{"paper_id":"dcf6ede3-51dc-4599-8a32-259f82fe2cd3","body_text":"Review Article\nISSN: 2574 -1241              DOI: 10.26717/BJSTR.2026.66.010350\nThe Role of Artificial Intelligence in Revolutionizing \nEndometriosis Diagnosis and Clinical Pathology: \nA Commentary Review of New Biomarkers (2020–2026)\nCopyright@ : Raouf Roshdy | Biomed J Sci & Tech Res |  BJSTR.MS.ID.010350. 58790\nRaouf Roshdy1* and Reham A Rashed2\n1Egypt Health Foundation , Egypt\n2Clinical Pathology Department, National Cancer Institute, Cairo University, Egypt\n*Corresponding author: Raouf Roshdy, Egypt Health Foundation , Egypt\nABSTRACT\nThe diagnosis of endometriosis has historically been hampered by an average diagnostic delay of seven to ten \nyears, owing largely to the field’s reliance on invasive surgical laparoscopy as the “gold standard.” The integra -\ntion of artificial intelligence (AI) and machine learning (ML) into clinical pathology, however, is now catalyzing a \nparadigm shift toward non-invasive, precision diagnostics. This report synthesizes research published between \n2020 and 2026, highlighting the emergence of high-accuracy diagnostic tools built on microRNA (miRNA) sig -\nnatures, plasma protein panels, and genomic markers. Key findings include the validation of blood- and sali -\nva-based miRNA signatures achieving area under the curve (AUC) values exceeding 0.95, rivaling the accuracy of \nsurgical diagnosis. AI methodologies including Random Forest, eXtreme Gradient Boosting (XGBoost), and sup-\nport vector machines (SVM) have identified novel biomarkers such as PNMA2, LRP1, and WISP2/CCN5. These \nadvances promise not only to shorten diagnostic latency but also to enable molecular stratification, supporting \nmore personalized treatment strategies. Challenges remain in clinical standardization and in validating certain \nprotein-based models, yet the accumulated evidence suggests that AI-driven clinical pathology will soon aug -\nment and in some cases replace traditional surgical diagnosis for many patients.\nAbbreviations: AI: Artificial Intelligence; ML: Machine Learning; miRNA: microRNA; AUC: Area Under the \nCurve; SVM: Support Vector Machines; ANN: Artificial Neural Networks; CNN: Convolutional Neural Networks; \nWGCNA: Weighted Gene Co-Expression Network Analysis\nARTICLE INFO\nReceived: \n   August 07, 2026\nPublished: \n   August 19, 2026\nCitation: Raouf Roshdy and Reham \nA Rashed. The Role of Artificial Intel -\nligence in Revolutionizing Endome -\ntriosis Diagnosis and Clinical Pathol -\nogy: A Commentary Review of New \nBiomarkers (2020–2026) . Biomed \nJ Sci & Tech Res 66(3)-2026. BJSTR. \nMS.ID.010350.\nIntroduction\nEndometriosis is a complex, multifactorial inflammatory disease \ncharacterized by the presence of endometrial-like tissue outside the \nuterine cavity. It affects millions of individuals worldwide and is a \nleading cause of chronic pelvic pain and infertility [1]. Despite this \nprevalence, the current diagnostic framework remains significantly \nflawed: laparoscopy an invasive surgical procedure carrying real risks \nand high costs remains the only definitive diagnostic method [2,3]. \nThis reliance on surgery contributes to substantial diagnostic delays \nand prolonged patient suffering. Clinical pathology is now undergo -\ning a transformation driven by artificial intelligence. By leveraging \nlarge multi-omics datasets (transcriptomics, proteomics, methylom -\nics) alongside high-resolution imaging, AI-based tools are uncovering \npatterns “crimson clues,” in the words of one recent review within \nperipheral blood, saliva, and tissue that are invisible to the human eye \n[4,5]. These technologies aim to shift endometriosis diagnosis away \nfrom a purely symptom-based or surgical approach and toward one \ngrounded in precision medicine and molecular stratification [3,5]. \nThis report examines the AI methodologies, novel biomarkers, and \nclinical impacts described in the recent literature, with the goal of \nproviding a comprehensive overview of the current state of the art.