Related Experiment Video
Updated: Mar 6, 2026
Author Spotlight: Investigating Early Events and Long-Term Effects of ACL Injuries for Osteoarthritis Progression
Published on: September 29, 2023
Visual assessment of AI-reconstructed knee MRI: A pilot study
S S Ghotra1, L Buttex2, L Gallus3
1School of Health Sciences (HESAV), University of Applied Sciences and Arts Western Switzerland (HES-SO), Lausanne 1011, Switzerland; Department of Diagnostic & Interventional Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne 1011, Switzerland; Radiography and Diagnostic Imaging, School of Medicine, University College Dublin, Ireland.
Introduction:
Artificial intelligence is increasingly influencing medical imaging workflows by enhancing image quality and reducing acquisition time. The purpose of this study is to evaluate the use of artificial intelligence (AI) reconstruction methods for knee magnetic resonance imaging (MRI) investigations.
Methods:
An exploratory comparison was performed between AI-enhanced and standard 3T knee MRI protocols. Sequences included sagittal T1-weighted (T1w) fast spin echo (FSE), proton density-weighted (PDw) FSE fat saturation (FS) and T2w FSE, with additional PDw FSE FS axial and coronal views. Parameters were adjusted to improve image quality (IQ) and shorten scan duration. The optimised AI protocol was tested on ten healthy volunteers. Three MRI experts independently assessed images visually using ViewDEX. Visual grading analysis (VGA), inter-observer agreement (Kappa), and visual grading characteristics (VGC) were utilised for evaluation.
Results:
VGA results demonstrated that AI reconstruction produced equal or superior scores across most anatomical and IQ criteria, with a time reduction of 36.9 % (8:22 min vs 13:15 min). All the AI-enhanced sequences were judged as clinically acceptable. Kappa values indicated moderate-to-good agreement for AI sequences, whereas agreement for standard sequences ranged from low to good. VGC confirmed statistically higher performance for AI images (AUCVGC 0.76-0.81, p ≤ 0.05).
Conclusion:
This study indicates that incorporating AI into knee MRI protocols can substantially enhance overall image quality while simultaneously reducing acquisition time by 36.9 %. However, further research is needed to reinforce these findings through clinical validation.
Implication For Practice:
AI can reduce scan time without lowering image quality, improving workflow and patient throughput. Understanding how acquisition parameters affect image quality and artefacts is crucial when integrating AI reconstruction into protocols.
