Artificial double inversion recovery images for (juxta)cortical lesion visualization in multiple sclerosis

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Cortical lesions are often subtle on conventional Magnetic Resonance Imaging (MRI).
• Double Inversion Recovery (DIR) sequences offer higher sensitivity but are challenging to acquire, potentially leading to missed diagnoses.
• This limitation impacts clinical care and the analysis of clinical trial data.

Purpose of the Study:

• To assess the effectiveness of artificially generated DIR (aDIR) images for detecting cortical lesions.
• To compare the diagnostic performance of aDIR with conventionally acquired DIR (cDIR) images.

Main Methods:

• A fully convolutional neural network was trained to generate 3D-aDIR images from 3D-T1 and 2D-proton density/T2-weighted images.
• The dataset comprised 73 patients with varying neurological conditions (49 Relapsing-Remitting, 20 Secondary Progressive, 4 Primary Progressive).
• A randomized blind scoring approach was employed to evaluate lesion detection reliability, precision, and recall on the test set.

Main Results:

• A total of 626 cortical lesions were identified on aDIR images compared to 696 on cDIR images, with high inter-rater reliability (ICC = 0.92).
• Artificial DIR demonstrated comparable precision (0.84 ± 0.06) and recall (0.76 ± 0.09) relative to conventional DIR.
• Discernibility varied across brain regions, with frontal and temporal lobes showing the most significant differences.

Conclusions:

• Artificially generated DIR images provide reliable detection of cortical lesions.
• This AI-driven approach can increase the accessibility of DIR imaging in clinical settings.
• aDIR offers a valuable method for retrospective analysis of cortical lesions in existing clinical trial datasets.