Thermal Radiomics for Improving the Interpretability of Breast Cancer Detection from Thermal Images

Abstract

Mortality rates of breast cancer are expected to increase over the next 10 years. Detection of cancers in their early stage can help in combating this disease and improving the survival rates. The traditional breast cancer detection techniques that worked in high income countries might not be applicable in low- and middle-income countries owing to implementation challenges. Thermography is re-emerging as an affordable imaging modality to enable screening in these countries. However, manual interpretation of thermograms is subjective and not accurate. Thermalytix is a novel fusion of machine learning and thermography to alleviate the subjectivity and improve the accuracy and interpretability of breast cancer detection. In this paper, we discuss three different thermal radiomics employed by Thermalytix to characterize different thermal patterns in the breast region. These thermal radiomics are interpretable and play an important role in clinical adaptation. When we tested Thermalytix with these radiomics on thermal data obtained from two different clinical studies involving 717 women, it resulted in an AUROC of 0.944 with a sensitivity and specificity of 90.6% and 85.3%, respectively. This shows the potential of Thermalytix as a promising tool for breast cancer detection.