NEW METHODS FOR THE AUTOMATED DETECTION OF ABNORMALITIES IN ULTRASOUND IMAGES OF THE OVARY

Polycystic ovary syndrome (PCOS), a hormonal disorder with diverse diagnostic criteria, is currently diagnosed through manual analysis of ultrasound images, requiring time-consuming and subjective tracking and measurement of follicles. This paper presents a novel automated approach to PCO detection. Our method employs a modified Otsu approach, where several optimization algorithms including Particle Swarm Optimization (PSO) and its variations, Pigeon Inspired Optimization (PIO), and Invasive Weed Optimization (IWO) maximize class variance to achieve optimal follicle segmentation. Following follicle extraction from ultrasound images, their features are automatically quantified using stereology, and these attributes are stored as feature vectors. These vectors are then used for binary classification based on PCO presence or absence. Correct segmentation performance is evaluated using standard quality metrics. This automated approach has the potential to significantly improve the accuracy, objectivity, and efficiency of PCO diagnosis, reducing reliance on manual assessment.