Machine learning in the early detection of endometriosis: a literature review on symptom clustering and imaging integration

Endometriosis is a gynecologic inflammatory condition that affects up to 10% of repro- ductive-aged women worldwide. The disease exhibits heterogeneous presentations and is associated with a prolonged diagnostic delay, often exceeding seven years, be- cause existing diagnostic modalities such as transvaginal ultrasound, magnetic reso- nance imaging, and the biomarker cancer antigen 125 (CA-125) are suboptimal. This re- view examines how machine learning (ML) is playing an increasingly significant role in early, non-surgical endometriosis diagnosis through two main approaches: symptom clustering and imaging integration. Unsupervised ML algorithms such as k-means, par- titioning around medoids, and Bayesian networks have demonstrated success in iden- tifying clinically informative endometriosis phenotypes from patient-reported symp- toms and electronic health records. Concurrently, ML models such as convolutional neural networks and radiomics approaches have achieved high accuracy in lesion de- tection from imaging data, in some cases surpassing human interpretation. Despite these advances, significant challenges remain, including limited access to large, anno- tated multimodal datasets, the absence of widely accepted evaluation standards, and concerns regarding interpretability and generalizability. Multicenter, integrative studies and the incorporation of explainability techniques are recommended as potential strat- egies to address these gaps. Finally, multimodal ML approaches that combine symp- tomatology and imaging data hold substantial promise for reducing diagnostic delays, facilitating early intervention, and improving clinical outcomes in the management of endometriosis.

Artificial intelligence; Endometriosis; Machine learning Precision and Future Medicine 2025;9(3):117-128 https://doi.org/10.23838/pfm.2025.00177 pISSN: 2508-7940 · eISSN: 2508-7959 1 / 1CROSSMARK_logo_3_Test 2017-03-16https://crossmark-cdn.crossref.org/widget/v2.0/logos/CROSSMARK_Color_square.svg Received: June 22, 2025 Revised: September 1, 2025 Accepted: September 6, 2025 Corresponding author: Shady Saud Khan General Medicine Practice Program, Batterjee Medical College, Jeddah 21442, Saudi Arabia Tel: +966-556550655 E-mail: [email protected] 118 http://pfmjournal.org Machine learning in endometriosis detection

Endometriosis is a chronic, inflammatory gynecological dis- order characterized by the presence of endometrial-like tis- sue outside the uterus, which leads to pain, infertility, and re- duced quality of life. It affects approximately 10% of repro- ductive-aged women worldwide, with some estimates sug- gesting an even higher prevalence due to underdiagnosis [1]. The most common symptoms include dysmenorrhea (painful menstruation), chronic pelvic pain, dyspareunia (pain during intercourse), and subfertility, although symptom severity does not always correlate with disease stage [2]. In addition to physical suffering, endometriosis significantly affects men- tal health, work productivity, and social relationships, with many patients reporting anxiety, depression, and reduced overall well-being [3]. The economic burden is also consider- able, with healthcare expenditures and productivity losses amounting to billions of dollars annually [4]. Given its wide- spread impact, enhancing diagnostic and therapeutic strate- gies is critical for mitigating the long-term consequences of this debilitating condition. Despite its prevalence, endometriosis remains notoriously difficult to diagnose, with an average delay of 7 to 10 years from symptom onset to definitive diagnosis [5]. The gold standard for diagnosis is laparoscopic surgery with histologi- cal confirmation, an invasive procedure that carries risks such as infection, bleeding, and complications related to anesthe- sia [6]. Noninvasive imaging techniques, including transvagi- nal ultrasound (TVUS) and magnetic resonance imaging (MRI), are frequently employed but demonstrate variable sensitivity and specificity depending on lesion type and oper- ator expertise [7]. Furthermore, biomarkers such as cancer antigen 125 (CA-125) lack sufficient accuracy for use as stand- alone diagnostic tools, leading to frequent misdiagnoses with conditions such as irritable bowel syndrome or pelvic inflam- matory disease [8]. These diagnostic limitations underscore the urgent need for more reliable and less invasive methods to reduce delays and improve patient outcomes. The pro- tracted diagnostic journey for endometriosis patients high- lights the importance of developing early, noninvasive diag- nostic tools. Early detection could prevent disease progres- sion, reduce chronic pain, and preserve fertility by enabling timely intervention [9]. Emerging research is investigating liq- uid biopsies, proteomics, and microRNA profiling as potential diagnostic strategies, although none have yet achieved wide- spread clinical adoption [10]. In addition, integrating pa- tient-reported symptoms with imaging findings may enhance diagnostic accuracy without immediate reliance on surgery [11]. Given the shortcomings of current methods, there is a pressing demand for innovative approaches that combine clinical data with advanced analytical techniques to stream- line diagnosis and improve patient care. Artificial intelligence (AI) and machine learning (ML) are transforming healthcare by enabling automated pattern rec- ognition, predictive modeling, and decision-support systems. In medical imaging, AI algorithms have demonstrated consid- erable success in detecting malignancies, cardiovascular ab- normalities, and neurological disorders [12]. These technolo- gies excel in processing large datasets, identifying subtle pat- terns, and enhancing diagnostic precision capabilities that could help address key challenges in endometriosis diagnosis [13]. For example, ML models trained on ultrasound or MRI datasets can improve lesion detection, while natural language processing can extract diagnostic information from electronic health records (EHRs) [14]. As AI continues to evolve, its appli- cation in endometriosis holds promise for reducing diagnos- tic delays and improving classification accuracy. This review aims to critically examine the current diagnos- tic challenges in endometriosis and explore the potential of AI-driven solutions to overcome these barriers. We evaluate the limitations of existing diagnostic tools, analyze recent de- velopments in biomarker research and imaging technologies, and discuss how AI can integrate multimodal data to improve accuracy. Furthermore, we highlight ongoing research, clini- cal applications, and future directions for AI in endometriosis care. By synthesizing the latest evidence, this review seeks to inform clinicians, researchers, and policymakers about the transformative potential of AI in addressing one of gynecolo- gy’s most persistent diagnostic challenges.

