GLAM: Glomeruli Segmentation for Human Pathological Lesions using Adapted Mouse Model

TL;DR

GLAM addresses the problem of cross-species segmentation of pathological glomeruli by adapting a mouse-derived model to human kidney lesions using zero-shot and hybrid transfer learning. It combines a residual U-Net backbone with a dynamic, class-aware head that uses a class vector $T_k$ to generate lesion-specific kernels, enabling fine-grained, partially labeled segmentation with the final mask $P = ((((M * \omega_1) * \omega_2) * \omega_3))$ where $\omega = \phi(\text{GAP}(F) || T_k; \Theta_\phi)$. Across mouse and human datasets, zero-shot (M2H) and hybrid (M&H2H) training improve performance over baselines, demonstrating the viability of leveraging mouse data for human pathology with reduced annotation burden. This work offers a practical, data-efficient path toward translational renal pathology tools and introduces a framework that can be extended to other cross-species medical imaging tasks.

Abstract

Moving from animal models to human applications in preclinical research encompasses a broad spectrum of disciplines in medical science. A fundamental element in the development of new drugs, treatments, diagnostic methods, and in deepening our understanding of disease processes is the accurate measurement of kidney tissues. Past studies have demonstrated the viability of translating glomeruli segmentation techniques from mouse models to human applications. Yet, these investigations tend to neglect the complexities involved in segmenting pathological glomeruli affected by different lesions. Such lesions present a wider range of morphological variations compared to healthy glomerular tissue, which are arguably more valuable than normal glomeruli in clinical practice. Furthermore, data on lesions from animal models can be more readily scaled up from disease models and whole kidney biopsies. This brings up a question: ``\textit{Can a pathological segmentation model trained on mouse models be effectively applied to human patients?}" To answer this question, we introduced GLAM, a deep learning study for fine-grained segmentation of human kidney lesions using a mouse model, addressing mouse-to-human transfer learning, by evaluating different learning strategies for segmenting human pathological lesions using zero-shot transfer learning and hybrid learning by leveraging mouse samples. From the results, the hybrid learning model achieved superior performance.

Figures (5)

• Figure 1: Overview of the learning framework. This figure provides an overview of transfer learning for glomerular segmentation from mouse to human, highlighting the strenth of using mouse samples compared to human samples.
• Figure 2: GLAM pipeline. This figure introduces the network structure of the proposed dynamics head. Specifically, it contains a residual U-Net backbone, a class-aware controller, and a dynamic segmentation head.
• Figure 3: Lesion Classes. This figure shows fine-grained classes of mouse and human glomerular images that are stained with PAS. Glomeruli with lesions: global sclerosis(GS), Hyalinosis(HN), mesangial lysis(ML), microaneurysm(MA), nodular sclerosis(NS), and segmental sclerosis(SS).
• Figure 4: Qualitative Results of different methods on human glomerular lesions. This figure displays the qualitative outcomes of various segmentation methods for six classes of glomerular lesions. The first column features the original, unannotated images, while the second column shows the manual segmentation results. Subsequent columns belong to three section: "Mouse-to-human","Human-to-human" and "Mouse&Human-to-human", where VM refers to model selection utilizing validation results from mouse data, whereas VH indicates model selection based on validation results from human data. The bold mark indicates the best performance.
• Figure 5: Qualitative Results of different methods on six mice lesions for diabetic nephropathy. This figure displays the qualitative outcomes of various segmentation methods for six classes of glomerular lesions. The first column features the original, unannotated images, while the second column shows the manual segmentation results. Subsequent columns are dedicated to the results from baseline models and the GLAM Mouse-to-Mouse approach, respectively.