Ultrasonic and magnetic‐resonance spectra in tissue‐type imaging for planning and monitoring prostate cancer treatment

Abstract

No reliable method of imaging prostate cancer currently exists. These prostate tissue‐typing studies aim to improve the effectiveness of biopsy guidance and treatment monitoring by developing better imaging methods for identifying cancerous prostate tissue. Success will reduce the false‐negative rate of biopsies and treatment side‐effects. Ultrasonic (US) radio‐frequency (rf) echo‐signal data acquired during biopsy examinations were used to compute spectral parameters. These spectral parameters along with clinical parameters, e.g., PSA, were used to train a neural network classifier using biopsy results as the gold standard. Cancer‐likelihood scores from a look‐up table were used to generate tissue‐type images (TTIs). The ROC‐curve area for US neural‐network‐based classification was 0.844±0.018 vs 0.638±0.031 for B‐mode‐based classification, and the sensitivity of neural‐network based classification was superior to that of B‐mode‐guided biopsies. These classification methods are being extended to include magnetic‐resonance spectral (MRS) techniques that use the choline to citrate ratio‐to‐distinguish cancerous from noncancerous prostate tissue. 3D renderings of prostatectomy histology, US, and MR images show encouraging correlations, and combining MRS parameters with US spectral parameters appears to have potential to further improve prostate‐cancer imaging. [Work supported in part by NIH Grant CA053561.]