Structure-Accurate Medical Image Translation based on Dynamic Frequency Balance and Knowledge Guidance

2025-04-13

AI Summary

Existing medical image translation methods suffer from anatomical distortion due to high-frequency overfitting and low-frequency attenuation, undermining clinical diagnostic reliability. To address structural fidelity challenges in multimodal medical image translation, we propose a dynamic frequency-domain balancing mechanism synergized with knowledge guidance. Specifically, we design a waveguide-adaptive frequency-domain balancing module that jointly optimizes low-frequency global anatomy and high-frequency local details. Additionally, we incorporate a vision-language model (VLM) to distill clinical prior knowledge, enforcing semantic consistency in anatomical generation. Our method integrates wavelet-based feature decomposition and dynamic frequency modulation into a diffusion model framework. Extensive experiments across multiple cross-modal datasets demonstrate significant improvements in FID, SSIM, and structural similarity metrics. Qualitative evaluation confirms markedly enhanced anatomical accuracy and clinical credibility.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

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

Abstract

Multimodal medical images play a crucial role in the precise and comprehensive clinical diagnosis. Diffusion model is a powerful strategy to synthesize the required medical images. However, existing approaches still suffer from the problem of anatomical structure distortion due to the overfitting of high-frequency information and the weakening of low-frequency information. Thus, we propose a novel method based on dynamic frequency balance and knowledge guidance. Specifically, we first extract the low-frequency and high-frequency components by decomposing the critical features of the model using wavelet transform. Then, a dynamic frequency balance module is designed to adaptively adjust frequency for enhancing global low-frequency features and effective high-frequency details as well as suppressing high-frequency noise. To further overcome the challenges posed by the large differences between different medical modalities, we construct a knowledge-guided mechanism that fuses the prior clinical knowledge from a visual language model with visual features, to facilitate the generation of accurate anatomical structures. Experimental evaluations on multiple datasets show the proposed method achieves significant improvements in qualitative and quantitative assessments, verifying its effectiveness and superiority.

Problem

• Addresses anatomical structure distortion in medical image translation
• Balances high and low frequency information dynamically
• Integrates clinical knowledge to improve cross-modality accuracy

Innovation

• Dynamic frequency balance for feature adjustment
• Wavelet transform for frequency decomposition
• Knowledge-guided fusion with clinical prior