Classification of terahertz-pulsed imaging data from excised breast tissue

We investigate the efficacy of using data reduction techniques to aid classification of terahertz (THz) pulse data obtained from tumor and normal breast tissue. Fifty-one samples were studied from patients undergoing breast surgery at Addenbrooke's Hospital in Cambridge and Guy's Hospital in London. Three methods of data reduction were used: ten heuristic parameters, principal components of the pulses, and principal components of the ten parameter space. Classification was performed using the support vector machine approach with a radial basis function. The best classification accuracy, when using all ten components, came from using the principal components on the pulses and principal components on the parameter, with an accuracy of 92%. When less than ten components were used, the principal components on the parameter space outperformed the other methods. As a visual demonstration of the classification technique, we apply the data reduction/classification to several example images and demonstrate that, aside from some interpatient variability and edge effects, the algorithm gives good classification on terahertz data from breast tissue. The results indicate that under controlled conditions data reduction and SVM classification can be used with good accuracy to classify tumor and normal breast tissue.