\nAI Methodologies and Approaches in the Sources\nThe literature reviewed here reveals a diverse array of AI and \nmachine learning methodologies, each tailored to a particular type \nof biological data. Broadly, these approaches fall into two categories: \nsupervised learning for classification, and unsupervised learning for \nsubtype identification.\n\nCopyright@ :  Raouf Roshdy | Biomed J Sci & Tech Res | BJSTR.MS.ID.010350.\nVolume 66- Issue 3\nDOI: 10.26717/BJSTR.2026.66.010350\n58791\nSupervised Learning and Ensemble Methods\nThe most common application of AI in the sources reviewed is \nsupervised learning used to differentiate patients with endometriosis \nfrom healthy controls.\n•\t Ensemble classifiers : “GenomeForest” is an advanced en -\nsemble machine learning classifier that uses chromosomal \npartitioning to analyze transcriptomic and methylomic data, \nachieving F1 scores as high as 0.968 [6].\n•\t Tree-based models : Random Forest, XGBoost, and Ada -\nBoost are frequently favored for their robustness in handling \nhigh-dimensional omics data. Random Forest algorithms, for \nexample, have been used to validate salivary miRNA signa -\ntures with 96.2% sensitivity [7] and to identify driver bio -\nmarkers from RNA-seq data [8].\n•\t Neural networks: Artificial neural networks (ANN) and con -\nvolutional neural networks (CNN) are increasingly applied to \ncomplex diagnostic models and imaging data [8,9].\nIntegrated Discovery Pipelines\nMany studies rely on hybrid pipelines that combine bioinformat -\nics with machine learning for biomarker screening.\n•\t WGCNA and ML: Weighted gene co-expression network anal-\nysis (WGCNA) is often paired with machine learning to iden -\ntify lipid metabolism-related biomarkers and molecular sub -\ntypes [10].\n•\t Multi-omics integration: Markov et al. describe the integra -\ntion of ML and deep learning with imaging, molecular bio -\nmarkers, and histopathology to build multi-omics diagnostic \npathways [11].\n•\t Stacked models: Stacked machine learning models have been \ndeveloped to distinguish endometriosis from adenomyosis us-\ning peripheral blood and coagulation data [12].\nFindings and Comparative Analysis\nThe synthesis of recent research points to several high-potential \nbiomarker categories, with miRNA signatures currently showing the \ngreatest diagnostic accuracy.\nmiRNA Signatures and Non-Invasive Diagnostics\nMicroRNAs have emerged as the most promising candidates for \nnon-invasive testing.\n•\t Blood-based signatures: Bendifallah et al. [13] developed an \n86-miRNA signature using algorithms such as Random Forest \nand XGBoost, achieving an AUC of 0.984 [13] a robust, repro -\nducible diagnostic that could potentially replace laparoscopy \n[13].\n•\t Saliva-based testing: A 109-miRNA saliva signature has been \nvalidated with 96.2% sensitivity and 95.1% specificity [7]. \nThis approach offers a genuinely non-invasive alternative that \ncould be administered in a primary care setting, meaningfully \nshortening the path to diagnosis [14].\n•\t Early-stage detection : Specific miRNA panels (e.g., miR-\n199a, miR-122) combined with CA-125 have demonstrated an \nAUC of 0.939 for detecting early-stage disease [15].\nProteomic and Immunologic Biomarkers\nWhile miRNAs dominate the literature, protein- and im -\nmune-based markers offer critical insight into disease pathophysiol -\nogy.