Ordinal Label-Distribution Learning with Constrained Asymmetric Priors for Imbalanced Retinal Grading

2025-09-30

AI Summary

Diabetic retinopathy (DR) grading is inherently ordinal and exhibits a severe long-tailed distribution, where late-stage samples are scarce, highly heterogeneous, and clinically critical—yet conventional methods employ isotropic Gaussian priors and symmetric losses, failing to capture the asymmetric clinical costs of misclassification. This work proposes the first framework integrating asymmetric latent priors with ordinal learning: (1) a constrained asymmetric prior explicitly models clinical risk gradients; (2) margin-aware orthogonal compactness loss and direction-aware ordinal loss jointly construct soft labels and enforce stronger penalties for under-grading; and (3) a Wasserstein autoencoder, lightweight decentralized prediction heads, and adaptive multi-task optimization enhance robustness and efficiency. Evaluated on public DR benchmarks, the model achieves significant improvements in quadratic weighted kappa, overall accuracy, and macro-F1. t-SNE visualizations confirm tighter, more ordered latent-space clustering with reduced inter-class overlap.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Diabetic retinopathy grading is inherently ordinal and long-tailed, with minority stages being scarce, heterogeneous, and clinically critical to detect accurately. Conventional methods often rely on isotropic Gaussian priors and symmetric loss functions, misaligning latent representations with the task's asymmetric nature. We propose the Constrained Asymmetric Prior Wasserstein Autoencoder (CAP-WAE), a novel framework that addresses these challenges through three key innovations. Our approach employs a Wasserstein Autoencoder (WAE) that aligns its aggregate posterior with a asymmetric prior, preserving the heavy-tailed and skewed structure of minority classes. The latent space is further structured by a Margin-Aware Orthogonality and Compactness (MAOC) loss to ensure grade-ordered separability. At the supervision level, we introduce a direction-aware ordinal loss, where a lightweight head predicts asymmetric dispersions to generate soft labels that reflect clinical priorities by penalizing under-grading more severely. Stabilized by an adaptive multi-task weighting scheme, our end-to-end model requires minimal tuning. Across public DR benchmarks, CAP-WAE consistently achieves state-of-the-art Quadratic Weighted Kappa, accuracy, and macro-F1, surpassing both ordinal classification and latent generative baselines. t-SNE visualizations further reveal that our method reshapes the latent manifold into compact, grade-ordered clusters with reduced overlap.

Problem

• Addresses ordinal imbalanced grading in diabetic retinopathy
• Corrects misaligned latent representations with asymmetric priors
• Ensures grade-ordered separability through structured loss functions

Innovation

• WAE with asymmetric prior preserves minority class structure
• MAOC loss ensures grade-ordered latent space separability
• Direction-aware ordinal loss penalizes under-grading more severely