 14.7 9 PLoS One 429 2.9 9 Fig. 4 ML in OC co-citation journals Top 10 journals with the most published Journal with the largest number of articles published and the journal with the largest number of citations. A Journal with the largest number of articles published. B Journals with the largest number of citations Top 10 journals with the most cited ML in OC co-citation journals We utilized the Bibliometrix package in R software to identify the top 20 most-cited references in the field of ML in OC (Table  6 ). The three most cited papers were: “A Review of Feature Selection Techniques in Bioinformatics,” “Calibration: The Achilles Heel of Predictive Analytics,” and “Genomic and molecular landscape of DNA damage repair deficiency across the cancer genome atlas”. Table 6 The top 20 cited references related to the ML in OC Paper DOI Total Citations TC per Year SAEYS Y, 2007, BIOINFORMATICS 10.1093/bioinformatics/btm344 3380 177.89 VAN CALSTER B, 2019, BMC MED 10.1186/s12916-019–1466-7 834 119.14 KNIJNENBURG TA, 2018, CELL REP 10.1016/j.celrep.2018.03.076 719 89.88 KOZAK KR, 2005, PROTEOMICS 10.1002/pmic.200500093 251 11.95 MAJUMDER B, 2015, NAT COMMUN 10.1038/ncomms7169 223 20.27 DENG LJ, 2022, MOL CANCER 10.1186/s12943-022–01510-2 203 50.75 GULHAN DC, 2019, NAT GENET 10.1038/s41588-019–0390-2 197 28.14 BIRGISDOTTIR V, 2006, BREAST CANCER RES 10.1186/bcr1522 186 9.3 WARD DG, 2006, BRIT J CANCER 10.1038/sj.bjc.6603188 181 9.05 LIANG MX, 2015, IEEE ACM T COMPUT BI 10.1109/TCBB.2014.2377729 180 16.36 CHATTERJEE M, 2006, CANCER RES 10.1158/0008–5472.CAN-04–2962 179 8.95 GRUS FH, 2005, INVEST OPHTH VIS SCI 10.1167/iovs.04–0448 172 8.19 KRISHNAN AV, 2010, ENDOCRINOLOGY 10.1210/en.2009–0855 154 9.63 KATOH M, 2020, INT J MOL MED 10.3892/ijmm.2019.4418 152 25.33 DLAMINI Z, 2020, COMPUT STRUCT BIOTEC 10.1016/j.csbj.2020.08.019 138 23 LU HN, 2019, NAT COMMUN 10.1038/s41467-019–08718-9 129 18.43 KAWAKAMI E, 2019, CLIN CANCER RES 10.1158/1078–0432.CCR-18–3378 123 17.57 DESBOIS M, 2020, NAT COMMUN 10.1038/s41467-020–19408-2 122 20.33 WEI SH, 2006, CLIN CANCER RES 10.1158/1078–0432.CCR-05–1551 119 5.95 HARTMANN LC, 2005, CLIN CANCER RES 10.1158/1078–0432.CCR-04–1673 115 5.48 The top 20 cited references related to the ML in OC To delve deeper into the exploration of the forefront and focal areas of ML in OC, we utilized CiteSpace to identify the top 15 most significant citation bursts related to ML in OC (refer to Fig.  5 ). The titles of these citations, along with their respective DOIs, are listed in Annex 3 . Remarkably, the three citations exhibiting the most pronounced citation bursts were: (a) “Use of Proteomic Patterns in Serum to Identify OC (strength: 14.25)”; (b) “OC Statistics, 2018 (strength: 7.89)”; and (c) “Integrated genomic analyses of ovarian carcinoma (strength: 6.42)”. Furthermore, the titles of the three most cutting-edge citation bursts were: (a) “Clinical-Grade Computational Pathology Using Weakly Supervised Deep Learning on Whole Slide Images”; (b) “MRI‐based machine learning for differentiating borderline from malignant epithelial ovarian tumors: A multicenter study”; and (c) “Radiomics of high-grade serous ovarian cancer: association between quantitative CT features, residual tumour and disease progression within 12 months”. Fig. 5 Top 15 references with the strongest citation bursts on ML in OC Top 15 references with the strongest citation bursts on ML in OC In general, through the analysis of cited literature and citation bursts, we identified three key research focuses in the field of ML and OC: (a) the discovery of OC biomarkers driven by ML and the development of personalized treatment strategies; (b) the application of ML in analyzing the OC tumor microenvironment and predicting treatment resistance; and (c) the application of ML in imaging-based OC diagnosis and risk stratification. Keyword clusters are essential for quickly grasping the primary research themes and trends within a specific area. In our study, we utilized VOSviewer to identify 3249 keywords. Table 7 presents the top 20 keywords that appear more than 22 times, highlighting the main hotspots in the research. The most frequently occurring keyword was "classification" (n = 94), followed by " breast cancer " (n = 91), " diagnosis" (n = 81), "survival analysis" (n = 81), "expression" (n = 74), and " biomarkers " (n = 72). Table 7 The top 20 keywords related to ML in OC Rank Keywords Count 1 Classification 94 2 Breast cancer 91 3 Diagnosis 81 4 Survival analysis 81 5 Expression 74 6 Biomarkers 72 7 Prediction 61 8 Prognosis 57 9 Chemotherapy 54 10 Risk 51 11 Models 47 12 Identification 41 13 Prostate cancer 40 14 Radiomics 39 15 Cells 37 16 Women 33 17 Mass spectrometry 32 18 Ca-125 30 19 Protein 29 20 Proteomics 28 The top 20 keywords related to ML in OC Through cluster analysis, we observe six distinct colored clusters in Fig.  6 . (a) The use of ML in exploring molecular mechanisms and precision medicine for OC treatment and prognosis (red dots) features 35 keywords, including expression, prognosis, immunotherapy, genes, and bioinformatics. (b) Prediction of genetic heterogeneity and treatment resistance in OC using ML (green dots) comprises 33 keywords, such as survival analysis, neoadjuvant chemotherapy, recurrence, mortality, and outcomes. (c) ML in personalized medicine and precision therapy for OC (blue dots) comprises 29 keywords, such as chemotherapy, resistance, validation, bevacizumab, and association. (d) ML in imaging diagnosis and preoperative evaluation in the field of OC (yellow dots) comprises 23 keywords, such as diagnosis, risk, models, radiomics, and ultrasonography. (e) ML in biomarker identification and multi-omics research in OC (purple dots) comprises 18 keywords, such as biomarkers, identification, mass spectrometry, CA-125, and protein. (f) Applications of ML and artificial intelligence in classification and prediction in OC research (cyan dots) comprises 18 keywords, such as classification, prediction, discovery, algorithm, and patterns. All keywords in the clusters are listed in Annex 4 . Fig. 6 Keywords co-occurrence map of publications on ML in OC research Keywords co-occurrence map of publications on ML in OC research Additionally, to project upcoming trends within this domain, we employed the bibliometrix toolkit within the R programming environment to create a dynamic thematic progression chart (Fig.  7 ). From 2005 to 2019, the application of ML in cancer research evolved from basic proteomic pattern recognition to the exploration of disease mechanisms and biomarkers. In the early phase (2005–2006), researchers concentrated on identifying cancer-associated proteins through proteomic analysis and initial ML models, thereby establishing a foundation for diagnosis. In 2007–2008, research shifted focus to artificial neural networks and biomarker discovery, employing neural network models for large-scale data analysis to enhance OC screening. From 2009 to 2017, the research expanded to identify serum markers and gene expression, with ML aiding in the detection of early cancer signs. In 2018–2019, the focus turned to pathological mechanisms, mass spectrometry, and network analysis, using these tools to uncover the mechanisms and biomarkers of OC. From 2018 to 2021, validation and algorithm optimization became central themes, emphasizing the enhancement of ML algorithms for clinical applications. The research also prioritized medical image analysis and gene expression data classification to improve early detection and diagnostic accuracy. From 2022 to 2024, studies explored how ML could predict the stage and prognosis of OC. Currently, artificial intelligence and multicenter research are key areas of investigation, reflecting a pursuit of complex algorithms and their robustness and applicability in large-scale clinical trials. Fig. 7 Temporal Distribution and Proportion of Top 30 Terms on ML in OC Research Temporal Distribution and Proportion of Top 30 Terms on ML in OC Research In summary, our comprehensive analysis, which includes citation burst detection, keyword frequency analysis, keyword clustering, and thematic evolution, has revealed emerging research frontiers at the intersection of ML and OC. Our findings indicate that the research hotspots in this field primarily focus on four key directions: (a) the discovery of OC biomarkers driven by ML and the development of personalized treatment strategies, (b) the application of ML in analyzing the OC tumor microenvironment and predicting treatment resistance, (c) the application of ML in imaging-based OC diagnosis and risk stratification, and (d) the application of ML in multicenter studies within the OC field.

