Prototype Instance-semantic Disentanglement with Low-rank Regularized Subspace Clustering for WSIs Explainable Recognition

Chentao Li; Pan Huang

cs.CVFeb 16, 2026

Prototype Instance-semantic Disentanglement with Low-rank Regularized Subspace Clustering for WSIs Explainable Recognition

Authors: Chentao Li, Pan Huang

Abstract

The tumor region plays a key role in pathological diagnosis. Tumor tissues are highly similar to precancerous lesions and non tumor instances often greatly exceed tumor instances in whole slide images (WSIs). These issues cause instance-semantic entanglement in multi-instance learning frameworks, degrading both model representation capability and interpretability. To address this, we propose an end-to-end prototype instance semantic disentanglement framework with low-rank regularized subspace clustering, PID-LRSC, in two aspects. First, we use secondary instance subspace learning to construct low-rank regularized subspace clustering (LRSC), addressing instance entanglement caused by an excessive proportion of non tumor instances. Second, we employ enhanced contrastive learning to design prototype instance semantic disentanglement (PID), resolving semantic entanglement caused by the high similarity between tumor and precancerous tissues. We conduct extensive experiments on multicentre pathology datasets, implying that PID-LRSC outperforms other SOTA methods. Overall, PID-LRSC provides clearer instance semantics during decision-making and significantly enhances the reliability of auxiliary diagnostic outcomes.

Figures & tables

Figure 1: The motivation of proposed PID-LRSC. Left : Instance entanglement caused by excessive proportion of non-tumor over tumor. Right : Semantic entanglement caused by high similarity between precancerous lesions and tumor.

Figure 2: Overview of PID-LRSC . Cropped instances extracted by encoder are clustered into three subspaces by L ow R ank S subspace C lustering method in two phases. They are then identified into tumor, non-tumor and background by P rototype I nstance-semantic D isentanglement. End-to-end optimization is performed for disentangled representation learning with refined instance features.

Models	AMU-LSCC		AMU-CSCC		DHMC-KIDNEY		DHMC-LUNG
Models	ACC	AUC	ACC	AUC	ACC	AUC	ACC	AUC
ABMIL [ 7 ]	0.6739	0.7888	0.8302	0.9547	0.7675	0.9421	0.7333	0.8123
CLAM-SB [ 15 ]	0.6957	0.8367	0.8302	0.9581	0.7807	0.9339	0.7556	0.7783
CLAM-MB [ 15 ]	0.6812	0.8031	0.8774	0.9336	0.7939	0.9372	0.7333	0.7684
TransMIL [ 16 ]	0.8261	0.9021	0.8868	0.9521	0.8465	0.9603	0.7333	0.8084
DTFD-MIL [ 26 ]	0.5435	0.7245	0.8113	0.9036	0.7061	0.9159	0.7333	0.8084

Table 1: Performance comparison of cancer diagnosis, and sub-typing on AMU-CSCC and AMU-LSCC with Swin-T pretrained features. DHMC-KIDNEY and DHMC-LUNG are for generalization.

Panel of Table 2 — Table 2: Ablation experiments for different components of PID-LRSC on AMU-LSCC.

Metrics in PID (AMU-LSCC)	ACC	AUC
MMD	0.8551	0.9441
CFD (ours)	0.9130	0.9590

Table 3: Ablation experiments on different distribution metrics in PID.

Figure 3: Visualization results on multicentre datasets. The first column is original WSI with pathologist-annotated boundaries. The second is the attention contribution scores corresponding to its category. The last column is the disentangling results.

Figure 4: ANOVA plots on four datasets where $\eta_{*}$ denote the value of effect size. Statistically, $\eta_{*}>0.4$ implies the model having large effects within categories.

Whole-Slide Image Analysis

Whole-slide image analysis uses AI to interpret gigapixel pathology slides for diagnosis, prognosis, tissue mapping, and report generation.

