A Hybrid AI Framework for Early Detection of PCOS using Questionnaire Data and Ultrasound Imaging

Abstract

Polycystic Ovarian Syndrome (PCOS) remains a prevalent endocrine challenge for women in their reproductive years. Because its clinical manifestations are so diverse, achieving an accurate diagnosis is frequently difficult, necessitating a comprehensive and detailed medical evaluation. A late diagnosis of PCOS can result in severe long-term health issues, such as infertility, hormonal imbalance, and cardiovascular disease. To address existing diagnostic gaps, this research introduces a multifaceted hybrid framework. It synergizes machine learning algorithms to process patient-reported questionnaire data with deep learning architectures specifically designed for the automated interpretation of ovarian ultrasound imagery. XGBoost is used to analyze structured questionnaire data to assess symptom patterns, lifestyle variables, and clinical variables related to PCOS. At the same time, DenseNet-121 is employed to classify ovarian ultrasound images by learning robust visual representations. For better interpretation of the model's decisions, Grad-CAM is used to point out the areas in the ultrasound images that impact the prediction results. Experimental findings show that the proposed method provides reliable classification accuracy with transparency in the decision-making process.