Utilizing Weak-to-Strong Consistency for Semi-Supervised Glomeruli Segmentation

Irina Zhang,Jim Denholm,Azam Hamidinekoo,Oskar Ålund,Christopher Bagnall,Joana Palés Huix,Michal Sulikowski,Ortensia Vito,Arthur Lewis,Robert Unwin,Magnus Soderberg,Nikolay Burlutskiy,Talha Qaiser

2024-05-30

Abstract:Accurate segmentation of glomerulus instances attains high clinical significance in the automated analysis of renal biopsies to aid in diagnosing and monitoring kidney disease. Analyzing real-world histopathology images often encompasses inter-observer variability and requires a labor-intensive process of data annotation. Therefore, conventional supervised learning approaches generally achieve sub-optimal performance when applied to external datasets. Considering these challenges, we present a semi-supervised learning approach for glomeruli segmentation based on the weak-to-strong consistency framework validated on multiple real-world datasets. Our experimental results on 3 independent datasets indicate superior performance of our approach as compared with existing supervised baseline models such as U-Net and SegFormer.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

This paper aims to address the issue of accurate segmentation of glomeruli in medical image analysis, particularly in the application of automated diagnosis and monitoring of chronic kidney disease (CKD). Current methods often perform poorly when dealing with real-world pathological images due to the differences and high variability in data annotation. Therefore, the paper proposes a semi-supervised learning approach based on a weak-to-strong consistency framework for glomerular segmentation to improve the generalization performance on multi-center, large-scale real-world datasets. The researchers compared this method with existing supervised learning models such as U-Net and SegFormer, and validated its superiority on three independent datasets. Diverse data sources including NURTuRE (the UK's first renal biobank), KPMP (Kidney Precision Medicine Project in the US), and HuBMAP (a publicly available dataset on renal and vascular anatomy) were used. By combining labeled and unlabeled data, their semi-supervised learning algorithm outperformed the supervised baseline model on external datasets. The core of the paper lies in utilizing high-confidence predictions from weakly augmented images as pseudo-labels, training the model to make consistent predictions on strongly augmented images, thus achieving weak-to-strong consistency regularization. This approach enhances the model's ability to capture detailed information across different augmented images. Experimental results demonstrate that the proposed semi-supervised learning method performs well on evaluation metrics such as Dice coefficient and recall on all three validation datasets, while maintaining comparable accuracy, confirming its potential in handling multi-center real-world pathological data. This method has the potential to be extended to identify different types of glomerular diseases or morphological changes, assisting in the pathological examination of kidney biopsies.

Utilizing Weak-to-Strong Consistency for Semi-Supervised Glomeruli Segmentation

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

A Deep Learning-Based Approach for Glomeruli Instance Segmentation from Multistained Renal Biopsy Pathologic Images

Unsupervised learning for labeling global glomerulosclerosis

Weakly supervised semantic segmentation of histological tissue via attention accumulation and pixel-level contrast learning

Inherent Consistent Learning for Accurate Semi-supervised Medical Image Segmentation

Glo-In-One: Holistic Glomerular Detection, Segmentation, and Lesion Characterization with Large-scale Web Image Mining

Leveraging Uncertainty for Deep Interpretable Classification and Weakly-Supervised Segmentation of Histology Images

Uncertainty-Guided Mutual Consistency Learning for Semi-Supervised Medical Image Segmentation

Weakly Supervised Histopathology Image Segmentation with Sparse Point Annotations

Segmentation and Thickness Calculation of Glomerular Basement Membrane Using RADS-Net in Glomerular Microscopic Images

Consistency learning with dynamic weighting and class-agnostic regularization for semi-supervised medical image segmentation

Semi-Supervised Neuron Segmentation via Reinforced Consistency Learning

Predicting clinical endpoints and visual changes with quality-weighted tissue-based renal histological features

Dual-consistency guidance semi-supervised medical image segmentation with low-level detail feature augmentation

Learning Multiscale Consistency for Self-supervised Electron Microscopy Instance Segmentation

Automatic fine-grained glomerular lesion recognition in kidney pathology

Semi-supervised segmentation of cardiac chambers from LGE-CMR using feature consistency awareness

Dual Consistency Enabled Weakly and Semi-Supervised Optic Disc and Cup Segmentation With Dual Adaptive Graph Convolutional Networks

Global-local consistent semi-supervised segmentation of histopathological image with different perturbations

MSMTSeg: Multi-Stained Multi-Tissue Segmentation of Kidney Histology Images via Generative Self-Supervised Meta-Learning Framework