Modeling for Semantic Segmentation and Classification of Breast Thermograms Using MobileNetV3 and Boosted Tree

Breast cancer is a leading cause of mortality among women worldwide, making early detection essential for improving survival rates. While various detection methods exist, infrared thermography offers a safe, non-invasive approach by analyzing skin surface temperature distribution. However, achieving accurate segmentation and classification remains challenging. This research proposes a breast cancer detection model using semantic segmentation with MobileNetV3 and classification with the Boosted Tree algorithm. Semantic segmentation isolates the breast area from the background, while classification identifies potential cancer presence within this area. The model achieved a segmentation accuracy of 91.06% (learning rate 10–3) and classification accuracy of 86.44%, with average processing times of 0.55 seconds for segmentation and 0.17 seconds for classification. These results indicate the model's suitability for clinical applications, with a total model size of just 7.2 MB.