Electronic databases, including PubMed/MEDLINE, Web of Science, and Google Scholar, were searched for case reports, case series, case-control studies, cohort studies, cross-sec- tional studies, and review articles from each database’s incep- tion to May 2025. The search terms included ‘Endometriosis,’ ‘Machine learning,’ ‘Symptom clustering,’ and ‘Artificial intelli- gence,’ among others. Studies were included if they investi- gated the application of AI or ML to the diagnosis or early de- tection of endometriosis, reported original data (including case-control, cohort, cross-sectional studies, or randomized studies) or systematic/narrative reviews directly relevant to ML in endometriosis, used symptom-based, imaging-based, 119https://doi.org/10.23838/pfm.2025.00177 Ziad Mumtaz Ramadan, et al. or multimodal datasets; and were published in English. Both prospective and retrospective studies were eligible, and no restrictions were placed on patient age or geographic setting. The majority of the studies included were small to medi- um-sized, often retrospective single-center registries with less than 100 up to several hundred patients. The patient popula- tions were enrolled predominantly from the gynecology spe- cialty clinics and may have had limited generalizability to more prevalent community practice. In particular, most stud- ies defined endometriosis by surgical/pathologic diagnosis, the current gold standard for diagnosis, while others used clinical diagnosis or imaging study findings such as ultra- sound or MRI. This was conducted as a narrative review rather than a systematic review; therefore, no formal quality assess- ment or strict exclusion criteria were applied. Study inclusion was primarily based on relevance to the research question and the authors’ judgment of contribution to the evolving field. DIAGNOSTIC CHALLENGES IN ENDOMETRIOSIS Heterogeneity of symptoms One of the primary obstacles in diagnosing endometriosis is the wide variability in clinical presentation. While some pa- tients experience severe pelvic pain and infertility, others re- main asymptomatic, with lesions discovered incidentally during surgery for unrelated conditions [15]. Symptom pat- terns also differ by lesion location; for instance, deep infiltrat- ing endometriosis (DIE) may cause bowel or urinary symp- toms, whereas ovarian endometriomas are more commonly associated with dysmenorrhea [16]. This heterogeneity com- plicates diagnosis, as clinicians must distinguish endometrio- sis from other conditions with overlapping symptoms, such as adenomyosis, interstitial cystitis, and gastrointestinal (GI) disorders [17]. The absence of a definitive symptom profile necessitates a multidisciplinary diagnostic approach, further delaying identification and treatment. Average diagnostic delay (typically 7 to 10 years) The prolonged diagnostic delay in endometriosis typically ranging from 7 to 10 years reflects systemic shortcomings in awareness, clinical evaluation, and diagnostic practices [18]. Studies show that adolescents and young women often expe- rience even longer delays, as their symptoms are frequently dismissed as ‘normal’ menstrual pain [18,19]. Contributing factors include clinician bias, normalization of pelvic pain, and limited access to specialized care [20]. These delays per- mit disease progression, thereby increasing the risk of chronic pain, infertility, and surgical complications [21]. Reducing this delay requires enhanced clinician education, standardized symptom-assessment tools, and improved diagnostic tech- nologies to support early intervention.

of current diagnostic tools (ultrasound, MRI, and CA-125) Although TVUS and MRI are widely used for endometriosis detection, their diagnostic accuracy depends on lesion type, operator expertise, and imaging protocols [22]. TVUS is effec- tive for detecting ovarian endometriomas but is less sensitive for superficial peritoneal or deep infiltrating lesions [23]. MRI provides superior soft-tissue contrast but is costly and not universally accessible [24]. Serum biomarkers such as CA-125 are often elevated in endometriosis but lack specificity, as they may also increase in ovarian cancer, pregnancy, and oth- er inflammatory conditions [25]. These limitations under- score the need for more reliable and cost-effective diagnostic tools that can detect endometriosis across its diverse mani- festations without invasive procedures. Importance of integrating symptomatology and imaging Given the limitations of individual diagnostic methods, a mul- timodal approach that combines symptom assessment, im- aging, and biomarker analysis may improve diagnostic accu- racy. Recent studies suggest that structured symptom ques- tionnaires (e.g., the Endometriosis Symptom Diary) can aid in stratifying patients for further evaluation [26]. When paired with AI-enhanced imaging analysis, these tools could facilitate earlier identification of high-risk individuals [27]. For instance, ML algorithms can process ultrasound or MRI data to detect subtle lesions that may be overlooked by human evaluators [28]. Integrating patient history with AI-driven imaging inter- pretation has the potential to reduce reliance on laparoscopy, thereby enabling faster and less invasive diagnoses. Future research should prioritize validating such integrated models in diverse clinical settings to ensure broad applicability. MACHINE LEARNING IN MEDICINE: A PRIMER ML is a subfield of AI that encompasses computational meth- ods enabling computers to learn from medical records and prior experiences to categorize or predict outcomes without 120 http://pfmjournal.org Machine learning in endometriosis detection explicit programming [29,30]. The three primary types of ML are supervised learning, unsupervised learning, and deep learning. Deep learning is a more advanced approach that builds upon the first two paradigms [31,32]. To predict out- comes for new, unseen data, supervised learning requires al- gorithms trained on labeled datasets, in which the correct outputs are already known. This method is frequently applied in diagnostic contexts for tasks such as regression and classi- fication, where prior patient data are used to train models to differentiate between diseased and healthy states [31,33]. Unsupervised learning, by contrast, does not utilize labeled data but instead identifies hidden patterns or clusters. This approach is particularly valuable for uncovering phenotypes or disease subtypes that may not be immediately apparent to clinicians [31,34]. Deep learning, also referred to as deep neu- ral network learning, is especially powerful for analyzing medical images and signals, employing multilayered neural networks to automatically extract features and recognize complex patterns [35,36]. ML is increasingly applied in clinical settings because it can identify patterns within large and complex medical datasets. ML algorithms can assist clinicians