\n•\t Plasma protein panels: Schoeman et al. [16] identified a pan-\nel of 10 plasma protein biomarkers. Using a Random Forest \nmodel, they achieved an AUC of 0.997 for severe endometrio-\nsis, though accuracy for milder cases was considerably lower \n(Model 2 AUC = 0.729) [16].\n•\t Immune screening : Non-invasive immune screening tests \ncombined with ML (logistic regression) have identified mark-\ners predictive of disease, including IL-10 in Tc cells, reduced \nNK cytotoxic activity, and CXCL1 [2].\n•\t Challenges in reproducibility: Notably, Fishman [17] found \nthat some cytokine-based ML models performed no better \nthan chance, underscoring the need for rigorous standardiza-\ntion in protein-based discovery [17].\nGenomic and Transcriptomic Drivers\nMachine learning has also proven instrumental in identifying the \nmolecular drivers underlying the disease.\n•\t Hub genes : Integrated identification methods have pointed \nto five key diagnostic genes FOS, EPHX1, DLGAP5, PCSK5,  and \nADAT1 [18]. A separate study identified LRP1 as a potential \ndiagnostic and immunomodulatory target, showing a strong \ncorrelation with M2 macrophage infiltration [19].\n•\t Novel biomarkers : AI-driven screening has highlighted \nPNMA2 (linked to malignant growth and anti-apoptosis) [20], \nWISP2/CCN5 (a potential early-screening marker) [21], and \nSPP1 (a modulator of macrophage polarization) [22].\n•\t Cuproptosis-related genes : ML algorithms have identified \ngenes such as GLS and PDHA1 as being associated with endo -\nmetriosis progression and immune infiltration [23].\nAI in Imaging and Digital Pathology\nAI’s impact extends to the visual characterization of the disease \nas well.\n\nCopyright@ :  Raouf Roshdy | Biomed J Sci & Tech Res |  BJSTR.MS.ID.010350.\n58792\nVolume 66- Issue 3\nDOI: 10.26717/BJSTR.2026.66.010350\n•\t Automated recognition : Proof-of-concept studies have \ndemonstrated the automatic visual recognition of endometri-\nosis lesions during laparoscopy using AI, a step that could help \nstandardize surgical assessment [24].\n•\t Tissue segmentation: AI-based tissue segmentation and cell \nidentification in histological sections for instance, identifying \nCD138+ cells are improving the precision of pathological diag-\nnosis in infertility-associated conditions [25,26].\n•\t Imaging integration: Deep learning models are increasing -\nly being integrated with ultrasound and MRI data to improve \ndetection of deep infiltrating endometriosis and ovarian endo-\nmetriomas [27,28].\nImpact on Clinical Pathology and Diagnostic \nWorkflows\nThe integration of AI into clinical pathology is reshaping the diag-\nnostic landscape in several ways:\n•\t Reducing diagnostic latency : By enabling high-accuracy \nblood or saliva tests, AI tools can identify candidate biomark -\ner genes and signatures that shorten the years-long wait many \npatients face for a diagnosis [6,21].\n•\t Molecular stratification: AI enables identification of disease \nsubtypes, moving the field away from a one-size-fits-all ap -\nproach and toward models informed by molecular, epidemio-\nlogic, and cell-specific data [5].\n•\t Precision medicine : The development of diagnostic mod -\nels that predict treatment response for example, PNMA2 as a \ntherapeutic target is helping guide more personalized thera -\npeutic strategies [3,20].\n•\t Augmenting the gold standard: While some sources suggest \nAI could ultimately replace laparoscopy [13,15], others view \nit instead as a powerful complement that improves the effi -\nciency and accuracy of existing histopathological and surgical \nworkflows [18,29,30].