Summary:
UCLA Researchers in the Department of Radiological Sciences have developed a deep learning-based approach for super-resolution magnetic resonance imaging. This method incorporates a novel slice-profile transformation super-resolution framework for through-plane super-resolution of multi-slice 2D MRI. This approach results in higher accuracy and improved clinical assessment when compared to current super-resolution MRI approaches.
Background:
Magnetic resonance imaging (MRI) is a medical imaging technique that uses a magnetic field and computer-generated radio waves to create detailed images of organs and tissues in the human body. MRI is the most frequently used imaging instrument for the brain, spinal cord and other organs and is widely accepted in diagnostic procedures. The data from conventional 2-dimensional (2D) MRI scans can be viewed into multiple imaging planes, such as coronal and axial images, but its clinical utility is limited due to a thicker through-plane resolution. Therefore, alternative approaches are to acquire other 2D MRI scans in orthogonal imaging planes to allow multiple imaging views, but this method suffers from extended MRI acquisition times, which reduce their clinical applicability. A common practice using deep learning super-resolution (SR) MRI to circumvent these issues has become increasingly popular but, in most cases, have only shown promise in achieving high in-plane resolution, not through-plane resolution. Applying deep learning to improve through-plane resolution is still challenging due to the difficulties of achieving proper training for deep learning models. Novel deep learning MRI techniques for high through-plane resolution could revolutionize the efficiency and accuracy of using MRI in diagnosis and guided treatment.
Innovation:
UCLA researchers led by Dr. Kyung Sung in the Department of Radiological Sciences have developed a novel slice-profile transformation super-resolution (SPTSR) framework with deep generative learning for through-plane super-resolution (SR) of multi-slice 2D MRI imaging. After training their framework with 3,453 clinical subjects, they validated the framework in 392 patients and had two genitourinary radiologists qualitatively evaluate images taken with the framework in 50 patients. They found that this approach overcomes some of the current limitations of conventional approaches and reduces the need for acquiring additional orthogonal imaging planes. Furthermore, this method helps address the physical discrepancies between slice-encoding and frequency/phase-encoding in 2D MRI. Their novel approach is applicable across various disease contexts and results in higher accuracy imaging and reduced MRI acquisition time, making it a valuable tool in clinical settings to increase patient care and reduce the time burden on clinicians.
Potential Applications:
• Improves multi-slice 2D MRI imaging
• Physics-informed down-sampled 2D MRI scans can be used for other data processing procedures
• Synthesizes 3D MRI volumes from serial multi-slice 2D MRI scans
Advantages:
• Faster MRI acquisition time
• Low noise level, including reduced smear effect
• Reduction of stair-case artifact
• Higher overall image quality and resolution for greater diagnostic confidence and depth
Development to Date:
The researchers trained and validated the framework in 3453, 392, and 50 clinical subjects, respectively. MRI images taken from 50 patients were evaluated by two abdominal radiologists, who found that the framework achieved excellent overall image quality with excellent sharpness, minimal artifacts, and low noise level.
UCLA Newsroom: A novel artificial intelligence technique for generating through-plane super-resolution MRI images