in developing more effec- tive treatment strategies and making more accurate diagno- ses by analyzing patient information. For example, an ML model can use hospital data on symptoms and treatment outcomes to predict which therapies are most likely to be ef- fective for specific patient groups [35,37]. These algorithms, often referred to as classifiers, can accurately distinguish be- tween diagnostic categories depending on the model archi- tecture and the quality of the data [37,38]. ML is also used for risk prediction, employing algorithms such as support vector machines (SVMs) and convolutional neural networks (CNNs) to identify high-risk patients at earlier stages [31,38]. When ap- plied collectively, these approaches enable clinicians to detect diseases sooner, design personalized treatment plans, and achieve improved outcomes across diverse medical fields. One area where ML has demonstrated particular value is gynecology, where it enhances the accuracy and efficiency of disease detection, risk assessment, and clinical decision-mak- ing, especially in cancer diagnosis and pregnancy-related complications [39,40]. In cancer care, ML and deep learning models are applied to diagnose cervical, ovarian, and endo- metrial cancers through image analysis, biomarker discovery, and individualized risk prediction [41]. These models fre- quently outperform conventional diagnostic methods in sen- sitivity and accuracy, thereby facilitating earlier detection and reducing treatment delays [41-43]. Beyond oncology, ML also plays an important role in pre- dicting and managing a wide range of gynecological condi- tions. Algorithms such as decision tree ensembles and naïve Bayes classifiers have demonstrated high effectiveness in pre- dicting conditions including postpartum depression, anemia, preeclampsia, gestational diabetes, and miscarriage achiev- ing accuracies of up to 86% using symptom and clinical data [44-46]. In maternal-fetal medicine, ML facilitates the early identification of complications such as preterm birth, cesare- an delivery, and perinatal death, often through the integration of clinical metrics with imaging data [40,44]. Furthermore, AI and ML are transforming gynecologic imaging by automating image classification, assisting radiologists in their diagnostic tasks, and promoting greater consistency and objectivity in clinical decision-making [42,43]. These developments illus- trate how the fundamental principles of ML are being applied to integrate clinical care with computational methods, there- by achieving tangible improvements in women’s health. SYMPTOM CLUSTERING IN ENDOMETRIOSIS USING MACHINE LEARNING ML-based symptom clustering has emerged as a viable strate- gy for characterizing the complex clinical manifestations of endometriosis. Infertility, pelvic pain, dysmenorrhea, dyspa- reunia, and GI symptoms are frequently included as key vari- ables for clustering and prediction in endometriosis research (Table 1). While GI symptoms are often misattributed, contrib- uting to diagnostic delays, pelvic pain and dysmenorrhea are particularly prevalent in both adolescent and adult popula- tions. These symptoms are central to patient-reported out- comes as well as clinical data analyses [47-50]. Building on this symptom-based framework, unsupervised ML techniques such as multivariate mixture models, parti- tioning around medoids (PAM), and k-means clustering have been employed to identify subtypes within endometriosis populations. K-means clustering, for example, has been ap- plied to classify women according to quality-of-life parame- ters, revealing discrete groups with either high or low quality of life [51]. Similarly, analyses of adolescent EHR data using PAM and mixture models have identified ‘classic’ (pelvic pain, dysmenorrhea, chronic pain), ‘GI’ (gastrointestinal-domi- nant), and ‘feature-absent’ phenotypes [49]. In addition, ana- tomical pain sites have been mapped using Bayesian network analysis, which has also been applied to identify patterns in pain symptomatology [47]. Furthermore, mixed-membership 121https://doi.org/10.23838/pfm.2025.00177 Ziad Mumtaz Ramadan, et al. models have shown promise in handling multimodal, self- tracked symptom data, thereby improving the clinical rele- vance of identified subtypes [50]. Increasingly, studies incorporate EHR data and patient-re- ported outcomes obtained from large-scale surveys and symptom-tracking applications to support these computa- tional approaches and capture the full heterogeneity of endo- metriosis. These data sources enable researchers to identify phenotypes directly from real-world experiences through un- supervised learning [50,51]. Clinical documentation has also been analyzed with EHR-based clustering, which has revealed a spectrum of symptom profiles and their associations with treatment patterns [49]. Crucially, these clustering models not only help define en- dometriosis subtypes but also aid in distinguishing the condi- tion from other pelvic pain disorders. ML techniques have demonstrated higher predictive value for diagnosing endo- metriosis compared with other pain syndromes by identifying specific symptom constellations, such as the co-occurrence of chronic pelvic pain, subfertility, and dyspareunia [47]. Fur- thermore, studies indicate that women with classic pheno- types typically receive more hormonal or pain-related inter- ventions, whereas women with non-classic or GI-dominant presentations may receive less clinical attention [49,50]. These findings underscore the clinical value of detailed symp- tom profiling in improving differential diagnosis and guiding individualized therapeutic approaches. IMAGING-BASED MACHINE LEARNING MODELS TVUS and MRI are among the most widely used noninvasive

for detecting endometriosis, with each offering dis- tinct advantages and disadvantages. TVUS is a first-line imag- ing modality for endometriosis because of its minimal risk, low cost, and high diagnostic performance, with reported sensitivity ranging from 71% to 98% and specificity from 92% to 100%. TVUS can detect endometriotic lesions in the uterine lining as small as 5 mm, typically appearing as hypoechoic masses, while peritoneal lesions larger than 5 mm are classi- fied as DIE. The ‘tenderness-guided’ technique and the ‘slid- ing sign’ are commonly used to assess lesion depth and the presence of endometriosis in the pouch of Douglas (POD) [52]. MRI is considered a second-line imaging modality following TVUS due to its higher cost and limited availability. However, MRI offers important advantages, including the ability to eval- uate larger pelvic structures such as the bowel, ureters, and extra-pelvic lesions, as well as high sensitivity for fibrin degra- dation products that facilitate the detection of obscured le- sions [52]. Although TVUS and MRI remain essential in endo- metriosis diagnosis, image interpretation is challenging. In this context, ML models