Terahertz Pulse Data Dimensional Reduction and Classification for Hepatic Tissue Samples

Terahertz (THz) pulse combining with chemometrics methods was presented for discerning the liver cancer tissues. Deconvolution technique was employed to extract the impulse function to overcome system echo and disturbance in the atmosphere. To avoid the redundancy and noise of the THz pulse data, two data compression techniques, principle component analysis (PCA) and isometric feature mapping (Isomap), were conducted in reducing the dimensionality of the pulse data. In comparison of two-dimensionality scatter diagram, the Isomap plot could divide two classes with the certain distances. Support vector machine (SVM) and probabilistic neural network (PNN) were performed to classify the two group based on dimensional reduction strategies. The accuracies of PNN and SVM based on Isomap were better than those of PCA. The proposed scheme has capable of identifying the hepatic tumors in a sense.