Comparing and Combining Artificial Intelligence and Spectral/Statistical Approaches for Elevating Prostate Cancer Assessment in Bi-Parametric MRI: A Pilot Study

Background Prostate cancer management optimally requires non-invasive, objective, quantitative, accurate evalua-tion of prostate tumors. Current research applies visual inspection and quantitative approaches, such as artificial intelligence (AI) focused in deep learning (DL) to evaluate MRI. Recently, a different spectral/statistical approach successfully evaluated spatially registered bi-parametric MRIs for prostate cancer. This study aimed to further as-sess and improve the spectral/statistics approach through benchmarking and combining with AI. Methods A zonal-aware self-supervised mesh network (Z-SSMNet) was applied to the same 42-patient cohort from previous spectral/statistical studies. The probability of clinical significance of prostate cancer (PCsPCa) and detec-tion map with affiliated tumor volume, eccentricity were computed for each patient. Linear and logistic regression were applied to International Society of Urological Pathology (ISUP) grade and PCsPCa, respectively. The R, p-value, Area Under the Curve (AUROC) from Z-SSMNet output was computed The Z-SSMNet output was combined with spectral/statistics output for multiple-variate regression. Results The R (p-value), AUROC [95% Confidence Interval] from Z-SSMNet algorithm relating ISUP to PCsPCa is 0.298 (0.06), 0.50 [0.08-1.0], to average blob volume is 0.51 (0.0005), 0.37 [0.0-0.91], to total tumor volume is 0.36 (0.02), 0.50 [0.0-1.0]. R (p-value), AUROC computations showed much poorer correlation for eccentricity derived from Z-SSMNet detection map with ISUP. Overall DL/AI performed less well relative to spectral/statistical approaches from previous studies. Multi-variable regression fitted AI average blob size and SCR results in R=0.70 (0.000003) significantly higher than univariate regression fits of AI, spectral/statistics alone. Conclusions. Spectral/statistical approaches performed well relative to Z-SSMNet. Combining Z-SSMNet with spectral/statistics approaches significantly enhanced tumor grade prediction, possibly providing an alternative to current prostate tumor assessment.