such as CNNs and radiomics ap - proaches have shown potential for improving diagnostic ac- curacy. CNNs and radiomics as tools in medical imaging Diagnosing endometriosis using noninvasive methods such as TVUS and MRI remains challenging, with inconsistencies even in expert interpretation. However, the emergence of AI-driven, imaging-based ML models such as CNNs and radio- mics offers the potential for more accurate and timely diag- nosis when trained on large datasets of annotated images that capture subtle patterns of endometriosis often imper- ceptible to the human eye [53]. CNNs are a class of deep learning models capable of processing vast amounts of data; they are specifically designed for tasks such as image pattern Table 1. Summary of symptom clustering in endometriosis using machine learning Key aspect Methods used Data source Main findings/insights Key symptoms Pelvic pain, dysmenorrhea, dyspareunia, infertility, GI issues Patient-reported, EHRs Central to clustering and prediction models [47-51] Unsupervised learning k-means, PAM, mixture models, Bayesian networks Surveys, EHRs, self-tracking Identified classic, GI, and feature-absent phenotypes; robust to data variability [47,49-51] Patient-reported/EHR data Surveys, self-tracking, EHR notes Large cohorts, clinical notes Enabled detection of symptom heterogeneity and less-recognized phenotypes [49,51] Distinguishing disorders Clustering of symptom constellations Bayesian networks, clustering Classic symptom clusters increase endometriosis risk; models help differentiate from other pain disorders [47,49] GI, gastrointestinal; EHR, electronic health record; PAM, partitioning around medoids. 122 http://pfmjournal.org Machine learning in endometriosis detection recognition. In medical contexts, CNNs are frequently applied for detection, classification, and segmentation. Among CNN- based models, U-Net is particularly well suited for medical image segmentation. This ‘U-shaped’ neural network has been extensively used in research on gynecological and onco- logical conditions. Notably, a 2024 study demonstrated that U-Net achieved highly accurate segmentation of endometri- otic lesions, with a Dice coefficient of 0.977 suggesting that CNN-based models trained on sufficiently large datasets can substantially improve the accuracy of noninvasive endome- triosis diagnosis [53]. In addition to CNNs, another imaging-based AI technique is radiomics, a process that extracts and analyzes quantitative features from medical images and transforms them into high-dimensional data such as shape, texture, and intensity. The radiomics workflow typically involves several key steps: image acquisition, preprocessing, segmentation, feature ex- traction, and model building [54,55]. Radiomics can detect subtle patterns and variations in endometriosis that are often overlooked during traditional interpretation, thereby provid- ing a more comprehensive diagnostic approach [54,55]. For example, in a study that employed radiomics to develop a model distinguishing ovarian endometriomas from dermoid cysts, the model achieved an area under the curve (AUC) of 0.981 and an accuracy of approximately 94% [56]. A recent advancement in this field is the Human–AI Collaborative Mul- timodal Multirater Learning (HAICOMM) method, which inte- grates radiomics with collaborative human–AI models to fur- ther enhance diagnostic performance in endometriosis. HAI- COMM demonstrated significant improvements in the classi- fication of POD obliteration on MRI images, outperforming in- terpretations made by either clinicians or AI models alone [57]. Together, CNNs and radiomics exemplify how AI holds substantial promise for improving endometriosis diagnosis through noninvasive methods, thereby reducing diagnostic delays and improving patient outcomes. ML models in detecting endometrioma, DIE, and peritoneal lesions Superficial peritoneal endometriosis In one study, several ML algorithms including random forest and gradient boosting were trained on data from clinical histo- ry and physical examinations in women with chronic pelvic pain but no detectable abnormalities on imaging. The best- performing models achieved a sensitivity of 79.3% and a spec- ificity of 74.2% for predicting superficial peritoneal endome- triosis (SPE), with subsequent surgical confirmation. Key pre- dictive features included oligomenorrhea, bladder pain syn- drome, and irritable bowel syndrome [58]. Additionally, an- other study reported that a random forest model using a saliva microRNA signature (89 miRNAs) achieved 100% sensitivity, specificity, and AUC for detecting the SPE phenotype. This finding suggests a highly accurate and noninvasive diagnostic tool that could potentially replace invasive procedures such as laparoscopy [59]. DIE and endometrioma In a study on DIE, the inclusion of 16 clinical symptoms in an ML algorithm yielded AUC values ranging from 0.91 to 0.95 in training datasets and from 0.66 to 0.92 in test datasets. These models may serve as screening tools for DIE in both gyneco- logical and general practice settings [28]. For endometriomas, differentiation from SPE and DIE relies on imaging modalities such as ultrasound and MRI, as well as distinct cytokine signa- tures in peritoneal fluid. Future ML models incorporating these biomarkers have the potential to further improve diag- nostic accuracy [60]. Human vs. ML diagnostic performance Although ML models have shown promise in the medical im- aging of endometriosis, a key question remains: do ML mod- els outperform clinicians in diagnosis? Comparative evalua- tion of human and ML diagnostic performance is still an emerging area of research. ML approaches aim to improve accuracy, speed, and noninvasiveness, whereas current clini- cal standards continue to rely heavily on clinical evaluation and invasive procedures such as laparoscopy. On one hand, several studies support the diagnostic value of ML models using noninvasive methods. For instance, a 2023 study employing a CNN-based deep learning model, Vi- sual Geometry Group (VGG-16), applied to MRI images report- ed a sensitivity of 84.15% and a specificity of 83.86%, indicat- ing that ML has the potential to reduce false negatives and di- agnostic delays [61]. Other studies applying ML algorithms to detect ultrasonographic signs reported sensitivities and spec- ificities between 70% and 80%. Well-known models such as random forest and extra-trees achieved an AUC of 0.76, sug- gesting moderate to good diagnostic performance [62]. On the other hand, a systematic review comparing TVUS and MRI reported sensitivity and specificity values across different le- sion locations (Table 2). TVUS demonstrated similar sensitivi- ty to MRI but slightly higher specificity. These findings suggest that both methods are reliable, though performance varies depending on lesion location [22]. 123https://doi.org/10.23838/pfm.2025.00177 Ziad Mumtaz Ramadan, et al.

