ESDPCOS: Effectiveness of Combined CNN and Handcrafted Features for Ovarian Cyst Detection in PCOS Patients using Ultrasound Images

Polycystic Ovary Syndrome (PCOS) is a significant contributor to infertility among women of reproductive age, along with other health conditions such as preterm abortions and anovulation, which impact the health of women worldwide. This endocrine condition, affecting 5-10% of women in their reproductive years, is characterized by irregular hormone levels, including an excess of male hormones. PCOS leads to ovarian dysfunction, increasing the risk of miscarriage and infertility. Additionally, PCOS symptoms include obesity, irregular menstrual cycles, acne, hirsutism, and various other manifestations. Therefore, diagnosing PCOS is challenging due to the diverse range of symptoms and the presence of several related gynecological disorders. Diagnosing PCOS requires many clinical tests and ovary scans, which can be time-consuming and costly for affected patients. Many of the existing models using Convolution Neural Network (CNN), for detecting PCOS reported moderate results in classifying PCOS and non-PCOS cases. However, further improvements in accuracy can be achieved by combining handcrafted features with CNN features. Therefore, this paper introduces an Expert System to Detect PCOS (ESDPCOS) that aims to accurately and efficiently classify PCOS and non-PCOS cases using ultrasound images. The system employs a hybrid approach, utilizing the CNN model for deep feature extraction and the Gray-Level Co-occurrence Matrix (GLCM) technique for local feature extraction. Then the combined features are processed through a fully connected layer to detect PCOS. The proposed model has achieved an impressive accuracy of 96.06%.