DGCD-3D: Difference-guided conditional diffusion model for low-field 3D MRI enhancement to assist stroke assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-field (LF) magnetic resonance imaging (MRI) is vital for rapid stroke diagnosis.
• LF MRI suffers from low signal-to-noise ratio (SNR) and poor image quality, hindering accurate stroke lesion identification.

Purpose of the Study:

• To develop a novel method for enhancing LF MRI image quality while preserving stroke lesion integrity.
• To improve the accuracy and reliability of stroke diagnosis using LF MRI.

Main Methods:

• Proposed a Difference-Guided Conditional Diffusion Model (DGCD-3D) incorporating a difference-adaptive forward diffusion process.
• Utilized multi-scale intrinsic features and prior spatial information of stroke lesions during training.
• Implemented a time-adaptive multi-loss optimization strategy balancing pixel-wise and perceptual losses.

Main Results:

• DGCD-3D significantly improved LF MRI image quality (PSNR = 28.26, SSIM = 0.896).
• Achieved higher consistency with high-field (HF) MRI in stroke lesion assessment (Spearman's ρ = 0.732 vs. 0.680 for LF MRI).
• Clinical authenticity assessment showed a low confusion rate (51.6%) and score (5.64), indicating clinical reliability.

Conclusions:

• DGCD-3D effectively enhances LF MRI quality and lesion assessment accuracy.
• The model demonstrates broad clinical applicability and reliability for stroke diagnosis.
• This approach offers a promising solution for improving diagnostic capabilities in resource-limited settings.