A Comparative Study of Deep Learning Techniques for Breast Cancer Detection Using Mammography, MRI, and Thermal Imaging

Abstract

Breast cancer remains a critical global health concern with early detection being vital for effective treatment. This study presents a comparative evaluation of three deep learning techniques—CNN, RCNN, and CNN-LSTM—across three distinct imaging modalities: mammography, MRI, and dynamic thermal imaging. The CNN model was applied to mammogram datasets (CBIS-DDSM, INbreast, MIAS), the RCNN was trained on annotated MRI scans, and the CNN-LSTM model utilized sequential thermal images (DMR-IR dataset). Evaluation metrics such as accuracy, precision, recall, F1-score, and AUC were used to assess model efficacy. Our findings reveal that dynamic thermal imaging with temporal modeling outperforms traditional mammography-based CNNs and even MRI-based RCNNs in classification performance. The study concludes with a discussion on the potential of radiation-free, non-invasive thermal imaging for widespread deployment in resource-constrained settings.