DensiThAI: A Multi-View Deep Learning Framework for Breast Density Estimation Using Infrared Images.

Breast tissue density is an established biomarker of breast cancer risk and an important determinant of mammographic sensitivity. Density assessment is typically performed using X-ray mammography. In this study, we investigate whether breast tissue density-related information may be reflected in infrared thermal images using artificial intelligence. The underlying hypothesis is that fibroglandular and adipose tissues differ in thermophysical and physiological properties, potentially giving rise to subtle surface temperature patterns. We propose DensiThAI, a multi-view deep learning framework that integrates information from five standardized thermal breast views to classify density. The framework was evaluated on a multi-center dataset of 3500 women using mammography-derived breast density labels as the reference standard. DensiThAI achieved a mean AUROC of 0.73 across 10 independent test splits, with statistically significant separation between density classes (p < 0.05). These findings suggest that density-associated thermal patterns may be detectable and motivate further investigation of thermal imaging as a complementary modality for breast tissue characterization in larger and independently validated cohorts.