Attention-based Transfer Learning Approach using Spatial Pyramid Pooling for Diagnosis of Polycystic Ovary Syndrome

A considerable percentage of pregnant women are affected by PCOS, or polycystic ovarian syndrome, a metabolic condition. It causes higher amounts of androgens, or male hormones, and irregular menstrual periods. Follicles, which are tiny sacs packed with fluid that form on the ovaries and impede the regular release of eggs, are a common side effect of PCOS. Initial detection and weight control might help prevent the detrimental effects of PCOS, even though the exact aetiology is still unknown. With the help of the Spatial Pyramid Pooling Network (ASPPNet), a novel attention-based transfer learning approach for PCOS identification from ultrasound pictures is developed in this work. Initially, ResNet-18 is used to extract fundamental features, and the attention module is applied to extract significant disease-related features. Later, the Spatial pyramid pooling technique is utilized to preserve multi-scale features by adopting a fixed-length illustration that is not affected by image size/scale. Spatial pyramid pooling is additionally resistant to image component distortions. The experimental results demonstrate that the ASPPNet exceeds other well-known algorithms with a $\mathbf{9 8 . 7 9 \%}$ accuracy, demonstrating its applicability in both medical and industrial applications.