of imaging-based ML Although ML techniques have shown promising results in di- agnostic accuracy, traditional methods remain important in clinical practice. The current standard for diagnosing endo- metriosis involves clinical evaluation, imaging modalities such as MRI and ultrasound, and confirmatory laparoscopy. This reliance often results in diagnostic delays due to the complex and heterogeneous nature of symptoms. A major

of ML is its tendency to overfit, particularly when trained on small datasets. Consequently, researchers empha- size the need for large, well-curated datasets to improve train- ing and to account for the variability of symptoms experi- enced throughout a patient’s lifetime. Addressing this chal- lenge requires a comprehensive strategy that integrates ML with imaging and additional diagnostic approaches, such as biomarker testing, to more effectively detect endometriosis [63]. Despite encouraging findings, the field still requires sys- tematic protocols for image preprocessing and feature ex- traction across imaging modalities, as well as clinical valida- tion through large-scale cohort studies to ensure reliability and generalizability. Collectively, current evidence suggests that human interpretation remains a necessary component of endometriosis diagnosis, despite the limitations faced by clinicians. Only once modern AI technologies are consistently validated and shown to outperform conventional methods can they be considered a standalone alternative. INTEGRATION OF SYMPTOM CLUSTERING WITH IMAGING DATA Although ML models such as CNNs and radiomics have shown substantial potential in detecting endometriotic le- sions, their accuracy is limited when the heterogeneous clini- cal presentations of patients are not considered. To address this limitation, recent studies have focused on integrating symptom clustering with advanced ML imaging data. By joint- ly evaluating patient-reported symptoms and radiological findings, ML models can provide a more holistic and individu- alized approach to diagnosing endometriosis. This section explores ML methodologies that emphasize such integrative strategies in endometriosis research [64]. Multimodal ML re- fers to the use of algorithms that analyze multiple types of data. In the context of endometriosis, this involves combining clinical symptoms with imaging data to improve diagnostic outcomes. The application of multimodal ML is particularly relevant for complex and heterogeneous diseases such as en- dometriosis. Recent advances highlight the importance of combining clinical, symptom, and imaging data using multi- modal ML to achieve more accurate and comprehensive diag- nostic solutions. For instance, in one study that developed an ML model incorporating both imaging data and patient-re- ported symptoms, the results demonstrated high sensitivity and specificity in diagnosing endometriosis [65]. Fusion strategies, benefits, and challenges in multimodal ML A systematic review of fusion techniques found that early fu- sion in which raw data from different modalities are com- bined before training the model is the most commonly used strategy compared with other fusion methods. This approach benefits the model by enabling it to learn cross-modal repre- sentations from the outset [66]. Both conventional ML models (e.g., decision trees and SVMs) and deep learning models (e.g., CNNs and transformers) can be integrated into multi- modal systems. Additionally, Bayesian networks have been applied to cluster pain symptoms and relate them to surgical diagnoses, thereby improving identification of symptom pat- terns in endometriosis [67]. Ensemble and decision tree

are also frequently employed to combine clinical and imaging data, making them useful in the broader context of leveraging multimodal ML for disease diagnosis [47]. The benefits of multimodal ML include improved diagnostic accuracy through the integration of symptom patterns, imag- ing features (e.g., MRI), and clinical evaluations, which helps overcome the limitations of standalone clinician assessments or ML models [16]. Multimodal ML can also facilitate the iden- tification of distinct subtypes of endometriosis, thereby sup- porting earlier and more personalized treatment. Further- more, ML models that integrate imaging with clinical and symptom data can achieve more accurate mapping of lesion location and depth, which is critical for surgical planning [51]. Nevertheless, several challenges remain. Variability in symp- tom reporting and imaging protocols may complicate the data processed by ML models. Many studies are also constrained by small or single-center cohorts, limiting generalizability and statistical significance. Finally, the complexity of advanced ML Table 2. Comparison between sensitivity and specificity of magnetic resonance imaging in various locations Location Sensitivity (%) Specificity (%) Recto-sigmoid 85 95 Recto-vaginal septum 66 97 Uterus 70 93 124 http://pfmjournal.org Machine learning in endometriosis detection models such as ensembles can make them difficult for clini- cians to interpret and trust in decision-making [68]. The integration of multimodal ML has the potential to ad- vance current diagnostic guidelines for endometriosis, but addressing existing challenges and utilizing large datasets will be essential for widespread clinical adoption.

AND CURRENT GAPS IN THE LITERATURE Although advances in technology applied to AI- and ML-based approaches can significantly reduce diagnostic delay and in- crease sensitivity of noninvasive endometriosis diagnosis, downstream clinical impact is complex. Treatment choice continues to remain symptomatic and empirical in practice today with excessive use of symptom suppression by hor- mones, analgesics, or surgery, rather than disease-pheno- type-guided tailored approaches [2,9,21]. As has been em- phasized by recent criticism, endometriosis continues to be managed largely as chronic pain syndrome, symptom-load driven treatment rather than overt diagnostic subtyping [16]. Thus, even if ML models enhance the likelihood of earlier or more accurate detection, their short-term impact on thera- peutic gain may be blunted unless diagnoses progress are in- tegrated into evidence-based treatment pathways. Follow-up analyses need to move beyond initial trend toward diagnosis itself to ascertain if early diagnosis with AI is being translated into better clinical results, such as reduced pain, preservation of fertility, better quality of life, or more personalized opera- tive planning. Without this translation step, better diagnostic equipment is technology advancement with no-holds-barred therapeutic correlations. Despite the promising advances in applying ML to the early diagnosis of endometriosis, several

and gaps remain in the current literature. One of the most significant limitations is the lack of large, labeled datasets that combine clinical symptoms with imaging data, which hinders the development and validation of robust pre- dictive models. Most recent studies have focused exclusively on either symptom clustering or imaging interpretation, rath- er than leveraging the synergistic benefits of multimodal data integration. For example, Wang et al. [57] introduced the HAICOMM model to address this gap by combining T1- and T2-weighted MRI images with domain-annotator labels, but such efforts remain scarce in the literature. Generalizability of existing models is further limited by reliance on single-institu- tion retrospective datasets, which may not adequately repre- sent heterogeneous patient populations. Maleki et al. [69] emphasized that methodological shortcomings such as data leakage and batch effects can lead to overestimation of mod- el performance and reduce applicability across different clini- cal settings. In addition, the lack of standardized evaluation metrics and benchmarking protocols hinders meaningful comparison of model performance across studies [69]. Sala- huddin et al. [70] advocated for uniform evaluation frame- works to ensure both interpretability and reliability of deep learning models in medical imaging. Interpretability itself re- mains a major challenge for complex models such as deep learning networks, with limited efforts directed toward incor- porating explainability tools that could foster clinician trust and usability [70]. Solutions such as InterpretML (Microsoft Research) have been proposed, but their application to endo- metriosis research is only beginning to emerge. Addressing these challenges will be critical to transitioning ML-based tools from experimental applications to routine clinical prac- tice in endometriosis diagnosis [71].

