Suppressing image blurring of PROPELLER MRI via untrained method

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• PROPELLER MRI is robust to motion artifacts but increases scan time.
• Existing acceleration methods for PROPELLER MRI often degrade image quality due to blurring.
• Deep learning for MRI reconstruction typically requires external training data, posing challenges with data distribution shifts.

Purpose of the Study:

• To introduce an untrained neural network (UNN) to accelerate PROPELLER MRI.
• To suppress image blurring and enhance image quality in accelerated PROPELLER MRI.
• To eliminate the need for external training data in deep learning-based PROPELLER MRI reconstruction.

Main Methods:

• An untrained neural network (UNN) was developed to suppress image blurring.
• The UNN was integrated into the blade k-space of PROPELLER MRI acquisition.
• The method was applied to brain MRI data with undersampling factors of 2, 3, and 4.

Main Results:

• The UNN method significantly improved blade image quality in brain MRI.
• Enhanced image sharpness was achieved compared to parallel imaging and supervised learning methods.
• PROPELLER MRI acquisition was successfully accelerated without compromising image quality.

Conclusions:

• The UNN-enhanced PROPELLER method effectively suppresses blurring and improves image quality.
• This approach removes the requirement for external training data, simplifying clinical implementation.
• The technique offers a viable solution for faster and higher-quality PROPELLER MRI without affecting patient care workflows.