Model Agnostic Preference Optimization for Medical Image Segmentation

2025-12-16

AI Summary

Medical image segmentation faces key bottlenecks in preference optimization: strong model dependency, limited prediction diversity, and reliance on high-quality annotations. To address these, this paper proposes a general unsupervised preference optimization framework. Methodologically, it introduces a dropout-driven stochastic segmentation hypothesis sampling mechanism—enabling model-agnostic and dimension-agnostic preference gradient construction—and integrates relative preference signal modeling with unsupervised gradient backpropagation, ensuring compatibility with arbitrary 2D/3D CNNs and Transformers. Evaluated across multi-center datasets, the framework significantly improves boundary accuracy, mitigates overfitting, and enhances training stability. It consistently outperforms standard supervised training across all metrics. By eliminating the need for explicit preference labels or architectural constraints, the approach establishes a scalable, annotation-efficient paradigm for medical image segmentation.

Technology Category

• Computer Vision: Segmentation
• Reasoning under Uncertainty: Stochastic Optimization
• Search and Optimization: Sampling/Simulation-based Search

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: Foundation models and LLMs for Web-related graphs

Abstract

Preference optimization offers a scalable supervision paradigm based on relative preference signals, yet prior attempts in medical image segmentation remain model-specific and rely on low-diversity prediction sampling. In this paper, we propose MAPO (Model-Agnostic Preference Optimization), a training framework that utilizes Dropout-driven stochastic segmentation hypotheses to construct preference-consistent gradients without direct ground-truth supervision. MAPO is fully architecture- and dimensionality-agnostic, supporting 2D/3D CNN and Transformer-based segmentation pipelines. Comprehensive evaluations across diverse medical datasets reveal that MAPO consistently enhances boundary adherence, reduces overfitting, and yields more stable optimization dynamics compared to conventional supervised training.

Problem

• Optimizes segmentation without ground-truth supervision
• Enhances boundary adherence and reduces overfitting
• Works across 2D/3D CNN and Transformer architectures

Innovation

• Model-agnostic training framework using dropout-driven stochastic hypotheses
• Constructs preference-consistent gradients without ground-truth supervision
• Supports 2D/3D CNN and Transformer architectures