A lightweight SVM-based framework for breast thermography using statistical descriptors

Breast cancer remains a major global health challenge, necessitating the development of non-invasive, efficient, and accessible diagnostic tools. While infrared thermography (IRT) has emerged as a promising complementary technique, most current approaches rely on complex deep learning architectures that function as “black boxes” and demand high computational resources. In this work, we propose a lightweight diagnostic framework based on three compact and interpretable thermal descriptors: maximum temperature (T_imax ), standard deviation (Sd_Ti), and skewness (Sk_Ti). These parameters capture the peak metabolic heat, thermal dispersion, and statistical asymmetry of the tissue with direct physical meaning. The framework was evaluated using the Database for Mastology Research with Infrared Images (DMR-IR) through a Support Vector Machine (SVM) classifier. Our results demonstrate that this low-dimensional representation achieves high diagnostic precision, reaching an Area Under the Curve (AUC) of 0.98 and an accuracy of 96.8%. The simplicity of the proposed method significantly reduces processing costs while maintaining robust performance, making it a viable candidate for integration into portable medical devices and point-of-care screening in low-resource settings.