Deep-Learning Generated Synthetic Double Inversion Recovery Images Improve Multiple Sclerosis Lesion Detection

Area of Science:

• Artificial Intelligence in Medical Imaging
• Neuroimaging
• Radiology

Background:

• Multiple sclerosis (MS) diagnosis relies on detecting brain lesions.
• Conventional MRI sequences like FLAIR have limitations in lesion detection.
• Synthetic imaging techniques offer potential for improved diagnostic performance.

Purpose of the Study:

• To implement a deep-learning tool for generating synthetic double inversion recovery (synthDIR) images.
• To compare the diagnostic performance of synthDIR with conventional MRI sequences for MS lesions.
• To evaluate the utility of artificial intelligence in enhancing MS imaging.

Main Methods:

• A retrospective analysis of 100 MS patients was conducted.
• An artificial neural network (DiamondGAN) was trained to create synthDIR images from T1, T2, and FLAIR acquisitions.
• Two independent readers assessed MS lesions on synthDIR, trueDIR, and FLAIR images, comparing lesion counts and contrast-to-noise ratios.

Main Results:

• SynthDIR detected significantly more MS lesions than FLAIR (P < 0.001), particularly juxtacortical lesions.
• Interrater reliability was excellent across all assessed modalities (FLAIR, synthDIR, trueDIR).
• SynthDIR exhibited a higher contrast-to-noise ratio than FLAIR (P = 0.009) and was comparable to trueDIR.

Conclusions:

• Computationally generated synthDIR images improve MS lesion depiction compared to standard MRI modalities.
• This study highlights the potential of artificial intelligence to enhance medical imaging for specific pathologies like MS.
• SynthDIR represents a promising AI-driven tool for more accurate MS lesion detection.