PCOS-WaveConvNet: A Wavelet Convolutional Neural Network for Polycystic Ovary Syndrome Detection using Ultrasound images
Shamik Tiwari, Piyush Maheshwari
Abstract
Women of reproductive age are susceptible to polycystic ovarian syndrome (PCOS), a hormonal condition. Multiple small follicles or cysts on the ovaries are one of the symptoms of PCOS and can be found using ultrasound imaging. Wavelet ConvNets have been applied in various applications, including image classification, object detection, and biomedical signal analysis. A Wavelet ConvNet is a type of deep learning model that applies wavelet transformation to input data before feeding it into a convolutional neural network. The wavelet transform is a mathematical technique that breaks down a signal or image into a series of sub-bands, each representing different frequency components of the original data. In this work, A 2D Discrete Wavelet Transform (2D-DWT) with the Haar wavelet is applied to each image. The resulting sub-bands namely Low-Low (LL), Low-High (LH), High-Low (HL), and High-High (HH) are concatenated to create a 4-channel feature map. Further, this concatenated feature map is fed into the ConvNet for classification. The PCOS-WaveConvNet classifier has attained 99.7% accuracy which is better than a usual ConvNet model.