Robust Functional Profile Identification for DSC Thermograms

Abstract

Differential scanning calorimetry is an emerging technique with an attempt to characterize a subject's disease status according to heat capacity profiles, which are called thermograms. However, thermograms exhibit large shape variations, and the sample size is typically small. Therefore, it is important to extract robust characterization of thermograms representing the clinical status for further study. The current study identifies the representative heat capacity profiles from functional principle components which are derived from the bootstrap distribution of the deepest heat capacity function according to the functional data depth, instead of the original thermogram data set. 71 thermograms are obtained from two groups (healthy, cervical carcinoma), and functional PCA are conducted with the original thermogram data set and the bootstrap data set of the deepest heat capacity functions. Examining the first three PCs of the two groups between the two data sets, the bootstrap data set shows more distinctive difference in modes of variation between the two groups in comparison with the original thermogram data set, and the representative heat profiles are reconstructed with the PCs which are derived from the bootstrap sample sets. 90% confidence intervals of the representative heat profiles can be directly obtained from the same bootstrap set.