Ultrasound-Based Characterization of Prostate Cancer Using Joint Independent Component Analysis

摘要: Objective: This paper presents the results of a new approach for selection RF time series features based on joint independent component analysis in vivo characterization prostate cancer. Methods: We project three sets extracted from spectrum, fractal dimension, and wavelet transform ultrasound data space spanned by five components. Then, we demonstrate that obtained mixing coefficients group patients can be used to train classifier, which applied characterize cancerous regions test patient. Results: In leave-one-patient-out cross validation, an area under receiver operating characteristic curve 0.93 classification accuracy 84% are achieved. Conclusion: Ultrasound accurately cancer, without need exhaustive search feature space. Significance: use systematic fusion multiple features, within machine learning framework, PCa study.