Introduction: We propose a novel approach for identification (ID) of clinical significant PCa on mpMRI based upon retrospective comparison of in vivo mpMRI images to spatially concordant digitally-scanned post-prostatectomy H&E images. Steps: 1) localisation and Gleason grading of tumour foci; 2) reconstruction of H&E slides in 3D; 3) alignment of reconstructed histology to mpMRI images; 4) labelling of aggressive prostate cancer on mpMRI images using reconstructed 3D histology; 5) training a deep learning-based model for unsupervised segmentation of aggressive PCa foci using mpMRI images and transferred labels. Using this approach, the extent of cancer can be mapped directly onto mpMRI enabling accurate segmentation of voxels corresponding to tumour foci, including ID of mpMRI-invisible lesions using radiomic features.
Methods: The step 2 is presented herein. Whole-mount histopathological sections from totally embedded radical prostatectomy specimens, with correspondent diagnostic pre-biopsy mpMRI, were used. Apex and base tissue blocks were cut perpendicular to the axial plane, with the central portion of the gland sliced in 4 mm thick sections. H&E whole-mount slides were digitised at 40x. 3D reconstruction was performed using a novel computational strategy that includes: 1) angular alignment of individual macrodissected tissue pieces using colour ink markers; 2) scale alignment to fit the pieces on a pre-defined bounding box; 3) generation of intermediate layers between two pieces; 4) normal vector estimation; and 5) Poisson reconstruction to generate the triangular mesh.
Results: The volume estimate from the original prostate specimen was compared to the reconstructed volume to assess the 3D reconstruction performance with a correlation of 85%-88%. Because the base and apex portions are not discarded, we establish a high correlation between the reconstructed 3D histopathological volume and actual prostate volume. Further, this methodology allowed ID not only of independent tumour foci within the gland but also 3D reconstruction of the different Gleason patterns with accurate estimation of tumour volume and prognostic Grade Group.
Conclusions: A method for reconstructing 3D prostate volumes from 2D histology images has been presented. The development of radiomic and deep learning algorithms to automatically detect prostate cancer on MRI will be aided by the accurate labelling of tumour foci on mpMRI images using our 3D reconstruction approach.