The data for this study were obtained from the WoSCC database (Guangxi Medical University Purchase Edition) on December 31st, 2024. The search strategy is detailed in Table  1 . All retrieved articles were saved in plain text format and exported as full records, including cited references. Table 1 The research retrieval fomular of the field of ML in OC Step Search expression Results #1 TS = ("Ovarian cancer*" OR "Ovarian Neoplasm*" OR "Ovary Neoplasm*" OR "Ovary Cancer*" OR "Cancer of the Ovary" OR "Cancer of Ovary") 99,566 #2 TS = ("machine learning" OR "Learning Machine" OR "Artificial intelligence" OR AI OR "Transfer Learning" OR "Learning Transfer" OR "Deep learning" OR "Neural networks" OR "Unsupervised learning" OR "Supervised learning" OR "Reinforcement learning") 779,200 #3 (((#1 AND #2) AND DOP = (2004–01-01/2024–12-31)) AND DT = (Article OR Review)) AND LA = (English) 777 The research retrieval fomular of the field of ML in OC The methodology employed in this study was based on previous research conducted by Li et al. [ 21 , 22 ]. To analyze annual publication trends, Origin 2018 was used. Additionally, the analysis utilized R software (version 3.6.3) with the bibliometrix package (version 4.0, http://www.bibliometrix.org ) [ 23 , 24 ], VOSviewer (version 1.6.17) [ 25 ], and CiteSpace (version 6.1.4) [ 26 ]. To ensure the accuracy and reliability of the data, two independent authors conducted data extraction and analysis management separately. The bibliometrix package was utilized for visualizing and mapping scientific knowledge. VOSviewer was employed to construct visual representations of country and institutional co-authorship networks, source co-citation analysis, and keyword co-occurrence. For the co-authorship network, a minimum threshold was set to include countries or institutions with at least 5 publications. In the co-citation analysis, sources with a minimum of 50 citations were included. For keyword co-occurrence analysis, keywords with at least 5 occurrences were considered, while terms such as "ovarian cancer (OC)," "machine learning (ML)," "Artificial intelligence (AI)" and their synonyms were excluded [ 21 ]. Journal impact factors (IFs) were obtained from the 2023 edition of the Journal Citation Reports (JCR).

This study employed bibliometric analysis to comprehensively examine the application of ML in OC from 2004 to 2024. The analysis revealed global developmental trends, research hotspots, and future directions in this domain, highlighting the potential of ML to enhance early diagnosis, treatment, and prognosis of OC. Based on the findings of this study, the following recommendations are proposed to guide future research on the application of ML in OC. Firstly, future research should prioritize the development of interpretable ML models to enhance transparency and provide insights into the decision-making process. This will help clinicians and patients better understand the basis for model predictions, thereby addressing the “black box” challenge often associated with complex algorithms. Secondly, future studies should expand international collaboration by including diverse patient populations and datasets from multiple regions. This approach will not only enhance the robustness and generalizability of ML models but also promote the development of globally applicable diagnostic and prognostic tools. Third, prospective and large-scale validation studies are essential. Future research should focus on validating ML models in real-world clinical settings through prospective, large-scale trials to ensure their accuracy and reliability in clinical practice. Finally, translating ML models into real-world clinical applications remains a key priority. Although numerous studies have demonstrated the potential of ML in OC research, its actual clinical application remains limited. Future research should focus on pilot studies and randomized controlled trials to assess the impact of ML on clinical outcomes and patient care, thereby bridging the gap between theoretical potential and practical implementation.

