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

Francisco Caetano; Lemar Abdi; Christiaan Viviers; Amaan Valiuddin; Fons van der Sommen

doi:10.1007/978-3-032-05472-2_4

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

Francisco Caetano (Corresponding author)
, Lemar Abdi
, Christiaan Viviers
, Amaan Valiuddin
, Fons van der Sommen

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

1 Citation (Scopus)

5 Downloads (Pure)

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.

Original language	English
Title of host publication	Deep Generative Models
Subtitle of host publication	5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings
Editors	Anirban Mukhopadhyay, Ilkay Oksuz, Sandy Engelhardt, Dorit Mehrof, Yixuan Yuan
Place of Publication	Cham
Publisher	Springer
Pages	35-45
Number of pages	11
ISBN (Electronic)	978-3-032-05472-2
ISBN (Print)	978-3-032-05471-5
DOIs	https://doi.org/10.1007/978-3-032-05472-2_4
Publication status	Published - 25 Sept 2025
Event	5th Workshop on Deep Generative Models for Medical Image Computing and Computer Assisted Intervention, DGM4MICCAI 2025, held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025 - Daejeon, Korea, Republic of Duration: 23 Sept 2025 → 23 Sept 2025

Publication series

Name	Lecture Notes in Computer Science (LNCS)
Volume	16128
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	5th Workshop on Deep Generative Models for Medical Image Computing and Computer Assisted Intervention, DGM4MICCAI 2025, held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025
Country/Territory	Korea, Republic of
City	Daejeon
Period	23/09/25 → 23/09/25

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

Access to Document

10.1007/978-3-032-05472-2_4

pdfFinal published version, 2.53 MBLicence: TAVERNE

Cite this

Caetano, F., Abdi, L., Viviers, C., Valiuddin, A., & van der Sommen, F. (2025). MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching. In A. Mukhopadhyay, I. Oksuz, S. Engelhardt, D. Mehrof, & Y. Yuan (Eds.), Deep Generative Models: 5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings (pp. 35-45). (Lecture Notes in Computer Science (LNCS); Vol. 16128). Springer. https://doi.org/10.1007/978-3-032-05472-2_4

Caetano, Francisco ; Abdi, Lemar ; Viviers, Christiaan et al. / MedSymmFlow : Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching. Deep Generative Models: 5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings. editor / Anirban Mukhopadhyay ; Ilkay Oksuz ; Sandy Engelhardt ; Dorit Mehrof ; Yixuan Yuan. Cham : Springer, 2025. pp. 35-45 (Lecture Notes in Computer Science (LNCS)).

@inproceedings{046d788ba1e94115ad62d0a35173c8b2,

title = "MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching",

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.",

author = "Francisco Caetano and Lemar Abdi and Christiaan Viviers and Amaan Valiuddin and \{van der Sommen\}, Fons",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 5th Workshop on Deep Generative Models for Medical Image Computing and Computer Assisted Intervention, DGM4MICCAI 2025, held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025 ; Conference date: 23-09-2025 Through 23-09-2025",

year = "2025",

month = sep,

day = "25",

doi = "10.1007/978-3-032-05472-2\_4",

language = "English",

isbn = "978-3-032-05471-5",

series = "Lecture Notes in Computer Science (LNCS)",

publisher = "Springer",

pages = "35--45",

editor = "Anirban Mukhopadhyay and Ilkay Oksuz and Sandy Engelhardt and Dorit Mehrof and Yixuan Yuan",

booktitle = "Deep Generative Models",

address = "Germany",

}

Caetano, F , Abdi, L , Viviers, C , Valiuddin, A & van der Sommen, F 2025, MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching. in A Mukhopadhyay, I Oksuz, S Engelhardt, D Mehrof & Y Yuan (eds), Deep Generative Models: 5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings. Lecture Notes in Computer Science (LNCS), vol. 16128, Springer, Cham, pp. 35-45, 5th Workshop on Deep Generative Models for Medical Image Computing and Computer Assisted Intervention, DGM4MICCAI 2025, held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025, Daejeon, Korea, Republic of, 23/09/25. https://doi.org/10.1007/978-3-032-05472-2_4

MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching. / Caetano, Francisco (Corresponding author); Abdi, Lemar ; Viviers, Christiaan et al.
Deep Generative Models: 5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings. ed. / Anirban Mukhopadhyay; Ilkay Oksuz; Sandy Engelhardt; Dorit Mehrof; Yixuan Yuan. Cham: Springer, 2025. p. 35-45 (Lecture Notes in Computer Science (LNCS); Vol. 16128).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - MedSymmFlow

T2 - 5th Workshop on Deep Generative Models for Medical Image Computing and Computer Assisted Intervention, DGM4MICCAI 2025, held in conjunction with the 28th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2025

AU - Caetano, Francisco

AU - Abdi, Lemar

AU - Viviers, Christiaan

AU - Valiuddin, Amaan

AU - van der Sommen, Fons

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

PY - 2025/9/25

Y1 - 2025/9/25

N2 - 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.

AB - 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.

UR - https://www.scopus.com/pages/publications/105018576368

U2 - 10.1007/978-3-032-05472-2_4

DO - 10.1007/978-3-032-05472-2_4

M3 - Conference contribution

AN - SCOPUS:105018576368

SN - 978-3-032-05471-5

T3 - Lecture Notes in Computer Science (LNCS)

SP - 35

EP - 45

BT - Deep Generative Models

A2 - Mukhopadhyay, Anirban

A2 - Oksuz, Ilkay

A2 - Engelhardt, Sandy

A2 - Mehrof, Dorit

A2 - Yuan, Yixuan

PB - Springer

CY - Cham

Y2 - 23 September 2025 through 23 September 2025

ER -

Caetano F , Abdi L , Viviers C , Valiuddin A , van der Sommen F. MedSymmFlow: Bridging Generative Modeling and Classification in Medical Imaging Through Symmetrical Flow Matching. In Mukhopadhyay A, Oksuz I, Engelhardt S, Mehrof D, Yuan Y, editors, Deep Generative Models: 5th MICCAI Workshop, DGM4MICCAI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025, Proceedings. Cham: Springer. 2025. p. 35-45. (Lecture Notes in Computer Science (LNCS)). doi: 10.1007/978-3-032-05472-2_4