Related Experiment Video
Updated: Jun 12, 2026
17:16
Registered Bioimaging of Nanomaterials for Diagnostic and Therapeutic Monitoring
Published on: December 9, 2010
10.7K
Anatomy-Aware Sketch-Guided Latent Diffusion Model for Orbital Tumor Multi-Parametric MRI Missing Modalities
IEEE Transactions on Medical Imaging
|December 29, 2025
Summary
This study introduces an anatomy-aware sketch-guided latent diffusion model (ASLDM) for synthesizing missing MRI modalities. ASLDM enhances anatomical accuracy and structural consistency in multi-parametric MRI (mpMRI) completion.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- Synthesizing missing modalities in multi-parametric MRI (mpMRI) is crucial for accurate tumor diagnosis but faces challenges with incomplete data and varying modalities.
- Conventional diffusion models in the image domain are memory-intensive and lack anatomical fidelity due to noise-space supervision.
- Standard latent diffusion models (LDMs) lack structural priors and struggle with effective multi-modal integration.
Purpose of the Study:
- To develop a novel latent diffusion model (LDM) framework for flexible and structure-preserving MRI synthesis.
- To improve the anatomical accuracy and cross-modal consistency of synthesized mpMRI data.
- To address limitations of existing methods in handling incomplete multi-modal MRI data.
Main Methods:
- Proposed the anatomy-aware sketch-guided latent diffusion model (ASLDM) incorporating an anatomy-aware feature fusion module.
- Utilized cross-attention with tumor region masks and edge-based anatomical sketches for structure-guided denoising.
- Implemented a modality synergistic reconstruction strategy for joint modeling of available and missing modalities.
- Introduced L1 and SSIM image-level losses for pixel-space supervision.
Main Results:
- ASLDM demonstrated superior synthesis quality and structural consistency compared to state-of-the-art methods on orbital tumor and BraTS2024 datasets.
- The proposed method effectively integrates multiple MRI modalities and handles arbitrary missing data scenarios.
- Pixel-space supervision significantly improved anatomical accuracy over pure noise-based loss training.
Conclusions:
- ASLDM offers a robust framework for clinically reliable multi-modal MRI completion.
- The integration of structural priors and synergistic reconstruction enhances the fidelity of synthesized MRI data.
- This approach holds significant potential for improving diagnostic accuracy in oncology through advanced MRI techniques.
