RESUMEN
BACKGROUND: Gleason Score (GS) is a validated predictor of prostate cancer (PCa) disease progression and outcomes. GS from invasive needle biopsies suffers from significant inter-observer variability and possible sampling error, leading to underestimating disease severity ("underscoring") and can result in possible complications. A robust non-invasive image-based approach is, therefore, needed. PURPOSE: Use spatially registered multi-parametric MRI (MP-MRI), signatures, and supervised target detection algorithms (STDA) to non-invasively GS PCa at the voxel level. METHODS AND MATERIALS: This study retrospectively analyzed 26â¯MP-MRI from The Cancer Imaging Archive. The MP-MRI (T2, Diffusion Weighted, Dynamic Contrast Enhanced) were spatially registered to each other, combined into stacks, and stitched together to form hypercubes. Multi-parametric (or multi-spectral) signatures derived from a training set of registered MP-MRI were transformed using statistics-based Whitening-Dewhitening (WD). Transformed signatures were inserted into STDA (having conical decision surfaces) applied to registered MP-MRI determined the tumor GS. The MRI-derived GS was quantitatively compared to the pathologist's assessment of the histology of sectioned whole mount prostates from patients who underwent radical prostatectomy. In addition, a meta-analysis of 17 studies of needle biopsy determined GS with confusion matrices and was compared to the MRI-determined GS. RESULTS: STDA and histology determined GS are highly correlated (Râ¯=â¯0.86, pâ¯<â¯0.02). STDA more accurately determined GS and reduced GS underscoring of PCa relative to needle biopsy as summarized by meta-analysis (pâ¯<â¯0.05). CONCLUSION: This pilot study found registered MP-MRI, STDA, and WD transforms of signatures shows promise in non-invasively GS PCa and reducing underscoring with high spatial resolution.