Glo-In-One-v2: Holistic Identification of Glomerular Cells, Tissues, and Lesions in Human and Mouse Histopathology

TL;DR

A convolutional neural network for multiclass segmentation of intraglomerular tissue and lesions and transfer learning from rodent to human for the glomerular lesion segmentation model has enhanced the average segmentation accuracy across different types of lesions by more than 3%.

Abstract

Segmenting glomerular intraglomerular tissue and lesions traditionally depends on detailed morphological evaluations by expert nephropathologists, a labor-intensive process susceptible to interobserver variability. Our group previously developed the Glo-In-One toolkit for integrated detection and segmentation of glomeruli. In this study, we leverage the Glo-In-One toolkit to version 2 with fine-grained segmentation capabilities, curating 14 distinct labels for tissue regions, cells, and lesions across a dataset of 23,529 annotated glomeruli across human and mouse histopathology data. To our knowledge, this dataset is among the largest of its kind to date.In this study, we present a single dynamic head deep learning architecture designed to segment 14 classes within partially labeled images of human and mouse pathology data. Our model was trained using a training set derived from 368 annotated kidney whole-slide images (WSIs) to identify 5 key intraglomerular tissues covering Bowman's capsule, glomerular tuft, mesangium, mesangial cells, and podocytes. Additionally, the network segments 9 glomerular lesion classes including adhesion, capsular drop, global sclerosis, hyalinosis, mesangial lysis, microaneurysm, nodular sclerosis, mesangial expansion, and segmental sclerosis. The glomerulus segmentation model achieved a decent performance compared with baselines, and achieved a 76.5 % average Dice Similarity Coefficient (DSC). Additional, transfer learning from rodent to human for glomerular lesion segmentation model has enhanced the average segmentation accuracy across different types of lesions by more than 3 %, as measured by Dice scores. The Glo-In-One-v2 model and trained weight have been made publicly available at https: //github.com/hrlblab/Glo-In-One_v2.

Figures (6)

• Figure 1: Involved glomerular classes. This figure presents fine-grained classes of intraglomerular tissue, including Bowman’s capsule (Cap), tuft (Tuft), mesangium (Mes), mesangial cells (Mec), and podocytes (Pod). It also highlights glomerular lesions observed in rodents and humans: adhesion (AH), capsular drop (CD), global sclerosis (GS), hyalinosis (HS), mesangial lysis (ML), microaneurysm (MA), nodular sclerosis (NS), mesangial expansion (ME), and segmental sclerosis (SS).
• Figure 2: Toolkit Overview. This figure provides an overview of the Glo-In-One-v2 toolkit. The proposed toolkit is able to achieve 14 segmentation classes using a single command Docker command line. The input consists of raw WSIs, and the output is a holistic segmentation of glomeruli. The detection module, inherited from the previous toolkit version, delivers quantitative detection of glomeruli. The segmentation module utilizes a trained model, developed from patches extracted from WSIs with manual annotations provided by medical experts.
• Figure 3: Overview of the Rodent-to-human transfer learning framework. This figure provides an overview of transfer learning for glomerular segmentation from rodent to human, where Fig \ref{['fig:zR2H']} illustrates the direct adaptation of a model trained on rodent data to human tasks without incorporating any knowledge from the human domain. In contrast. Fig \ref{['fig:RH2H']} demonstrates the use of a model that integrates knowledge learned from both rodent and human data for improved performance on human tasks. In the figures, the black arrows represent training paths, while the red arrows indicate testing paths.
• Figure 4: Our pipeline. This figure introduces the network structure of the dynamics head. Specifically, it contains a residual U-Net backbone, a class-aware controller, and a dynamic segmentation head.
• Figure 5: Qualitative Results of different methods on glomerular segmentation. This figure displays the qualitative outcomes of various segmentation methods for all classes of glomeruli. The first column features the original, unannotated images, while the second column shows the manual segmentation results.