(Sunday, 7/29/2018) 1:00 PM - 1:55 PM
Room: Karl Dean Ballroom B1
Purpose: Random biopsies may miss the presence of high Gleason grade prostate cancer. We developed a quantitative MRI (qMRI)-based model for spatial mapping of the high-grade Gleason Pattern (GP) at a voxel-level to improve the accuracy of targeted biopsies to detect high-grade GP.
Methods: At three centers, qMRI consisting of T2 mapping, diffusion-weighted, and dynamic contrast-enhanced MRI was acquired for 30 patients prior to prostatectomy. We used generalized linear mixed-effect modelling for spatial mapping of high-grade GP (GP≥ 4). To ensure that the model is trained on voxels with a homogeneous Gleason component, we used segmentations of GP 3, 4 and 5 at a resolution of 0.5 µm/pixel on whole-mount histology sections to map to the qMRI. We averaged the voxel-level probabilities of high-grade GP per patient and used it as a measure for the probability of high-grade prostate cancer (GS≥ 4+3). We assessed the accuracy of the model with the area under the curve (AUC) from leave-one-institute-out validation, training the model on two of the institutes and calculating its performance on the third.
Results: A combination of all qMRI parameters was most predictive for localizing high-grade GP. Based on the mapping of high-grade GP inside the lesions per patient, we accurately derived the Gleason Score (GS): probabilities in the range of 0-36% for GS≤ 3+4 and 67-100% for GS≥ 4+3. The model distinguished high-grade GP, on voxel-level with AUC of 0.67 and high-grade prostate cancer, GS≥ 4+3, on patient-level with AUC of 0.70.
Conclusion: Based on qMRI-mapping of high-grade GP at voxel-level we could accurately derive the GS per patient and distinguish high-grade prostate cancer on a patient level. Use of this method is expected to improve the accuracy of qMRI-based targeted biopsies and benefit monitoring of disease progression during active surveillance.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
