fbpx Skip to main content

Researchers were able to use AI-generated images as accurately as MRI-acquired images to identify lesions. 

Although cortical multiple sclerosis (MS) lesions are clinically significant, they are often hard to evaluate using conventional clinical MRI. Double inversion recovery (DIR) and phase-sensitive inversion recovery (PSIR) are more sensitive but tend to be unavailable in many clinical settings. In the past several years, artificial intelligence (AI) tools have been used to generate DIR and PSIR from standard clinical sequences. However, further validation from a variety of sources is needed before this technique can be applied more broadly. This study, published in Radiology, evaluated the cortical and juxtacortical MS lesion detection for diagnostic and disease monitoring on AI-generated DIR and PSIR images. These were then compared with MRI-acquired DIR and PSIR images.

Creating AI-Generated Images to Detect MS Lesions

The authors used generative adversarial networks to generate AI-based DIR and PSIR images. A total of 50 DIR and 43 PSIR images were used for this study. A randomized blinded scoring was conducted on seven readers, all with over 10 years of experience in lesion assessment. The readers were tasked with detecting lesions among both the AI-generated and MRI-acquired images. In order to precisely compare the two image sources, between-reader reliability was determined by calculating the intraclass correlation coefficient, and differences in number of lesions at the lesion subtype level were analyzed using Wilcoxon signed-rank tests. 

MRI scans of 202 patients with MS from seven centers were used. A total of 1154 lesions were detected on AI-generated DIR  images, versus 855 on MRI-acquired DIR images and 814 on MRI-acquired PSIR images. Reliability was good for both DIR and PSIR images overall. 

You May Also Like::  Brain Age Gap in NMOSD Patients Compared to Healthy Controls

The Difference Between AI-Generated Images and MRI-Acquired Images for Lesion Detection

A few differences were noted between the two different image sources. More juxtacortical lesions were found on AI-generated DIR images compared to those from MRI, but fewer of these lesions were found on AI-generated PSIR images compared to MRI-acquired PSIR images. The authors conclude the amount of detectable cortical and juxtacortical lesions in AI-generated DIR and PSIR images is at least the same as in MRI-acquired images. They also state that these data show that these AI-generated images can serve as an alternative for visualizing cortical pathologic abnormalities in patients with MS in clinical care and that they can also be used in retrospective clinical studies when MRI-acquired images are desired but unavailable.


Bouman, P. M., Noteboom, S., Nobrega Santos, F. A., Beck, E. S., Bliault, G., Castellaro, M., Calabrese, M., Chard, D. T., Eichinger, P., Filippi, M., Inglese, M., Lapucci, C., Marciniak, A., Moraal, B., Morales Pinzon, A., Mühlau, M., Preziosa, P., Reich, D. S., Rocca, M. A., . . . Steenwijk, M. D. (2023). Multicenter Evaluation of AI-generated DIR and PSIR for Cortical and Juxtacortical Multiple Sclerosis Lesion Detection. Radiology, 221425. https://doi.org/10.1148/radiol.221425