Feasibility and image quality evaluation of accelerated T2-weighted and time-of-flight MRI using AI-assisted compressed sensing and deep reconstruction.

Authors

Abstract

Magnetic Resonance Imaging (MRI) provides high-resolution, non-invasive diagnostic images, but long acquisition times increase patient burden and motion artefacts. Compressed sensing helps reduce scan duration, yet further image quality improvements remain necessary. Recent advances in artificial intelligence (AI) have shown promise in enhancing image quality and accelerating MRI acquisition. In this study, we aim to evaluate the feasibility of accelerated MRI acquisition using AI-assisted compressed sensing (ACS) and deep reconstruction by assessing scan time, quantitative image quality, and radiologist-rated image quality in T2-weighted and time-of-flight (TOF) imaging. Healthy volunteers underwent accelerated 3D T2-weighted (n = 11) and TOF (n = 10) MRI using a vendor-implemented framework integrating AI-assisted compressed sensing for data acquisition and deep learning-based reconstruction for image enhancement. Multiple acceleration factors and reconstruction levels were assessed. Conventional accelerated imaging without AI assistance served as the reference. Quantitative evaluation included signal-to-noise ratio (SNR) measurements in gray matter, white matter, and cerebrospinal fluid. Qualitative assessment was independently performed by three radiologists using a five-point Likert scale evaluating overall image quality, contrast, sharpness, noise, and artifacts. Maximum intensity projection (MIP) images derived from TOF datasets were also analyzed. For T2-weighted imaging, the combined ACS and AI-based reconstruction framework increased SNR by up to 1.3-fold in gray matter, 2.7-fold in white matter, and 3.5-fold in cerebrospinal fluid compared with conventional accelerated imaging. In TOF imaging, SNR increased by up to 2.8-fold across acceleration factors. Qualitative assessment demonstrated significant improvements in overall image quality, sharpness, and noise reduction for T2-weighted images, as well as enhanced overall image quality, contrast, and artifact suppression for TOF images. Similar improvements were observed in maximum intensity projection (MIP) images reconstructed from TOF datasets. Accelerated acquisition was achieved with substantially reduced scan times at higher acceleration factors. These findings suggest that ACS combined with deep reconstruction offers practical advantages over conventional compressed sensing in terms of both image quality and acquisition speed, supporting its potential for improving clinical workflow efficiency and reducing patient burden. However, this study was limited to a small cohort of healthy volunteers, and further validation in patient populations with diverse pathologies is required to establish diagnostic performance. Caution is also warranted regarding potential AI-related risks such as over-smoothing or artificial signal generation. AI-assisted compressed sensing combined with deep reconstruction enabled accelerated T2-weighted and TOF MRI while maintaining or improving quantitative and qualitative image quality. The proposed framework supported higher acceleration factors with favorable image quality performance, demonstrating the technical feasibility of accelerated MRI acquisition in healthy subjects. Although these findings support the feasibility of AI-accelerated MRI, caution is warranted regarding potential AI-related risks, including over-smoothing and artificial signal generation. MRI scans can take a long time, which may be difficult for some people. This study tested a faster MRI method that combines artificial intelligence with advanced image processing and compared it with standard fast MRI scans. This study found that the new method shortened scan time and improved image quality in healthy volunteers. This matters because faster scans with clear images may improve patient comfort and support more efficient healthcare, although more studies in patients are needed.

Tags

Topics