ML represents a transformative approach to advancing early diagnosis of endometriosis by integrating symptom cluster- ing with imaging data. Current models show strong potential; however, their translation into clinical practice is significantly limited by challenges in data availability, interpretability, and standardization. To encourage widespread clinical adoption at an accelerated pace, studies must continue to tackle; creat- ing multicenter, multimodal data consortia with wide and heterogeneous datasets, standardized benchmarking plat- forms and assessment measures to make studies compara- ble, incorporating explainability tools to establish clinician trust and acceptance into routine operating practice, con- ducting prospective external validation studies to determine real-world generalizability and conducting cost-effectiveness analyses to inform policy planning and the use of healthcare resources. By meeting these priorities, ML-based diagnostic tools can move on from proof-of-concept to clinically viable solutions and ultimately reduce diagnostic delays and im- proved outcomes for endometriosis in women. CONFLICTS OF INTEREST No potential conflict of interest relevant to this article was re- ported. 125https://doi.org/10.23838/pfm.2025.00177 Ziad Mumtaz Ramadan, et al. ACKNOWLEDGMENTS The authors would like to thank Dr Noha Farag for her valu- able time and contribution. ORCID Ziad Mumtaz Ramadan https://orcid.org/0009-0001-7436-2958 Saaid Mounzer Mouazen https://orcid.org/0009-0001-1013-9343 Shady Saud Khan https://orcid.org/0009-0005-7684-5426 Sariya Khan https://orcid.org/0009-0003-9809-872X Noha S. Farag https://orcid.org/0009-0001-9549-6832 AUTHOR CONTRIBUTIONS Conception or design: ZMR, SMM, SSK, SK, NSF. Acquisition, analysis, or interpretation of data: ZMR, SMM, SSK, SK, NSF. Drafting the work or revising: ZMR, SMM, SSK, SK, NSF. Final approval of the manuscript: ZMR, SMM, SSK, SK, NSF.

1. As-Sanie S, Mackenzie SC, Morrison L, Schrepf A, Zonder- van KT, Horne AW, et al. Endometriosis: a review. JAMA 2025;334:64-78. 2. Taylor HS, Kotlyar AM, Flores VA. Endometriosis is a chron- ic systemic disease: clinical challenges and novel innova- tions. Lancet 2021;397:839-52. 3. Brauner EV, Koch T, Juul A, Doherty DA, Hart R, Hickey M. Prenatal exposure to maternal stressful life events and earlier age at menarche: the Raine Study. Hum Reprod 2021;36:1959-69. 4. Simoens S, Dunselman G, Dirksen C, Hummelshoj L, Bokor A, Brandes I, et al. The burden of endometriosis: costs and quality of life of women with endometriosis and treated in referral centres. Hum Reprod 2012;27:1292-9. 5. Agarwal SK, Chapron C, Giudice LC, Laufer MR, Leyland N, Missmer SA, et al. Clinical diagnosis of endometriosis: a call to action. Am J Obstet Gynecol 2019;220:354. 6. Johnson NP, Hummelshoj L, Adamson GD, Keckstein J, Taylor HS, Abrao MS, et al. World Endometriosis Society consensus on the classification of endometriosis. Hum Reprod 2017;32:315-24. 7. Guerriero S, Condous G, van den Bosch T, Valentin L, Le- one FP, Van Schoubroeck D, et al. Systematic approach to sonographic evaluation of the pelvis in women with sus- pected endometriosis, including terms, definitions and measurements: a consensus opinion from the Interna- tional Deep Endometriosis Analysis (IDEA) group. Ultra- sound Obstet Gynecol 2016;48:318-32. 8. Nisenblat V, Prentice L, Bossuyt PM, Farquhar C, Hull ML, Johnson N. Combination of the non-invasive tests for the diagnosis of endometriosis. Cochrane Database Syst Rev 2016;7:CD012281. 9. Vercellini P, Somigliana E, Vigano P, Abbiati A, Daguati R, Crosignani PG. Endometriosis: current and future medical therapies. Best Pract Res Clin Obstet Gynaecol 2008;22: 275-306. 10. Suryawanshi S, Vlad AM, Lin HM, Mantia-Smaldone G, Las- key R, Lee M, et al. Plasma microRNAs as novel biomark- ers for endometriosis and endometriosis-associated ovar- ian cancer. Clin Cancer Res 2013;19:1213-24. 11. Leonardi M, Uzuner C, Mestdagh W, Lu C, Guerriero S, Za- jicek M, et al. Diagnostic accuracy of transvaginal ultra- sound for detection of endometriosis using International Deep Endometriosis Analysis (IDEA) approach: prospec- tive international pilot study. Ultrasound Obstet Gynecol 2022;60:404-13. 12. Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, et al. A guide to deep learning in healthcare. Nat Med 2019;25:24-9. 13. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019;25:44- 56. 14. Bini SA. Artificial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care? J Arthroplas- ty 2018;33:2358-61. 15. Coccia ME, Nardone L, Rizzello F. Endometriosis and infer- tility: a long-life approach to preserve reproductive integ- rity. Int J Environ Res Public Health 2022;19:6162. 16. Chapron C, Marcellin L, Borghese B, Santulli P. Rethinking mechanisms, diagnosis and management of endometri- osis. Nat Rev Endocrinol 2019;15:666-82. 17. Stratton P, Berkley KJ. Chronic pelvic pain and endometri- osis: translational evidence of the relationship and impli- cations. Hum Reprod Update 2011;17:327-46. 