Dynamic Thermography-Based Early Breast Cancer Detection Using Multivariate Time Series

Abstract

A computational approach for early breast cancer detection using Dynamic Infrared Thermography (DIT) was developed. Thermograms are represented by multivariate time series extracted from thermal hotspots in the breast, capturing five features: maximum and mean temperature, spatial heterogeneity, heat flux, and tumor depth, over 20 thermograms. Features are estimated based on the inverse solution of the Pennes bio-heat equation. Classification is performed using a Time Series Forest (TSF) and a Long Short-Term Memory (LSTM) network. The TSF achieved an accuracy of 86%, while the LSTM reached 94% accuracy. These results indicate that dynamic thermal responses under cold-stress conditions reflect tumor angiogenesis and metabolic activity, demonstrating the potential of combining multivariate thermographic sequences, biophysical modeling, and machine learning for non-invasive breast cancer screening.