SynPoC: a high-quality generative diffusion model for transforming ultra-low-field point-of-care MRI using high-field...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Ultra-low-field (ULF) point-of-care (PoC) Magnetic Resonance Imaging (MRI) offers increased accessibility but suffers from lower image quality (SNR, resolution, contrast).
• Current limitations hinder the diagnostic utility of ULF MRI, necessitating image enhancement techniques.

Purpose of the Study:

• To introduce SynPoC, a generative diffusion model designed to synthesize high-field MRI-like images from ULF MRI data.
• To enhance the image quality of ULF MRI, improving anatomical clarity and structural detail.

Main Methods:

• Developed SynPoC, a conditional adversarial diffusion model leveraging noise and contrast features for inter-field image synthesis.
• Evaluated SynPoC on a multi-site dataset of 180 participants (healthy and various brain conditions).
• Utilized quantitative and volumetric analyses to compare enhanced ULF MRI with high-field MRI.

Main Results:

• SynPoC successfully enhanced ULF MRI images, improving anatomical clarity and structural alignment with high-field MRI.
• Quantitative and volumetric analyses confirmed the improvements in image quality.
• The model shows potential for research applications in medical imaging.

Conclusions:

• SynPoC demonstrates promise for enhancing ULF MRI image quality, potentially broadening its applications.
• Generative models like SynPoC carry a risk of hallucinated features, requiring careful inspection and validation.
• Further validation is essential before SynPoC can be considered for diagnostic decision-making.