Comparing various methods of artificial intelligence in the diagnosisof polycystic ovarian syndrome

Background and aims: polycystic ovarian syndrome (PCOS) is a common endocrine disorderin reproductive-aged women with the risk of complex long-term complications. However, earlydetection of PCOS is one of the most critical concerns in the field of women’s healthcare. Owingto the complexity and various heterogenic profiles of the syndrome, identification, and differentialdiagnosis remain challenging despite the widely accepted criteria and tests. With the rapiddevelopment of artificial intelligence (AI), using machine learning and deep learning to assistwith PCOS detection has attracted much more attention. In the current study, different AI methodsapplied to PCOS diagnosis are reviewed to highlight the best prediction model(s).Method: The systematic review was performed on all published studies that have investigated theAI technology on PCOS detection based on the PRISMA statement, PubMed and Ovid’s databaseswere searched up to November ۲۰۲۲ using the terms ‘polycystic ovarian disease/syndrome’,‘PCOS’, ‘Stein Leventhal syndrome’, ‘Rotterdam’, ‘ESHRE/ASRM’, ‘criteria’, and ‘AI’, ‘deeplearning’, and ‘machine learning’ algorithms.Results: Based on the research design, algorithm type(s), number, and types of clinical parameters,the detection accuracy of each method varied greatly, ranging from ۷۹-۹۸%. Comparing thevarious applied AI methods showed that hybrid approaches are very much effective in the detectionof PCOS, especially SVM/KNN/ Logistic Regression hybrid model with a ۹۸% accuracyscore. Moreover, XG Boost and CatBoost have been also proposed to function as strong modelsfor the detection of PCOS with an accuracy score of ۹۶% and ۹۵%, respectively.Conclusion: AI deep learning technology provides a powerful tool for detecting PCOS at an earlystage and then early treatment of the patient. Ultimately, in order to achieve the highest detectionaccuracy of PCOS, hybrid approach-based algorithms on the most important genetic, epigenetics,transcriptome, clinical, metabolic, and immunological characteristics should be assessed.