MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging through Symmetrical Flow Matching

2025-07-25

AI Summary

To address insufficient reliability in both predictive accuracy and uncertainty quantification for medical image classification in high-stakes clinical settings, this paper proposes a generative-discriminative hybrid framework that jointly performs classification, image generation, and uncertainty estimation. Our method introduces three key innovations: (1) symmetric flow matching for expressive latent-space modeling; (2) a semantic mask conditioning mechanism to align generative processes with clinically relevant diagnostic semantics; and (3) high-fidelity uncertainty estimates derived naturally from generative sampling. Evaluated end-to-end on four MedMNIST subsets, our approach achieves classification accuracy and AUC scores on par with or superior to state-of-the-art baselines. Crucially, selective prediction experiments demonstrate significantly improved uncertainty calibration compared to existing methods. This work establishes a novel paradigm for trustworthy AI-assisted diagnosis by unifying discriminative decision-making with semantically grounded generative modeling and principled uncertainty quantification.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Reliable medical image classification requires accurate predictions and well-calibrated uncertainty estimates, especially in high-stakes clinical settings. This work presents MedSymmFlow, a generative-discriminative hybrid model built on Symmetrical Flow Matching, designed to unify classification, generation, and uncertainty quantification in medical imaging. MedSymmFlow leverages a latent-space formulation that scales to high-resolution inputs and introduces a semantic mask conditioning mechanism to enhance diagnostic relevance. Unlike standard discriminative models, it naturally estimates uncertainty through its generative sampling process. The model is evaluated on four MedMNIST datasets, covering a range of modalities and pathologies. The results show that MedSymmFlow matches or exceeds the performance of established baselines in classification accuracy and AUC, while also delivering reliable uncertainty estimates validated by performance improvements under selective prediction.

Problem

• Unify classification, generation, and uncertainty quantification in medical imaging
• Enhance diagnostic relevance with semantic mask conditioning mechanism
• Provide reliable uncertainty estimates in high-stakes clinical settings

Innovation

• Symmetrical Flow Matching hybrid model
• Latent-space scales to high-resolution inputs
• Semantic mask enhances diagnostic relevance