Pathological Features Associated with False-Negative Multiparametric MRI in Prostate Biopsies

Introduction: The clinical adoption of multiparametric MRI (mpMRI) has improved prostate cancer management by reducing unnecessary biopsies and improving detection of clinically significant prostate cancer (csPCa). However, false-negative (FN) rates in mpMRI are notable, potentially resulting in delayed diagnosis and disease progression. This study explores how computational analysis of pathological microscopic tissue composition can explain these diagnostic failures in mpMRI interpretation.MethodsWe analyzed 2,152 H&E stained digital pathology slides (247 positive biopsy blocks) from 124 patients at Northwestern Memorial Hospital. Patients with unilateral mpMRI lesions but subsequent bilateral positive biopsies were selected by linking pathology records to imaging reports. Biopsy blocks were categorized as: mpMRI-visible (csPCa ipsilateral to the mpMRI lesion) or mpMRI-invisible (csPCa contralateral to the mpMRI lesion). Using CellVIT, a foundational model for cell detection, we identified four cell types, including malignant neoplastic, inflammatory, connective, and epithelial cells. We then quantified five patient-level features based on cellular densities, specifically neoplastic, inflammatory, connective, and epithelial cell densities, along with overall cellularity to compare mpMRI-visible and mpMRI-invisible biopsies and investigate factors that may contribute to false-negative mpMRI findings. Pearson correlation coefficients (r) were calculated to assess relationships among cellular features and their potential redundancy. Statistical analyses were based on two-sided Mann-Whitney, with a significance level set at 0.05.ResultsCellularity emerged as the most significant discriminator between MRI-visible and MRI-invisible positive biopsies (p-value=0.0015, AUC=0.6169). Neoplastic cell density showed modest significance (p-value=0.0473, AUC=0.573), and connective, epithelial, and inflammatory cell densities did not reach statistical significance. Correlation analysis also revealed strong negative relationships between neoplastic density and connective tissue density (r=-0.68), while cellularity showed minimal correlation with other tissue components (|r|<=0.09).ConclusionOverall cellularity and neoplastic cell density measured from patholgical biopsy samples substantially correlate with the visibility of prostate cancer lesions on mpMRI. Specimens with higher cellularity were more likely to be detected in mpMRI, potentially due to their impact on tissue density and contrast characteristics. This integrative approach of linking computational pathology with imaging could advance our understanding of the biological basis for false-negative mpMRI readings and may guide refinements in imaging protocols and interpretations.FundingPolsky Urologic Cancer Institute Award, SPORE grant in Prostate Cancer.

Learning Objectives: 

1. Understand how computational digital pathology can reveal pathological signatures, such as cellularity and malignant cell density, that can explain false-negative results in multiparametric MRI (mpMRI).
2. Prostate specimens with higher cellularity were more likely to be detected in mpMRI, potentially due to their impact on tissue density and contrast characteristics.
3. AI can characterize digitized pathology slides to address clinical questions and uncover imaging biomarkers/pathological signatures that help explain diagnostic discrepancies.