Real-Time Image Reconstruction with Deep Denoising for Low-Field MRI-Guided Interventions

Congenital heart disease (CHD) affects thousands of infants annually, often necessitating repeated catheterizations guided by X-ray fluoroscopy, leading to cumulative radiation exposure. Low-field MRI guided intervention can reduce this risk; however, image noise is problematic. To reduce this risk, We present a real-time image reconstruction pipeline for low-field (0.55T) Magnetic Resonance Imaging (MRI)-guided interventions, using a 0.55T MRI scanner. Our method integrates the GRAPPA with a modified DnCNN denoising network in a deep learning framework, achieving high-quality, low-latency image reconstructions. The proposed pipeline was validated in preclinical in vivo experiments, demonstrating a 15% improvement in signal-to-noise ratio (SNR) and a 30% reduction in image artifacts compared to conventional techniques, while maintaining real-time performance.