With the advancement of medical big data and computational technologies, ML has been extensively utilized in cancer research, and the OC field is no exception. To delve into the research emphases and trends of ML in OC, we conducted a bibliometric and visualization analysis of relevant studies from 2004 to 2024, encompassing a total of 777 papers published during this period. Our findings indicate a notable upsurge in the number of papers on ML in OC since 2020, reflecting a growing interest in applying ML within the academic and clinical research communities. However, the volume of publications in this domain remains comparatively lower than in other cancer areas [ 10 , 27 ], suggesting that research in this field is still in its nascent stages, with ample opportunities for growth. Geographically, China has emerged as the country with the highest number of published papers. This trend is similar to the ML research observed in several other cancer types [ 28 , 29 ]. Notably, other leading countries in terms of publications include the United States, the United Kingdom, Italy, and India. While China leads in the volume of literature published, the United Kingdom demonstrates a broader and more diverse international collaboration network, with Fudan University and the Chinese Academy of Sciences serving as key hubs for collaboration. These results indicate that Chinese scholars have shown a strong interest in the application of ML to OC treatment. The 777 papers are distributed across 357 journals, with Cancers , Scientific Reports , Frontiers in Oncology , Gynecologic Oncology , and Bioinformatics being the top five journals in terms of publication volume. Notably, the top five journals based on citation counts include Gynecologic Oncology , Nature , Clinical Cancer Research , Bioinformatics , and Scientific Reports , underscoring the significance of these journals in the research field. Our comprehensive analysis, which includes citation burst detection, keyword frequency analysis, keyword clustering, and topic evolution, reveals emerging research frontiers at the intersection of ML and OC. The results indicate that research hotspots in this field are primarily concentrated in four key directions: (a) the discovery of OC biomarkers driven by ML and the development of personalized treatment strategies; (b) the application of ML in analyzing the OC tumor microenvironment and predicting treatment resistance; (c) the application of ML in imaging-based OC diagnosis and risk stratification; and (d) the application of ML in multicenter studies within the OC field. Cancer antigen 125 (CA125) is a commonly used biomarker for OC, with an elevated CA125 concentration (≥ 30 U/mL) serving as a predictive indicator for subsequent OC risk [ 30 ]. However, elevated CA125 levels in the range of 35–65 U/mL are challenging to distinguish between early-stage OC and benign gynecological conditions, such as endometriosis, since benign diseases can also elevate tumor marker levels [ 31 ]. Studies have shown that even the combined screening of vaginal ultrasound and CA125 has not been effective in reducing the mortality rate of OC, and there is a certain rate of false positives [ 32 ]. This indicates that the role of CA125 in OC screening is limited, highlighting the need for biomarkers with higher accuracy and specificity. ML algorithms can analyze vast amounts of biomedical data to identify novel biomarkers associated with OC and provide personalized treatment plans for patients. For instance, Gu et al. [ 33 ] found that OC exhibits a unique postprandial serum CA125 increase compared to benign ovarian diseases. Based on this finding, they constructed an SVM-based CA125 increment algorithm, reportedly achieving a sensitivity of 91.7% and a specificity of 99.2% in detecting early-stage OC. Similarly, Jerry Z et al. [ 34 ] screened 34 metabolites that showed significant differences in the urine of healthy adults and OC patients, establishing a classification model that achieved a maximum accuracy rate of 85.29%. Additionally, ML models play a crucial role in predicting the response of OC patients to chemotherapy drugs, such as platinum, aiding physicians in selecting more effective personalized treatment methods [ 15 , 35 , 36 ]. However, the "black box" issue in using ML to discover OC biomarkers limits practical application, as physicians and patients need to understand the basis of the model's predictions; thus, model interpretability is vital. To address this challenge, an increasing number of studies are applying Shapley analysis to quantify the importance of features in ML models, thereby enhancing their interpretability [ 37 , 38 ]. ML in the analysis of the OC tumor microenvironment and prediction of treatment resistance primarily involves analyzing a vast array of data from the tumor microenvironment, including genomic, transcriptomic, proteomic, and clinical data [ 4 , 39 , 40 ], to identify patterns and biomarkers associated with tumor growth, invasion, and treatment response. These patterns are crucial for forecasting treatment efficacy and patient prognosis. For instance, Zhao et al. [ 41 ] screened macrophage-related markers from The Cancer Genome Atlas OC dataset using weighted gene co-expression network analysis. These markers were then submitted to 10 ML algorithms to construct a prognostic prediction model, which reportedly outperformed traditional grading and staging in predicting overall survival rates in OC. Wu et al. [ 42 ] screened genes associated with the tumor microenvironment and prognosis from seven datasets, including The Cancer Genome Atlas. They submitted these genes to ML algorithms to develop a risk scoring model related to the tumor microenvironment. Patients with low scores exhibited BRCA1 mutations, immune activation, and a favorable immune response, while those with high scores were significantly associated with deficiencies in C-X-C motif chemokine ligands and the activation of oncogenic pathways. Current research indicates