ROISegNet: A Novel Deep Learning Model for Automatic ROI Segmentation in Breast Thermogram Imagery

Abstract

Breast cancer is a leading cause of death for women worldwide. Breast thermography, a non-invasive imaging technique, can detect breast irregularities that may lead to breast cancer in the long run. Early identification of the risk of breast cancer can potentially prevent the disease and save lives. AI has emerged as a technological breakthrough in the healthcare sector and can be used with breast thermography and deep learning (DL) techniques for early breast cancer diagnosis.This study uses DL for breast region of interest (ROI) segmentation to facilitate breast cancer screening. Semantic segmentation using the atrous convolution DL model can effectively separate the ROI from breast thermography images. The deep learning system, ROISegNet, includes a decoder module that utilizes atrous convolution in the encoder, enhancing the efficiency of semantic segmentation. The framework's realization is demonstrated through an algorithm called Intelligent Segmentation of Breast ROI (ISBROI). According to our empirical investigation using the benchmark dataset DMR-IR, ROISegNet achieved a maximum accuracy of 98.63%, outperforming other models such as VGG19, ResNet50, InceptionV3, and Atrous Convolution.