\nLimitations of the Evidence\nDespite rapid progress in the field, several limitations must be \nacknowledged:\n•\t Validation gaps : Many of the biomarkers identified so far \nremain in the discovery or interim-validation stage [4,14]. \nLarge-scale, multi-center prospective trials will be necessary \nto confirm their clinical utility.\n•\t Data heterogeneity: AI model performance varies consider -\nably depending on the dataset and disease stage, as reflected \nin the gap between severe- and mild-endometriosis detection \nin protein-based models [16].\n•\t Interpretability and fallibility : The “black box” nature of \nsome ML models remains a challenge in diagnostic patholo -\ngy, where clinical decisions demand transparent reasoning \n[31,32].\n•\t Standardization: Standardized protocols for biomarker col -\nlection and processing are urgently needed to ensure that \nAI-driven findings remain reproducible across different clin -\nical settings [4].\nResearch Gaps and Future Directions\nFuture research in this area should focus on:\n•\t Longitudinal studies: Understanding how AI-identified bio -\nmarkers change over the course of the disease and in response \nto treatment.\n•\t Integration with clinical data : Combining molecular signa -\ntures with symptom clustering and imaging data to build a \nholistic diagnostic “score” [33].\n•\t Cost-effectiveness analysis: Evaluating the economic impact \nof replacing laparoscopy with AI-driven non-invasive tests.\n•\t Pathologist training : Developing educational frameworks \nthat help pathologists integrate AI assistance into daily prac -\ntice [34].\nConclusion\nArtificial intelligence is no longer a peripheral technology in the \nstudy of endometriosis it is the engine driving the next generation of \nclinical pathology. Through the identification of highly accurate miR-\nNA signatures, novel genomic drivers such as LRP1 and PNMA2, and \nthe automation of digital pathology, AI is supplying the tools needed \nto overcome the field’s historical challenges of delayed diagnosis and \ninvasive surgery. Rigorous clinical validation remains the final hur -\ndle, but the evidence accumulated between 2020 and 2026 strongly \nsuggests that AI-powered non-invasive diagnostics will soon become \na cornerstone of endometriosis management, ushering in an era of \nprecision medicine for millions of patients..\nReferences\n1. A Ashish, K Kusum, S Rai, R Singh (2020) Endometriosis a brief review: \nevaluation of crucial risk factors and current treatment regimes. Interna -\ntional Journal of Advances in Medicine 7(12): 1896-1905.\n2. N Attia, M Franzoi, M P McHale, A R Gargiulo, M S Abrão, et al. (2023) Iden-\ntification of immunologic predictive biomarkers of endometriosis using a \nnon-invasive immune screening test in association with machine learning \nmodeling. Fertility and Sterility 120(4): e226-e227.\n3. L N Onah, M N Ezenwaeze, C N Nwankwo (2026) Endometriosis: From \ndelayed diagnosis to precision medicine. International Journal of Science \nand Research Archive 18(1): 306-314.\n4. T Gibbons, N Rahmioglu, K T Zondervan, C M Becker (2024) Crimson \n\nCopyright@ :  Raouf Roshdy | Biomed J Sci & Tech Res | BJSTR.MS.ID.010350.\nVolume 66- Issue 3\nDOI: 10.26717/BJSTR.2026.66.010350\n58793\nclues: advancing endometriosis detection and management with novel \nblood biomarkers. Fertility and Sterility 121(2): 145-163.\n5. C R Nezhat, Tomiko T Oskotsky, Joshua F R, Susan J F, Angie T , et al. (2025) \nReal World Perspectives on Endometriosis Disease Phenotyping Through \nSurgery, Omics, Health Data, and Artificial Intelligence. npj Women’s \nHealth 3(1): 8.\n6. S Akter, D Xu, S C Nagel, J J Bromfield, K E Pelch, et al. (2020) GenomeFor-\nest: An Ensemble Machine Learning Classifier for Endometriosis, p. 33-42.