A Diagnostic Biomarker for Breast Cancer Screening <i>via</i> Hilbert Embedded Deep Low-Rank Matrix Approximation

Thermography has been used extensively as a complementary diagnostic tool in breast cancer detection. Among thermographic methods matrix factorization (MF) techniques show an unequivocal capability to detect thermal patterns corresponding vasodilation cases. One of the biggest challenges such is selecting best representation basis. In this study, embedding method proposed address problem and Deep-semi-nonnegative (Deep-SemiNMF) for thermography introduced, then tested 208 screening First, we apply Deep-SemiNMF infrared images extract low-rank representations each case. Then, embed bases obtain one basis patient. After that, 300 imaging features, called thermomics, decode information automatic model. We reduced dimensionality thermomics by spanning them onto Hilbert space using RBF kernel select three most efficient features block Schmidt Independence Criterion Lasso (block HSIC Lasso). The preserved heterogeneity successfully classified asymptomatic versus symptomatic patients applying random forest model (cross-validated accuracy 71.36% (69.42%-73.3%)).