that ML models can effectively distinguish different subtypes of OC, assess the impact of the tumor microenvironment, and predict resistance to chemotherapy or immunotherapy [ 43 , 44 ]. However, ML also faces limitations in analyzing the OC tumor microenvironment and predicting therapeutic resistance [ 45 ]. Firstly, the accuracy of the model is highly dependent on the quality and quantity of the training data; biased or incomplete training data can compromise predictive power. Secondly, the selected genes related to the OC microenvironment require further in vitro and in vivo experiments to confirm their functions. Additionally, the heterogeneity and dynamic changes of the tumor microenvironment complicate predictions. Despite these challenges, advancements in technology and the availability of higher-quality data make the application prospects of ML in this field promising. By leveraging medical imaging data, such as computed tomography (CT), ultrasound imaging, and magnetic resonance imaging (MRI) [ 46 ], ML models can identify and extract features that are valuable for the diagnosis and prognostic prediction of OC. These models have shown great potential in enhancing the accuracy [ 15 , 47 , 48 ], efficiency, and prognostic assessment [53–54]of diagnostic procedures.Research indicates that ML models have made significant progress in imaging-based diagnosis of OC, pathological classification, guidance for targeted biopsies, and prognosis prediction [ 15 , 47 , 50 ]. For instance, through radiomic features, ML models can distinguish between different types of OC, such as epithelial OC and borderline epithelial ovarian tumors [ 51 ], thereby aiding clinical decision-making. Furthermore, by analyzing preoperative and postoperative imaging features, ML is also utilized to predict recurrence and prognosis of OC, as well as to forecast progression-free survival and overall survival for patients [ 52 – 54 ]. In terms of risk stratification, ML models can more accurately predict treatment response and prognosis for OC patients by integrating clinical, pathological, and radiomic data [ 49 , 50 , 55 ]. For example, by analyzing the imaging and pathological characteristics of tumors, quantitative features associated with prognosis can be identified, leading to more personalized treatment plans for patients [ 52 ]. Current research indicates that ML models can effectively distinguish different subtypes of OC, assess the impact of the tumor microenvironment, and predict resistance to chemotherapy or immunotherapy [ 43 , 44 ].Despite the significant potential of ML in the diagnosis and risk stratification of OC, most current studies are retrospective and have limited sample sizes. Future research needs to further test and verify the performance of these models in prospective, large-scale studies [ 56 ]. Additionally, enhancing the models' generalizability and interpretability will be a key focus for future research [ 57 ]. Overall, ML technology is playing an increasingly important role in the imaging diagnosis and risk stratification of OC and is expected to have a greater impact on future clinical practice. The application of ML in multicenter studies on OC is progressively expanding. For example, Gao et al. [ 47 ] developed a deep convolutional neural network model using pelvic ultrasound images from seven hospitals in China. They not only validated the model's accuracy on an internal validation set but also tested its generalization ability using data from two additional hospitals as an external validation set. Leng et al. [ 58 ] utilized data from patients with epithelial OC across three centers to develop an integrated model through ML algorithms. This model incorporated radiomic features along with clinical features to predict the FIGO staging of OC patients, demonstrating exceptional predictive performance in EOC staging, outperforming both the clinical feature model and the radiomic model. Li et al., through the analysis of MRI images from 501 confirmed OC cases across eight clinical centers, utilized ML algorithms to establish a model that distinguishes between borderline epithelial ovarian tumors and malignant ovarian tumors, demonstrating robust performance superior to the subjective assessment of radiologists [ 59 ]. Despite these advancements, multicenter studies in this domain also have certain limitations. Firstly, most current research is retrospective, necessitating prospective data for more compelling evidence. Secondly, there is a lack of international collaboration in current studies; incorporating data from patients of diverse ethnicities globally could enhance the model's generalizability and performance. In summary, multicenter collaboration is a significant trend for future applications of ML in the OC field, emphasizing the need for increased international cooperation and more prospective studies to improve model accuracy and generalizability. In this study, we conducted the first bibliometric analysis of the application of ML in the field of OC. Through the analysis of current and future trends in this field by this study, researchers and clinicians in this domain will be able to systematically understand the research priorities and future trends in this area. However, several limitations must be acknowledged. Firstly, the data were sourced exclusively from the WoSCC, which may lead to the omission of relevant literature from other databases and introduce potential biases. Additionally, the study was restricted to English-language publications, which may limit the comprehensiveness and representativeness of the findings. Thirdly, while bibliometric methods are effective in identifying research hotspots and trends, they have limitations in conducting in-depth analyses of specific study contents and quality. Therefore, future research should incorporate multiple databases (such as PubMed and Scopus) and include publications in various languages to provide a more comprehensive and unbiased overview of the research landscape.

