Accurate Classification of Burn Injuries Using Support Vector Machines and the Wavelet Shannon Entropy of the THz-TDS Waveforms

The accuracy of clinical assessment of partial-thickness burn injuries has remained as low as 60% in the first few days post burn induction. Here, we present the implementation of a wavelet Shannon entropy technique for noninvasive characterization of burn injuries in an in vivo porcine burn study. Supervised machine learning using the support vector machines (SVM) based on the energy to Shannon entropy ratio (ESER) in the wavelet packet transform of the THz-TDS waveform yielded accuracy rates above 91% in differentiation between superficial, intermediate, and full-thickness burn categories.