18. Surrey E, Soliman AM, Trenz H, Blauer-Peterson C, Sluis A. Impact of endometriosis diagnostic delays on healthcare resource utilization and costs. Adv Ther 2020;37:1087-99. 19. Ballard K, Lowton K, Wright J. What’s the delay?: a quali- 126 http://pfmjournal.org Machine learning in endometriosis detection tative study of women’s experiences of reaching a diagno- sis of endometriosis. Fertil Steril 2006;86:1296-301. 20. Soliman AM, Fuldeore M, Snabes MC. Factors associated with time to endometriosis diagnosis in the United States. J Womens Health (Larchmt) 2017;26:788-97. 21. Vercellini P, Vigano P, Somigliana E, Fedele L. Endometrio- sis: pathogenesis and treatment. Nat Rev Endocrinol 2014; 10:261-75. 22. Guerriero S, Saba L, Pascual MA, Ajossa S, Rodriguez I, Mais V, et al. Transvaginal ultrasound vs magnetic reso- nance imaging for diagnosing deep infiltrating endome- triosis: systematic review and meta-analysis. Ultrasound Obstet Gynecol 2018;51:586-95. 23. Bazot M, Bharwani N, Huchon C, Kinkel K, Cunha TM, Guerra A, et al. European Society of Urogenital Radiology (ESUR) guidelines: MR imaging of pelvic endometriosis. Eur Radiol 2017;27:2765-75. 24. Nisenblat V, Bossuyt PM, Farquhar C, Johnson N, Hull ML. Imaging modalities for the non-invasive diagnosis of endo- metriosis. Cochrane Database Syst Rev 2016;2:CD009591. 25. Maksym RB, Hoffmann-Mlodzianowska M, Skibinska M, Rabijewski M, Mackiewicz A, Kieda C. Immunology and immunotherapy of endometriosis. J Clin Med 2021;10: 5879. 26. Nnoaham KE, Hummelshoj L, Kennedy SH, Jenkinson C, Zondervan KT. Developing symptom-based predictive models of endometriosis as a clinical screening tool: re- sults from a multicenter study. Fertil Steril 2012;98:692- 701. 27. Cetera GE, Tozzi AE, Chiappa V, Castiglioni I, Merli CE, Ver- cellini P. Artificial intelligence in the management of wom- en with endometriosis and adenomyosis: can machines ever be worse than humans? J Clin Med 2024;13:2950. 28. Bendifallah S, Puchar A, Suisse S, Delbos L, Poilblanc M, Descamps P, et al. Machine learning algorithms as new screening approach for patients with endometriosis. Sci Rep 2022;12:639. 29. Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism 2017;69S:S36-40. 30. Waring J, Lindvall C, Umeton R. Automated machine learn- ing: review of the state-of-the-art and opportunities for healthcare. Artif Intell Med 2020;104:101822. 31. Sidey-Gibbons JA, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Meth- odol 2019;19:64. 32. Garg A, Mago V. Role of machine learning in medical re- search: a survey. Comput Sci Rev 2021;40:100370. 33. Dallora AL, Eivazzadeh S, Mendes E, Berglund J, Ander- berg P. Machine learning and microsimulation techniques on the prognosis of dementia: a systematic literature re- view. PLoS One 2017;12:e0179804. 34. Shameer K, Johnson KW, Glicksberg BS, Dudley JT, Sen- gupta PP. Machine learning in cardiovascular medicine: are we there yet? Heart 2018;104:1156-64. 35. Gopinath N, Suresh Anand M, Ishwarya MV. Machine learn- ing for medical images. In: Ganesh Babu TR, Saravanaku- mar U, Pattanaik B, editors. Computational imaging and analytics in biomedical engineering. Apple Academic Press; 2024. p. 95-104. 36. Kim H. Artificial intelligence for 6G. Springer International Publishing; 2022. 37. Karako K, Tang W. Applications of and issues with machine learning in medicine: bridging the gap with explainable AI. Biosci Trends 2025;18:497-504. 38. Kononenko I. Machine learning for medical diagnosis: his- tory, state of the art and perspective. Artif Intell Med 2001; 23:89-109. 39. Fiste O, Liontos M, Zagouri F, Stamatakos G, Dimopoulos MA. Machine learning applications in gynecological can- cer: a critical review. Crit Rev Oncol Hematol 2022;179: 103808. 40. Khan I, Khare BK. Exploring the potential of machine learn- ing in gynecological care: a review. Arch Gynecol Obstet 2024;309:2347-65. 41. Idlahcen F, Idri A, Goceri E. Exploring data mining and ma- chine learning in gynecologic oncology. Artif Intell Rev 2024;57:20. 42. Taddese AA, Tilahun BC, Awoke T, Atnafu A, Mamuye A, Mengiste SA. Deep-learning models for image-based gy- necological cancer diagnosis: a systematic review and meta-analysis. Front Oncol 2024;13:1216326. 43. Shrestha P, Poudyal B, Yadollahi S, Wright DE, Gregory AV, Warner JD, et al. A systematic review on the use of artifi- cial intelligence in gynecologic imaging: background, state of the art, and future directions. Gynecol Oncol 2022;166: 596-605. 44. Mennickent D, Rodriguez A, Opazo MC, Riedel CA, Castro E, Eriz-Salinas A, et al. Machine learning applied in maternal and fetal health: a narrative review focused on pregnancy diseases and complications. Front Endocrinol (Lausanne) 2023;14:1130139. 45. Cellini P, Pigoni A, Delvecchio G, Moltrasio C, Brambilla P. Machine learning in the prediction of postpartum depres- sion: a review. J Affect Disord. 2022;309:350-7. 127https://doi.org/10.23838/pfm.2025.00177 Ziad Mumtaz Ramadan, et al. 46. Islam A, Parvin R, Sultana T. Gynecological disease predic- tion by machine learning. World J Adv Res Rev 2024;23: 2107-12. 