Multimodal radiothermomic biomarkers for breast cancer screening

Abstract

Dynamic thermography has been widely used as a diagnostic tool in breast cancer screening before mammography and with clinical breast examination (CBE). Thermal imaging biomarkers, thermomics, are proven to highlight the heterogeneous thermal patterns and vasodilation indicating abnormalities in the area due to angiogenesis blood vessel formation. This study shows two sets of analyses. The first set is a feasibility study involving a combined multimodal imaging biomarker using mammographic and thermographic imaging for 11 cases of breast cancer screening. The second part of this paper shows the application of the t-distributed stochastic neighbor embedding (tSNE) method to provide a low dimensional representation of thermal sequence and tested for 55 breast cancer screening participants. We extracted high dimensional radiomics and thermomics and reduced the dimensionality of these features using spectral embedding technique, and trained a random forest model with tuned hyperparameters to perform diagnostic prediction. The results of tSNE combining clinical and demographics yield 77.4% (69.8%, 86.8%), while the highest accuracy belonged to Sparse PCT + Clinical with 79.3% (73.6%, 84.9%). The proposed method results indicated that the tSNE can preserve thermal patterns driven radiothermomics, which leads to significantly aid in CBE and early detection of breast cancer.