Objectives
The aim of this study was to assess the diagnostic accuracy of peripheral zone prostate cancer localization by multiparametric magnetic resonance (MR) at 3 T using segmental matching of histopathology and MR images to avoid bias by image features in selection of cancer and noncancer regions.
Materials and Methods
Forty-eight patients underwent multiparametric MR imaging (MRI) on a 3 T system using a phased array body coil and spine coil elements for signal detection before prostatectomy. The examination included T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), dynamic contrast-enhanced imaging (DCE-MRI), and MR spectroscopic imaging (MRSI). Histopathology slides were correlated to T2W images and a stringent matching procedure was performed to define cancer and noncancer areas of the peripheral zone without influence of the MR image appearance. Mean T2W signal intensity, apparent diffusion coefficient, area under the enhancement curve, and choline + creatine-to-citrate signal ratio were calculated for cancer and noncancer areas. Receiver operating characteristic (ROC) analysis was performed on MR-derived parameters from the selected areas. Logistic regression was used to create models based on best combination of parameters. A simplified approach assigning a parametric score to each segment based on cutoff values from ROC analysis was also explored.
Results
By using the stringent matching procedure, 138 noncancer and 41 cancer segments were selected in the T2W images and transferred to the images of the other MR methods. A significant difference between mean values in cancer and noncancer segments was observed for all MR parameters analyzed (P < 0.001). Apparent diffusion coefficient was the best performing single parameter, with an area under the ROC curve Az,DWI of 0.90 for prostate cancer detection. Any combination of T2WI, DWI, and DCE-MRI was significantly better than T2WI alone in separating cancer from noncancer segments (Az,T2WI + DWI + DCE-MRI = 0.94, Az,T2WI + DWI = 0.92, Az,T2WI + DCE-MRI = 0.91, Az,T2WI = 0.85). The combination of T2WI and MRSI was also better than T2WI alone (Az, T2WI + MRSI = 0.90); however, the logistic regression models including MRSI did not have significant parameters. The simplified approach combining all parameters gave similar results to logistic regression combining all parameters (Az = 0.95 and 0.97, respectively).
Conclusion
By selecting histopathology defined cancer and noncancer areas without influence of image contrast, this study objectively reveals that all investigated MR parameters have the ability to separate cancer from noncancer areas in the peripheral zone individually and that any combination is better than T2WI alone.