See all 214 papers

Hierarchical Prototype-based Domain Priors for Multiple Instance Learning in Multimodal Histopathology Analysis
Apr 27, 2026Xuemei Qiu, Dawei Fan, Yebin Huang +2Cross-Modal Alignment Prototype Learning
College of Computer and Information Science, Fujian Agriculture and Forestry University, Fuzhou, 350002, China · Fujian Agriculture and Forestry University, Fuzhou, 350002, China · Department of Pathology, Clinical Oncology School of Fujian Medical University, and Fujian Cancer Hospital, Fuzhou, 350014, China +2
MacNet: An End-to-End Manifold-Constrained Adaptive Clustering Network for Interpretable Whole Slide Image Classification
Feb 16, 2026Mingrui Ma, Chentao Li, Pan Huang +1Histopathology Image Classification Explainable Medical Image Analysis
School of Computer Science and Technology, Xinjiang University, Urumqi 830046, China · Affiliated Tumor Hospital of Xinjiang Medical University, Xinjiang 830011, China · Fu Foundation School of Engineering and Applied Science, Columbia University, New York, NY 10027, USA +1
LRMIL: Efficient Low-Resolution Multiple Instance Learning via High-Resolution Knowledge Distillation for Whole Slide Image Classification
Jun 5, 2026Yonghan Shin, Won-Ki JeongHistopathology Image Classification Knowledge Distillation
Department of Computer Science and Engineering, Korea University, Seoul, Korea

Apr 27, 2026J/K move · Enter open · S save

Hierarchical Prototype-based Domain Priors for Multiple Instance Learning in Multimodal Histopathology Analysis

Xuemei Qiu, Dawei Fan, Yebin Huang, Yanping Chen+1

College of Computer and Information Science, Fujian Agriculture and Forestry University, Fuzhou, 350002, China · Fujian Agriculture and Forestry University, Fuzhou, 350002, China · Department of Pathology, Clinical Oncology School of Fujian Medical University, and Fujian Cancer Hospital, Fuzhou, 350014, China+2

Cross-Modal Alignment Prototype Learning Histopathology Image Classification Multimodal Fusion Multiple Instance Learning Whole-Slide Image Analysis

Open paper AskPDF arXiv

Digital pathology has fundamentally altered diagnostic workflows by enabling the computational analysis of gigapixel Whole Slide Images (WSIs), yet effectively deciphering their complex tumor microenvironments remains a formidable challenge. Existing Multiple Instance Learning (MIL) frameworks typically treat Whole Slide Images as unstructured bags of patches, discarding critical morphological semantics and spatial geometry. This lack of inductive bias often leads to overfitting on background noise and fails to align visual features with high-level diagnostic knowledge. To overcome these limitations, we propose the Hierarchical Prototype-based Domain Priors (HPDP) framework, a unified multimodal approach for joint histopathology diagnosis and prognosis. HPDP mitigates the data-driven "black box" issue by introducing a Morphologically Anchored Prototype System (MAPS), which anchors learning to interpretable morphological clusters, and a Sinusoidal Positional Encoder (SPE) to explicitly model tissue architecture. Furthermore, we bridge the semantic gap via a Hierarchical Cross-Modal Alignment (HCMA) module, using Large Language Model (LLM)-generated descriptions to contextually refine visual representations. Extensive experiments across seven cancer cohorts demonstrate that HPDP consistently achieves state-of-the-art performance with superior robustness and interpretability.

Abstract

Whole-Slide Image Analysis

Explore similar work

Hierarchical Prototype-based Domain Priors for Multiple Instance Learning in Multimodal Histopathology Analysis

MacNet: An End-to-End Manifold-Constrained Adaptive Clustering Network for Interpretable Whole Slide Image Classification

LRMIL: Efficient Low-Resolution Multiple Instance Learning via High-Resolution Knowledge Distillation for Whole Slide Image Classification