47. Kiser AC, Schliep KC, Hernandez EJ, Peterson CM, Yandell M, Eilbeck K. An artificial intelligence approach for investi- gating multifactorial pain-related features of endometrio- sis. PLoS One 2024;19:e0297998. 48. Goldstein A, Cohen S. Self-report symptom-based endo- metriosis prediction using machine learning. Sci Rep 2023;13:5499. 49. Cohen RM, Leventhal E, Nukavarapu N, Lazarov V, Hanif S, Elovitz MA, et al. Phenotyping adolescent endometriosis: characterizing symptom heterogeneity through note-and patient-level clustering. medRxiv [Preprint] 2025 Feb 12. https://doi.org/10.1101/2025.02.10.25321215 50. Urteaga I, McKillop M, Elhadad N. Learning endometriosis phenotypes from patient-generated data. NP J Digit Med 2020;3:88. 51. Vallee A, Arutkin M, Ceccaldi PF, Ayoubi JM. Quality of life identification by unsupervised cluster analysis: a new ap- proach to modelling the burden of endometriosis. PLoS One 2025;20:e0317178. 52. Harshe G. Magnetic resonance imaging in evaluation of endometriosis: case-based pictorial essay. Med J Armed Forces India 2024;80:25-32. 53. Ithani QT, Mokri SS, Zulkarnain N, Ahmad MFB. Deep learn- ing model for endometrium segmentation in transvaginal ultrasound (TVUS) images. Adv Eng Res 2024;245:66-80. 54. Zhang W, Guo Y, Jin Q. Radiomics and its feature selec- tion: a review. Symmetry 2023;15:1834. 55. Ibrahim A, Primakov S, Beuque M, Woodruff HC, Halilaj I, Wu G, et al. Radiomics for precision medicine: current challenges, future prospects, and the proposal of a new framework. Methods 2021;188:20-9. 56. Liu L, Cai W, Zhou C, Tian H, Wu B, Zhang J, et al. Ultrasound radiomics-based artificial intelligence model to assist in the differential diagnosis of ovarian endometrioma and ovarian dermoid cyst. Front Med (Lausanne) 2024;11: 1362588. 57. Wang H, Butler D, Zhang Y, Avery J, Knox S, Ma C, et al. Hu- man-AI collaborative multi-modal multi-rater learning for endometriosis diagnosis. Phys Med Biol 2024;70:015008. 58. Santos LL, de Azevedo MC, Shimamura LK, Nogueira AA, Candido-dos-Reis F J, Schor E, et al. Machine learning as a clinical decision support tool for diagnosing superficial peritoneal endometriosis in women with dysmenorrhea and acyclic pelvic pain. Med Res Arch 2024;12:1-24. 59. Bendifallah S, Dabi Y, Suisse S, Ilic J, Delbos L, Poilblanc M, et al. Saliva-based microRNA diagnostic signature for the superficial peritoneal endometriosis phenotype. Eur J Obstet Gynecol Reprod Biol 2024;297:187-96. 60. Zhou J, Chern BS, Barton-Smith P, Phoon JW, Tan TY, Viar- dot-Foucault V, et al. Peritoneal fluid cytokines reveal new insights of endometriosis subphenotypes. Int J Mol Sci 2020;21:3515. 61. Ribeiro WR, da Silva IFS, Diniz JOB, Silva AC, de Paiva AC, Brandao Salomao ACC, et al. Deep learning-based com- putational approach for the diagnosis of deep endometri- osis using magnetic resonance imaging. In: Proceedings of the 23rd Brazilian Symposium on Computing Applied to Health (SBCAS 2023); 2023 Jun 27–30; São Paulo, Brazil. 62. Nouri B, Hashemi SH, J Ghadimi D, Roshandel S, Akhlagh- doust M. Machine learning-based detection of endometri- osis: a retrospective study in a population of Iranian fe- male patients. Int J Fertil Steril 2024;18:362-6. 63. Shrestha P, Shrestha B, Sherestha J, Chen J. Current sta- tus and future potential of machine learning in diagnostic imaging of endometriosis: a literature review. JNMA J Ne- pal Med Assoc 2025;63:205-11. 64. Sivajohan B, Elgendi M, Menon C, Allaire C, Yong P, Bedai- wy MA. Clinical use of artificial intelligence in endometrio- sis: a scoping review. NP J Digit Med 2022;5:109. 65. Enamorado-Diaz E, Morales-Trujillo L, Garcia-Garcia JA, Marcos AT, Navarro-Pando J, Escalona-Cuaresma MJ. A novel machine learning-based proposal for early predic- tion of endometriosis disease. Expert Syst Appl 2025;271: 126621. 66. Mohsen F, Ali H, El Hajj N, Shah Z. Artificial intelligence- based methods for fusion of electronic health records and imaging data. Sci Rep 2022;12:17981. 67. Ben-Miled Z, Shebesh JA, Su J, Dexter PR, Grout RW, Boustani MA. Multi-modal fusion of routine care electron- ic health records (EHR): a scoping review. Information (Basel) 2025;16:16010054. 68. Quesada J, Harma K, Reid S, Rao T, Lo G, Yang N, et al. En- dometriosis: a multimodal imaging review. Eur J Radiol 2023;158:110610. 69. Maleki F, Ovens K, Gupta R, Reinhold C, Spatz A, Forghani R. Generalizability of machine learning models: quantita- tive evaluation of three methodological pitfalls. Radiol Ar- tif Intell 2022;5:e220028. 70. Salahuddin Z, Woodruff HC, Chatterjee A, Lambin P. Trans- parency of deep neural networks for medical image anal- ysis: a review of interpretability methods. Comput Biol 128 http://pfmjournal.org Machine learning in endometriosis detection Med 2022;140:105111. 71. Jenkins S, Nori H, Koch P, Caruana R. interpret 0.1.34: fit interpretable machine learning models [Internet]. The Comprehensive R Archive Network; 2019 [cited 2025 Sep 1]. Available from: https://cran.r-project.org/src/contrib/ Archive/interpret