| Abstract: |
Spectrum analysis of radiofrequency (RF) ultrasonic echo signals often can sense tissue differences that are not visible on conventional ultrasonic images. Spectrum-analysis parameter values combined with other variables, such as serum prostate specific antigen (PSA) concentration, can be classified by neural networks to distinguish effectively between cancerous and noncancerous prostate tissues. Images based on neural network classification of spectral parameters and clinical variables can be advantageous for biopsy guidance, staging, and treatment planning and monitoring. A study based on 644 biopsies from 137 patients showed that these methods are significantly superior to B-mode image interpretation for differentiating cancerous from noncancerous prostate tissues. Using the histologic determination of tissue types as the gold standard, the area under the receiver-operator characteristic (ROC) curve for neural network classification based on spectrum analysis and PSA value for the 644 biopsies was 0.87 +/- 0.04, and the ROC curve are for a level-of-suspicion (LOS) assignment based on B-mode imaging was 0.64 +/- 0.04, Color-encoded and gray-scale images derived from neural network assignment of suspicion for cancer at each pixel location showed remarkable detail and suggested potential clinical value for biopsy guidance using real-time two-dimensional (2D) images and staging, treatment planning, and monitoring using three-dimensional (3D) images. |
