GenMed: A Pairwise Generative Reformulation of Medical Diagnostic Tasks

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Hantao, Guo, Weidong, Liu, Yuhe, Yang, Jiancheng, Bhagavan, Sathvik, Shi, Danli, Xu, Mingda, Fua, Pascal
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914554183680000
author Zhang, Hantao
Guo, Weidong
Liu, Yuhe
Yang, Jiancheng
Bhagavan, Sathvik
Shi, Danli
Xu, Mingda
Fua, Pascal
author_facet Zhang, Hantao
Guo, Weidong
Liu, Yuhe
Yang, Jiancheng
Bhagavan, Sathvik
Shi, Danli
Xu, Mingda
Fua, Pascal
contents Data-driven medical AI is traditionally formulated as a discriminative mapping from input $X$ to output $Y$ via a learned function $f$, which does not generalize well across heterogeneous data and modalities encountered in real-world clinical settings. In this work, we propose a fundamentally different, generative paradigm. We model the joint distribution $P(X,Y)$ using diffusion models and reframe inference as a test-time output optimization problem. By guiding the generative process to match observed inputs, our framework enables flexible, gradient-based conditioning at inference time without architectural changes or retraining, effectively supporting arbitrary and previously unseen combinations of observations. Extensive experiments demonstrate strong performance across standard and cross-modality medical image segmentation, few-shot segmentation with only 2 or 4 training samples, degraded-input segmentation, shape completion from sparse and partial observations, and zero-shot application to demonstrate generality. To support these evaluations, we curated and released a large-scale text-shape dataset derived from MedShapeNet. Our results highlight the versatility of generative joint modeling as a foundation for reusable, task-agnostic medical AI systems.
format Preprint
id arxiv_https___arxiv_org_abs_2605_10645
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle GenMed: A Pairwise Generative Reformulation of Medical Diagnostic Tasks
Zhang, Hantao
Guo, Weidong
Liu, Yuhe
Yang, Jiancheng
Bhagavan, Sathvik
Shi, Danli
Xu, Mingda
Fua, Pascal
Computer Vision and Pattern Recognition
Data-driven medical AI is traditionally formulated as a discriminative mapping from input $X$ to output $Y$ via a learned function $f$, which does not generalize well across heterogeneous data and modalities encountered in real-world clinical settings. In this work, we propose a fundamentally different, generative paradigm. We model the joint distribution $P(X,Y)$ using diffusion models and reframe inference as a test-time output optimization problem. By guiding the generative process to match observed inputs, our framework enables flexible, gradient-based conditioning at inference time without architectural changes or retraining, effectively supporting arbitrary and previously unseen combinations of observations. Extensive experiments demonstrate strong performance across standard and cross-modality medical image segmentation, few-shot segmentation with only 2 or 4 training samples, degraded-input segmentation, shape completion from sparse and partial observations, and zero-shot application to demonstrate generality. To support these evaluations, we curated and released a large-scale text-shape dataset derived from MedShapeNet. Our results highlight the versatility of generative joint modeling as a foundation for reusable, task-agnostic medical AI systems.
title GenMed: A Pairwise Generative Reformulation of Medical Diagnostic Tasks
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.10645