OC is one of the most prevalent malignant tumors in women globally, characterized by poor prognosis and high mortality rates [ 1 ]. The mortality rate of OC is significantly correlated with the stage at diagnosis; for instance, the 5-year survival rate for women diagnosed at Stage I is 90% [ 2 ]. In cases of locally metastatic OC, this rate drops to approximately 80%, while those with distant metastasis face a further decline to around 25%. Regrettably, over 70% of OC cases are diagnosed at Stage III or IV, when the disease is already advanced [ 3 ].Therefore, the search for screening initiatives capable of identifying early-stage changes in OC is crucial for reducing mortality rates among patients. Moreover, the primary treatment for OC consists of cytoreductive surgery followed by adjuvant chemotherapy, with the first-line therapy typically involving a combination of platinum-based drugs and paclitaxel. The initial response rate to platinum-based chemotherapy ranges from 60 to 80%; however, some patients may develop drug resistance [ 4 ], and approximately 70% of those with platinum-resistant disease are likely to relapse within two years [ 5 ]. Before completing chemotherapy, the response of OC patients to platinum-based treatment remains uncertain, making the prediction of their response crucial. In summary, enhancing the survival rates of OC patients may require extensive research into various aspects, including early diagnosis, risk screening, and therapeutic decision-making [ 6 , 7 ]. Throughout this research process, a significant amount of biomedical data is generated. Efficiently integrating and analyzing this data to provide clinical assistance presents a contemporary challenge, and ML is a vital tool in addressing this issue. ML is a branch of artificial intelligence. In simple terms, ML algorithms analyze large volumes of data to mimic the human brain's processes of thinking, reasoning, and decision-making, ultimately addressing real-world issues [ 8 , 9 ]. With the advancement of medical big data and computer technology, ML has been extensively utilized in contemporary cancer research [ 10 – 13 ]. ML methods have demonstrated significant potential in improving the accuracy of predictions related to cancer susceptibility, recurrence, and survival, with evidence indicating that their application has enhanced the accuracy of cancer forecasting by 15% to 20% in recent years [ 14 ]. With the advancement of ML in other oncological domains, an increasing number of researchers are utilizing ML for early screening, diagnosis, and treatment decision-making in OC. For instance, a recently published review highlighted the use of ML methods such as logistic regression, extreme gradient boosting, and support vector machines to predict the response of OC patients to platinum-based chemotherapy [ 4 ]. Another review highlighted the strong performance of ML algorithms in diagnosing OC through medical imaging; however, further external validation is needed to assess their accuracy [ 15 ]. Due to the limited number of publications, current reviews may not fully explore the latest research trends and hotspots in the application of ML in OC. Additionally, there is a lack of quantitative analysis across the existing literature in this field. Therefore, summarizing and quantitatively analyzing the global development trends and research hotspots of ML in OC is crucial for guiding future research. Bibliometrics is an academic discipline that combines quantitative analysis (such as publication counts and citation frequencies) with qualitative analysis (such as thematic summarization) to analyze scholarly literature and its metadata (including authors, keywords, and citation relationships) [ 16 ]. Its primary objective is to elucidate the knowledge structure, developmental patterns, and underlying connections within a research domain [ 17 ]. Unlike traditional statistical methods that focus on relationships and causal inferences between variables, bibliometrics leverages techniques such as information visualization, knowledge mapping, co-word analysis, and co-citation analysis to intuitively depict the developmental trajectory, current status, hotspots, and trends of a research field, thereby providing a macroscopic overview of the research landscape. Bibliometrics has been widely applied for identifying research hotspots and predicting developmental trends across various disciplines [ 18 – 20 ]. Therefore, this study collects literature related to ML in the field of OC from the WoSCC database and employs bibliometric methods to quantitatively analyze the research process and current status over the past 20 years, while also predicting potential future research trends. This research will assist researchers and clinicians in this field in gaining a more systematic understanding of research focuses and upcoming trends.

Additional file 1: Annex 1. Publication Rank of Journals on ML in OC. Additional file 2: Annex 2. Citation Rank of Journals on ML in OC. Additional file 3: Annex 3. Titles and DOIs of the Top 15 Most Significant Citation Bursts. Additional file 4: Annex 4. Keywords in the Clusters. Additional file 1: Annex 1. Publication Rank of Journals on ML in OC. Additional file 2: Annex 2. Citation Rank of Journals on ML in OC. Additional file 3: Annex 3. Titles and DOIs of the Top 15 Most Significant Citation Bursts. Additional file 4: Annex 4. Keywords in the Clusters.