Optimizing 3D Diffusion Models for Medical Imaging via Multi-Scale Reward Learning

Generative deep learning methods such as Variational Autoencoders, Generative Adversarial Networks or Diffusion Models have achieved remarkable results in image synthesis. In particular, Diffusion Models have demonstrated superior image fidelity and sample diversity compared to other approaches on standard benchmarks, while also offering more flexible training dynamics as evidenced both in natural image domains and in medical applications. In medical imaging including Computed Tomography, Magnetic Resonance Imaging or Positron Emission Tomography, such models have various important use-cases including data augmentation, domain transfer, image reconstruction, inpainting, image augmentation and artifact correction. Further, generated images can be conditioned on patient characteristics, medical reports, segmentation masks or other images. Specific challenges arise due to the 3D volumetric nature of medical images, which translate to substantial demands in terms of memory, compute, and training data. This article provides a comparative overview of 3D Diffusion Models which have been developed for medical images, with a focus on current performance, limitations, and an outlook for future work.

We introduce TCAM-Diff, a novel 3D medical image generation model that reduces the memory requirements to encode and generate high-resolution 3D data. This model utilizes a decoder-only autoencoder method to learn triplane representation from dense volume and leverages generalization operations to prevent overfitting. Subsequently, it uses a triplane-aware cross-attention diffusion model to learn and integrate these features effectively. Furthermore, the features generated by the diffusion model can be rapidly transformed into 3D volumes using a pre-trained decoder module. Our experiments on three different scales of medical datasets, BrainTumour 128 x 128 x 128, Pancreas 256 x 256 x 256, and Colon 512 x 512 x 512, demonstrate outstanding results. We utilized MSE and SSIM to assess reconstruction quality and leveraged the Wasserstein Generative Adversarial Network (W-GAN) critic to assess generative quality. Comparisons with existing approaches show that our method gives better reconstruction and generation results than other encoder-decoder methods with similar-sized latent spaces.

Recent image-to-3D generation models built on flow-matching diffusion Transformers (DiT) can produce high-fidelity meshes, yet their post-training strategy remains largely unexplored. There exist several critical bottlenecks in reinforcement learning: the inherent difficulty of defining comprehensive rewards for 3D geometric quality, and the gradient interference that arises when jointly optimizing heterogeneous objectives. Inspired by the practicability of on-policy distillation (OPD) in large language models and image generation, we propose Flow3D-OPD, a two-stage post-training framework that introduces multi-teacher distillation into 3D geometry generation. In the first stage, we utilize the semi-policy to enhance the foundational capability of the pretrained model and then design an agentic verifier for 3D geometric quality evaluation. Based on the verifier, we could cultivate domain-specialized teacher models via direct preference optimization (DPO). In the second stage, we consolidate heterogeneous expertise into a unified student model through on-policy distillation with hard task-routing sampling and gradient accumulation, which could mitigate the gradient interference in joint optimization. Without relying on elaborate modifications, our straightforward yet effective design achieves consistent improvements across all geometric quality dimensions and surpasses all teacher models in the average metric. Extensive experiments demonstrate that our approach provides an effective paradigm for reinforcement learning in 3D generation.