Ultrasound RF Time Series for Detection of Prostate Cancer: Feature Selection and Frame Rate Analysis

This paper provides the recent results of in-vitro clinical studies to evaluate the performance of a tissue typing method, based on ultrasound RF time series, for detection of prostate cancer. In our approach, we continuously record RF echo signals backscattered from tissue, while the imaging probe and the tissue are stationary in position. The continuously recorded frames generate a time series of echo values for each spatial sample of RF signals. We extract the fractal dimension, and six spectral features from the RF time series and use them with neural networks for tissue typing. We analyze the performance of this method in detecting prostate cancer in 16 patients and demonstrate that a subset of five parameters selected from the proposed features is optimal for the diagnosis. We also study the performance of the extracted features at various frame rates and show that 22 frames per second is sufficient for efficient cancer detection with an area under ROC curve of 0.89.