Plug-and-Play Self-Supervised Denoising for Pulmonary Perfusion MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Pulmonary dynamic contrast-enhanced (DCE) MRI is crucial for assessing lung perfusion but suffers from limited signal-to-noise ratio (SNR).
• Image noise in pulmonary perfusion MRI hinders accurate diagnosis and quantitative analysis.

Purpose of the Study:

• To develop and evaluate a novel self-supervised learning-based plug-and-play (PnP) denoising model, PnP-BSN, for improving pulmonary perfusion MRI quality.
• To compare the performance of PnP-BSN against traditional denoising methods and assess its impact on quantitative imaging metrics.

Main Methods:

• A self-supervised learning network, the asymmetric pixel-shuffle downsampling blind-spot network (AP-BSN), was trained on background-subtracted pulmonary perfusion images.
• The AP-BSN was integrated into a PnP framework (PnP-BSN) to balance noise reduction and image fidelity.
• Model performance was quantitatively assessed using SNR, sharpness, fractal dimension, and k-means segmentation, and qualitatively evaluated by two radiologists.

Main Results:

• PnP-BSN achieved significantly higher reader scores for SNR, sharpness, and overall image quality compared to a denoising convolutional neural network (DnCNN) and a Gaussian filter (p < 0.05).
• Radiologist scores for PnP-BSN were 3.56 ± 0.73 for SNR, 3.38 ± 0.64 for sharpness, and 3.53 ± 0.51 for overall image quality.
• Denoising with PnP-BSN improved quantitative fractal analysis of pulmonary perfusion MRI.

Conclusions:

• The PnP-BSN model effectively denoises pulmonary perfusion MRI, leading to superior image quality.
• This AI-driven approach enhances diagnostic accuracy and quantitative analysis in pulmonary perfusion imaging.
• PnP-BSN represents a significant advancement in medical image processing for pulmonary applications.