Design and Application of Signal Modeling, Segmentation and Classification Methods for High-Frequency Ultrasound Backscatter Signals

In this study, we explore the possibility of monitoring program cell death (apoptosis) and classifying clusters of apoptotic cells based on the changes in high frequency ultrasound backscatter signals from these cells. One of the hallmarks of cancer is that the fail [sic] in the apoptosis mechanism in cells. Therefore this research carries the promise of designing more refined and more effective cancer therapies. The ultrasound signals are modeled through the Autoregressive (AR) modeling technique. The proper model order is calculated by tracking the error criteria derived from statistical properties of the original and modeled signal. In the next stage, five machine learning classifiers are developed to classify backscatter signals based on their AR coefficients. In clinical applications ultrasound backscatter signals from tissues and tumors are most likely to be non-stationary. Therefore analyzing such signals requires signal segmentation techniques. We developed recursive least square lattice filter for adaptive segmentation of ultrasound backscatter signals from multiple cell types into blocks of stationary segments, and model and classify the segments individually. In this thesis we demonstrate the accuracy of modeling, segmentation and classification techniques to detect signals from different cell pellets based on the signal processing and machine learning techniques.