Validation of Direct Registration of Whole-Mount Prostate Digital Histopathology to ex vivo MR Images

摘要

Accurate determination of cancer stage and grade from in vivo prostate imaging could improve biopsy guidance, therapy selection and, possibly, focal therapy guidance. Validating prostate cancer imaging ideally requires accurate 3D registration of in vivo imaging to histopathology, which is facilitated by intermediate histology- ex vivo imaging registration. This work introduces and evaluates a direct registration with fiducial-based local refinement of digital prostate histopathology to ex vivo magnetic resonance (MR) images that obviates three elements typical of existing methods: (1) guidance of specimen slicing, (2) imaging/photography of sliced tissue blocks, and (3) registration guidance based on anatomical image features. The mean target registration error (TRE) of 98 intrinsic landmarks across 21 histology images was calculated for the proposed direct registration (0.7 mm) and compared to existing approaches: indirect using tissue block MR images (0.8 mm) and image-guided-slicing-based (1.0 mm). The local refinement was also shown to improve existing approaches to achieve a similar mean TRE (0.7 mm).