• prostate cancer
• PI-RADS
• MAI
• ADC

Objective - Multiparametric magnetic resonance imaging (mp-MRI) combines morphological and functional information and is being increasingly used to detect prostate cancer (PCa). Computer aided detection (CAD) systems have the potential to support the radiologist by indicating suspicious regions and reducing oversight and perception errors. Our objective is to evaluate the diagnostic performance of a computer-assisted diagnostic system (Watson Elementary®), in comparison to human readers, in identifying and localizing potentially malignant lesions within the prostate gland, based on the analysis of MR images. We also investigated the relationship between apparent diffusion coefficient (calculated from diffusion-weighted MR images), Malignancy Attention Index (calculated from Watson Elementary®) and Gleason score.
Materials and Methods - 22 patients with suspected prostate gland malignancy were studied mp-MRI (using T2‐weighted, T1‐weighted dynamic contrast enhanced and diffusion‐weighted sequences). A fully automated analysis tool was used to perform a retrospective analysis on sets of mp-MR images. The resulting malignancy predictions were compared to the original pathologic assessments of a combination of cognitive and fusion targeted biopsies, performed with TRUS/MRI‐fusion based image guided biopsy equipment. The diagnostic performance of the automated malignancy prediction algorithm was then compared to results reported in human reader studies.
Results - The automated analysis yielded diagnostic accuracies that were comparable to those of human readers. With a sensitivity of 94,4% and a specificity of 100% in detecting Gleason score ≥ 6 PCa, the automated analysis results correspond well within the reported confidence intervals with diagnostic accuracies reported in literature.
Conclusions - Our preliminary data reveal comparable diagnostic accuracies of an MAI based automated analysis and a PI‐RADS scoring based expert human reader analysis of multiparametric MRI data for the detection of PCa. This study also suggests a possible correlation between MAI maps and ADC values. Further large scale studies are certainly needed to confirm our initial results.