Correlating Histopathology and Radiologic Imaging to Validate Prostate Cancer Detection

Radiologic imaging is entrenched in standard practice for screening men with suspicion of prostate cancer (PCa) and to inform treatment decisions. Correlating in-vivo imaging to the radical prostatectomy (RP) specimen, however, is a challenging task due to the vast differences in resolution, scale, and deformation of the tissue. Microscopic examination of the RP specimen represents the gold-standard for diagnosis of PCa, so correlation with radiologic imaging is necessary in order to validate PCa detection competency. In this work, a methodology and program are presented with three aims: one, match histopathology slices to their corresponding radiologic images; two, register histopathology images to their corresponding radiologic image; and three, validate PCa detection in the radiologic image based on the registration with histopathology. Diffusion weighted imaging (DWI) (b50-b2000) was used in this work for proof of concept. Histopathology slides were scanned and processed in order to be compatible with a novel registration program. A slice matching method was established and a program, process_H, was created in MatLab (v2017b. Natick, MA) in order to register reconstructed histopathology slices to their closest corresponding DWI slice. A cohort of three RP patients, 29 total histopathological slices, with pre-operative multi-parametric magnetic resonance images (mpMRIs) was identified. Using this cohort, a set of histological images was prepared and process_H registered the respective images, creating validation sets for the registration as well as binary pixel-by-pixel data sets defining PCa by Gleason Score (GS). These pixel-by-pixel data sets were then used to validate PCa detection from the in-vivo DWIs by means of receiver operating characteristic (ROC) curves. This work specifically compares the diffusion coefficient (D), distributed diffusion coefficient (DDC), and a heterogeneity index (α) to GS. Our results show utility for our methodology and program in the validation of PCa. Further development will make this tool a valuable addition for validating new imaging sequences and those in current use.