\n7. S Bendifallah, Y Dabi, S Suisse, L Delbos, A Spiers, et al. (2023) Validation \nof a Salivary miRNA Signature of Endometriosis — Interim Data. NEJM \nEvidence 2(7).\n8. Zhen G, T Chu, J She, S Chen, P Wang, et al. (2026) Exploration of driver \nbiomarkers and immune microenvironment in patients with endometri -\nosis: Evidence from RNA-seq and machine learning. Journal of Reproduc -\ntive Immunology 175: 104872.\n9. H Malvezzi, E B Marengo, S Podgaec, C de A Piccinato (2020) Endometri -\nosis: current challenges in modeling a multifactorial disease of unknown \netiology. Journal of Translational Medicine 18(1): 1-21.\n10. Y Guo, Y Hou, J Wu, N Lou, D Yang, et al. (2025) Identification and Subtype \nAnalysis of Lipid Metabolism-Related Diagnostic Biomarkers for Endo -\nmetriosis Based on WGCNA and Machine Learning. American Journal of \nReproductive Immunology, p. 94.\n11. D Markov, J Gurung, U Khalid, K Bechev, V Aleksiev, et al. (2026) Emerging \nPathways to Non-Invasive Diagnosis in Endometriosis: Integrating Ma -\nchine Learning, Deep Learning and Multi-Omics Biomarkers. Diagnostics \n16(12): 1823.\n12. W Wang, W Zeng, S Yang (2024) A stacked machine learning-based clas -\nsification model for endometriosis and adenomyosis: a retrospective co -\nhort study utilizing peripheral blood and coagulation markers. Frontiers \nin Digital Health 6: 1463419.\n13. S Bendifallah, Y Dabi, S Suisse, L Jornea, D Bouteiller, et al. (2022) MicroR-\nNome analysis generates a blood-based signature for endometriosis. Sci -\nentific Reports 12(1).\n14. Gaia-Oltean AI, Boitor D, Laura-Ancuta P , Geanina G, Teodora T , et al. \n(2025) Non-Invasive Methods for Early Diagnosis of Endometriosis-A \nComprehensive Narrative Literature Review. Healthcare (Basel, Switzer -\nland) 13(24): 3276.\n15. M K Gałuszka, K K Figurny, P P Jagodziński, A Pławski (2025) Potential \nbiomarkers for early detection of endometriosis: current state of art (what \nwe know so far). Journal of Applied Genetics.\n16. E M Schoeman, S Bringans, K Peters, T Casey, C Andronis, et al. (2024) \nIdentification of plasma protein biomarkers for endometriosis and the de-\nvelopment of statistical models for disease diagnosis. Human Reproduc -\ntion 40(2): 270-279.\n17. D Fishman (2022) Developing a data analysis pipeline for automated pro-\ntein profiling in immunology.\n18. H Zhang, H Zhang, H Yang, A N Shuid, D Sandai, et al. (2023) Machine \nlearning-based integrated identification of predictive combined diagnos -\ntic biomarkers for endometriosis. Frontiers in Genetics 14: 1290036.\n19. Xie Liu (2026) LRP1 as a potential diagnostic and immunomodulatory tar-\nget in endometriosis: evidence from multi-omics and single-cell analyses. \nFrontiers in Immunology p: 17.\n20. Mengjun Zhang, Zidi Zhang, Jialin Wang, Haodi Yue, Xueling Lou, et al. \n(2025) Unveiling the Role of PNMA2 in Endometriosis: From Proliferation \nand Apoptosis to Immunomodulation. Journal of Cellular and Molecular \nMedicine 29(9): e70576.\n21. Sheng Dou, Shaohua Ling, Weihua Nong, Bixiao Wei, Yuehua Huang, et \nal. (2025) WISP2/CCN5 revealed as a potential diagnostic biomarker for \nendometriosis based on machine learning and single-cell transcriptomic \nanalysis. Functional & Integrative Genomics 25(1): 131.\n22. Sijia Zhang, Yi Yang, Chen Zhang, Xiaoyao Che, Fawei Li, et al. (2025) SPP1 \nas a key modulator of M2 macrophage polarization promotes endometri -\nosis progression via activation of the FAK/PI3K/AKT pathway: A bioinfor-\nmatics and experimental study. International Immunopharmacology 166: \n115563.