Application of Machine Learning for Early Diagnosis in Women’s Health: A Case Study on Polycystic Ovary Syndrome (PCOS)

This thesis investigates the potential of machine learning techniques to support early diagnosis of Polycystic Ovary Syndrome (PCOS), a highly prevalent yet frequently underdiagnosed condition in women’s health. Motivated by persistent diagnostic delays, overlapping symptoms with other endocrine and metabolic disorders, and structural gender biases in clinical assessment, the study combines structured-data modelling and deep learning approaches applied to ultrasound imaging. Using a publicly available clinical dataset, multiple supervised learning algorithms were compared following an extensive preprocessing and feature selection pipeline. To complement this analysis, a deep learning framework based on a ResNet50V2 backbone was developed for automated PCOS detection from transvaginal ultrasound images. Overall, the findings highlight both the promise and the limitations of data-driven methods for PCOS identification. While results are encouraging, the thesis emphasizes the need for larger, standardized, and multicenter datasets to ensure external validity and clinical applicability. The study contributes to ongoing efforts toward more objective, accessible and equitable diagnostic pathways in women’s health.