JS divergence Weibull embedding for very early breast cancer diagnosis

Dynamic thermography has emerged as a reliable adjunctive tool in breast cancer screening, complementing traditional methods such as mammography and clinical breast examination (CBE). Thermographic imaging, yielding thermal biomarkers termed thermomics, exhibits promise in identifying vasodilation within breast tissue, signaling potential abnormalities and lesions. Moreover, the observation of heterogeneous thermal patterns facilitates the detection of angiogenesis, the process of new blood vessel formation. This study investigates the application of thermal imaging biomarkers and thermographic imaging in breast cancer screening by applying Jensen‐Shannon (JS) divergence calculation on the low rank representation of thermal image sets obtained through Uniform Manifold Approximation and Projection (UMAP) to select images best for feature extraction and employing Weibull embedding to highlight heterogenous patterns in the thermal sequences. The results are used to extract high‐dimensional thermomics and spectral clustering is used to reduce feature abundance. The model, trained with consistent hyperparameters across comparisons, demonstrated favorable preliminary performance in predicting abnormality. These optimal biomarkers have the potential to capture thermal heterogeneity effectively, thereby facilitating the early detection of breast cancer and serving as a non‐invasive aid to CBE.