\n23. Y Lu, J Wang, G Sun, Z Fang, Z Xing, et al. (2023) Machine learning algo -\nrithms for a novel cuproptosis-related gene signature of diagnostic and \nimmune infiltration in endometriosis. Scientific Reports 13(1): 21603.\n24. A Netter, S Noorzadeh, F Duchateau, H Abrao, J Desternes, et al. (2025) Ini-\ntial results in the automatic visual recognition of endometriosis lesions by \nartificial intelligence during laparoscopy: a proof-of-concept study. Jour -\nnal of Minimally Invasive Gynecology 32(12): 1118-1125.\n25. S E Korman, G Vissers, M A J Gorris, K Verrijp (2024) Artificial intel -\nligence-based tissue segmentation and cell identification in multi -\nplex-stained histological endometriosis sections. Human Reproduction \n40(3): 450-460.\n26. S Lee, R K Arffman, E K Komsi, O Lindgren, J A Kemppainen, et al. (2024) \nAI-algorithm training and validation for identification of endometrial \nCD138+ cells in infertility-associated conditions; polycystic ovary syn -\ndrome (PCOS) and recurrent implantation failure (RIF). Journal of Pathol-\nogy Informatics 15: 100380-100380.\n27. N Jiang, H Xie, J Lin, Y Wang, Y Yin, et al. (2022) Diagnosis and Nursing \nIntervention of Gynecological Ovarian Endometriosis with Magnetic Res -\nonance Imaging under Artificial Intelligence Algorithm. Computational \nIntelligence and Neuroscience, p. 1-10.\n28. J Avery, A Deslandes, S M Freger, M Leonardi, G Lo, et al. (2023) Non-inva-\nsive Diagnostic Imaging for Endometriosis Part 1: A Systematic review of \nrecent developments in Ultrasound. Combination Imaging and Artificial \nIntelligence. Fertility and Sterility 121(2): 164-188.\n29. Y Yu, W Hing W, Y Hu, Y Shen, X Xu, et al. (2025) Translation of miRNA \nblood-based discovery to molecular testing for clinical diagnosis of endo-\nmetriosis. medRxiv.\n30. M J Iqbal, Z Javed, H Sadia, I A Qureshi, A Irshad, et al. (2021) Clinical appli-\ncations of artificial intelligence and machine learning in cancer diagnosis: \nlooking into the future. Cancer Cell International 21(1): 270-270.\n31. P Mishra, A Panda, M Mahapatra, P Dakshinakabat, A Mohanty, et al. \n(2024) Application and fallibility of Artificial Intelligence and machine \nlearning in Diagnostic Pathology. Bangladesh Journal of Medical Science \n23(10): S32-S37.\n32. J H Harrison, John R Gilbertson, M G Hanna, N H Olson, J N Seheult, et al. \n(2021) Introduction to Artificial Intelligence and Machine Learning for \nPathology. Archives of Pathology & Laboratory Medicine 145(10): 1228-\n1254.\n33. S M Mouazen, S Khan, N S Farag (2025) Machine learning in the early de -\ntection of endometriosis: a literature review on symptom clustering and \nimaging integration. Precision and Future Medicine 9(3): 117-128.\n34. Z Ahmad, S Rahim, M Zubair, J A Ghafar (2021) Artificial intelligence (AI) \nin medicine, current applications and future role with special emphasis \non its potential and promise in pathology: present and future impact, ob -\nstacles including costs and acceptance among pathologists, practical and \nphilosophical considerations. A comprehensive review. Diagnostic Pathol-\nogy 16(1): 24-24.\n\nCopyright@ :  Raouf Roshdy | Biomed J Sci & Tech Res |  BJSTR.MS.ID.010350.\n58794\nVolume 66- Issue 3\nDOI: 10.26717/BJSTR.2026.66.010350\nSubmission Link: https://biomedres.us/submit-manuscript.php\nAssets of Publishing with us\n• Global archiving of articles\n• Immediate, unrestricted online access\n• Rigorous Peer Review Process\n• Authors Retain Copyrights\n• Unique DOI for all articles\nhttps://biomedres.us/\nThis work is licensed under Creative\nCommons Attribution 4.0 License\nISSN: 2574-1241\nDOI: 10.26717/BJSTR.2026.66.010350\nRaouf Roshdy . Biomed J Sci & Tech Res","source_license":"CC0","license_restricted":false}