Towards Automated Banff Lesion Scoring: Tissue Segmentation in Kidney Transplant Biopsies using Deep Learning

Sebastiaan Ram, Dominique van Midden, Jeroen van der Laak, Linda Studer

Keywords: segmentation, nnUNet, segment anything model (SAM), kidney transplant biopsies, deep learning, Banff classification

TL;DR: We present a multi-class segmentation approach for kidney transplant biopsies that outperforms foundation models and enables reliable, automated identification of all diagnostically relevant tissue structures to support Banff lesion scoring.

Abstract: Inflammation and chronic changes in the different tissue structures (e.g., glomeruli, tubuli, interstitium) are major contributors to kidney transplant failure. Kidney transplant biopsy diagnostics is based on the Banff classification system, in which pathologists assess these changes. However, many of these factors have suboptimal reproducibility and the scoring is labor-intensive. To address this, we developed a multi-class segmentation approach that covers all tissue structures relevant for diagnostics. Our dataset comprises 99 Periodic-acid Schiff (PAS)-stained kidney transplant biopsy slides from two pathology departments. An expert pathologist manually annotated >17,000 structures across eight classes (glomeruli, sclerotic glomeruli, empty Bowman space, proximal tubuli, distal tubuli, atrophic tubuli, capsule, arteries/arterioles, and interstitium). We compared two segmentation approaches: (1) a combination of two nnU-Nets (one for tissue segmentation and one specialized for structure boundary detection) and (2) the SAM-Path foundation model. For the peritubular capillary segmentation, we used a previously developed U-Net. The nnU-Nets achieved a per-class average Dice score of 0.80, outperforming SAM-Path (0.69) and providing a reliable solution for all tissue structures relevant for kidney transplant biopsy diagnostics. Next, the nnU-Nets will be used in a reader study aimed at investigating the impact of AI on pathologists’ performance in Banff lesion scoring. The algorithm is publicly available on https://grand-challenge.org/algorithms/kidney-tissue-segmentation.

Submission Number: 26