An Automated Thermography-Based Breast Cancer Detection and Localization System

Globally, breast cancer is the primary cause of death for women, illustrating the importance of early detection to increase treatment effectiveness and survival rates. Even though traditional mammography is still the most widely used method of breast cancer detection, it has drawbacks such radiation exposure and patient discomfort. A non-invasive diagnostic method called breast thermography has gained traction as a possible supplement to breast cancer screening. This paper introduces an innovative approach that harnesses deep learning, specifically employing the Mask Scoring Region-based Convolutional Neural Network, to analyze breast thermography images for identifying abnormal tissue patterns. In our proposed method, the preprocessing of thermographic images is used to improve their quality and extract features, which in turn, facilitate subsequent deep learning analysis. Our model utilizes Mask Scoring R-CNN to recognize and segment regions of interest in breast thermograms, enabling a precise delineation of temperature irregularities within breast tissue. Further, validation tests on a diverse range of breast thermography images, including those depicting both benign and malignant lesions. The experimental results clearly show that our technique is successful in identifying and localizing aberrant thermal patterns in breast thermograms. These results indicate that deep learning-based breast thermography analysis can function as a non-invasive and radiation-free complement to traditional breast cancer screening methods. The suggested method has a high potential for early breast cancer diagnosis and may improve patient outcomes.