Enhancing Lesion Detection in Inflammatory Myelopathies: A Deep Learning-Reconstructed Double Inversion Recovery MRI Approach.

Abstract

Background and purpose: The imaging of inflammatory myelopathies has advanced significantly over time, with MRI techniques playing a pivotal role in enhancing lesion detection. However, the impact of deep learning (DL)-based reconstruction on 3D double inversion recovery (DIR) imaging for inflammatory myelopathies remains unassessed. This study aims to compare acquisition time, image quality, diagnostic confidence, and lesion detection rates among sagittal T2WI, standard DIR, and DL -reconstructed DIR in patients with inflammatory myelopathies.

Materials and methods: In this observational study, patients diagnosed with inflammatory myelopathies were recruited between June 2023 and March 2024. Each patient underwent sagittal conventional turbo spin-echo sequences and standard 3D DIR (T2WI and standard 3D DIR used as references for comparison), followed by an undersampled accelerated DIRDL examination. Three neuroradiologists evaluated the images using a four-point Likert scale (from 1 to 4) for overall image quality, perceived signal-tonoise ratio, sharpness, artifacts, and diagnostic confidence. The acquisition times, and lesion detection rates were also compared between the acquisition protocols.

Results: A total of 149 participants were evaluated (mean age 40.6 ± 16.8 years; 71 females). The median acquisition time for DIRDL was significantly lower than for standard DIR (298 seconds [interquartile range (IQR), 288-301] vs 151 seconds [IQR, 148-155]; P < 0.001), showing a 49%time reduction. DIRDL images scored higher in overall quality, perceived signal-to-noise ratio, and artifact noise reduction (all P < 0.001). There were no significant differences in sharpness (P = 0.07), or diagnostic confidence (P = 0.06) between the standard DIR and DIRDL protocols. Additionally, DIRDL detected 37% more lesions compared to T2WI (300 vs. 219; P < 0.001).

Conclusions: DIRDL significantly reduces acquisition time and improves image quality compared to standard DIR without compromising diagnostic confidence. Additionally, DIRDL enhances lesion detection in patients with inflammatory myelopathies, making it a valuable tool in clinical practice. These findings underscore the potential for incorporating DIRDL into future imaging guidelines.

Abbreviations: DL = deep learning; DIR = double inversion recovery; IQR = interquartile range; MS = multiple sclerosis; AQP4+NMOSD = AQP4-IgG positive neuromyelitis optica spectrum disorders; MOG = myelin oligodendrocyte glycoprotein; MOGAD = MOG antibody-associated diseases.