Abstract
A computer-aided diagnostic system for imaging prostate cancer has been developed in order to supplement today's conventional methods for the early detection of prostate carcinoma. The system is based on analysis of the spectral content of radiofrequency ultrasonic echo data in combination with evaluations of textural, contextual, morphological and clinical features in a multiparameter approach. A state-of-the-art, non-linear classifier, the so-called adaptive network-based fuzzy inference system, is used for higher-order classification of the underlying tissue-describing parameters. The system has been evaluated on radio-frequency ultrasound data originating from 100 patients using histological specimens obtained after prostatectomy as the gold standard. Leave-one-out cross-validation over patient data sets results in areas under the ROC curve of 0.86 ± 0.01 for hypoechoic and hyperechoic tumors and of 0.84 ± 0.02 for isoechoic tumors, respectively.