Optimizing 3D Diffusion Models for Medical Imaging via Multi-Scale Reward Learning
This work provides an incremental improvement for medical imaging researchers and practitioners by generating more clinically relevant 3D medical images.
This paper addresses the challenge of aligning 3D diffusion model outputs with clinical relevance in medical imaging by fine-tuning a pretrained model using Reinforcement Learning (RL) with a multi-scale reward system. The method significantly improves Fréchet Inception Distance (FID) and enhances the utility of synthetic data for downstream tumor and disease classification tasks compared to non-optimized baselines.
Diffusion models have emerged as powerful tools for 3D medical image generation, yet bridging the gap between standard training objectives and clinical relevance remains a challenge. This paper presents a method to enhance 3D diffusion models using Reinforcement Learning (RL) with multi-scale feedback. We first pretrain a 3D diffusion model on MRI volumes to establish a robust generative prior. Subsequently, we fine-tune the model using Proximal Policy Optimization (PPO), guided by a novel reward system that integrates both 2D slice-wise assessments and 3D volumetric analysis. This combination allows the model to simultaneously optimize for local texture details and global structural coherence. We validate our framework on the BraTS 2019 and OASIS-1 datasets. Our results indicate that incorporating RL feedback effectively steers the generation process toward higher quality distributions. Quantitative analysis reveals significant improvements in Fréchet Inception Distance (FID) and, crucially, the synthetic data demonstrates enhanced utility in downstream tumor and disease classification tasks compared to non-optimized baselines.