MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching
- ,
- ,
- ,
- ,
Open access
Information
Output type
Host publication Subtitle
5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, ProceedingsOriginal language
EnglishPages from-to (Number of pages)
Pages 35-45 (11 pages)Publication milestones
- Published - 25/09/2025
Publication status
Place of publication
ChamPublisher
Springer, Netherlands, United States, United Kingdom, Germany, Italy, France, Austria, India, Switzerland, China, JapanPublication series
- Publication series name: Lecture Notes in Computer Science (LNCS)
ISSN (Print): 0302-9743
ISSN (Electronic): 1611-3349
Volume: 16128
ISBN (Print)
978-3-032-05471-5ISBN (Electronic)
978-3-032-05472-2Host publication title
Deep Generative ModelsHost publication editors
- Anirban Mukhopadhyay
- Ilkay Oksuz
- Sandy Engelhardt
- Dorit Mehrof
- Yixuan Yuan
Related Event
Title
Event type
ConferenceDate
23/09/2025 - 23/09/2025Location
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. The code is available at github.com/caetas/MedSymmFlow.
