MAISI-v2: Accelerated 3D High-Resolution Medical Image Synthesis with Rectified Flow and Region-specific Contrastive Loss

2025-08-07

AI Summary

To address three key challenges in medical image synthesis—poor generalization, slow inference, and weak conditional alignment—this paper proposes the first rectified flow-accelerated generative framework tailored for 3D high-resolution medical images. Methodologically, it integrates latent diffusion models with rectified flows to enable rapid sampling; introduces a region-specific contrastive loss to enhance anatomical fidelity in regions of interest; and employs a 3D convolutional backbone to improve spatial modeling capacity. Experiments demonstrate strong cross-modality and cross-anatomy generalization, state-of-the-art image quality, and a 33× speedup in inference over conventional latent diffusion models. Moreover, generated images significantly boost downstream segmentation performance, validating the framework’s practical utility for medical data augmentation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Medical image synthesis is an important topic for both clinical and research applications. Recently, diffusion models have become a leading approach in this area. Despite their strengths, many existing methods struggle with (1) limited generalizability that only work for specific body regions or voxel spacings, (2) slow inference, which is a common issue for diffusion models, and (3) weak alignment with input conditions, which is a critical issue for medical imaging. MAISI, a previously proposed framework, addresses generalizability issues but still suffers from slow inference and limited condition consistency. In this work, we present MAISI-v2, the first accelerated 3D medical image synthesis framework that integrates rectified flow to enable fast and high quality generation. To further enhance condition fidelity, we introduce a novel region-specific contrastive loss to enhance the sensitivity to region of interest. Our experiments show that MAISI-v2 can achieve SOTA image quality with $33 imes$ acceleration for latent diffusion model. We also conducted a downstream segmentation experiment to show that the synthetic images can be used for data augmentation. We release our code, training details, model weights, and a GUI demo to facilitate reproducibility and promote further development within the community.

Problem

• Limited generalizability across body regions and voxel spacings
• Slow inference speed in diffusion models
• Weak alignment with input conditions in medical imaging

Innovation

• Rectified flow for fast 3D image synthesis
• Region-specific contrastive loss for condition fidelity
• 33x acceleration for latent diffusion model