Multiple sclerosis (MS) is an inflammatory demyelinating disease of the central nervous system. Disease monitoring relies on the detection of new lesions or lesions increasing in size through regular MRI scans. However, interpreting essential MRI signals (T2/FLAIR) is challenging and time-consuming due to variability in lesion distribution and burden among patients. Today, artificial intelligence software exists to assist in this process. But how effective are these tools?
Two retrospective studies
Collaborations led by Myriam Edjlali (BioMaps laboratory at SHFJ) conducted two retrospective studies to test and compare the performance of platforms with distinct features: Jazz®, which automates the reading process and image display, and Pixyl.Neuro.MS®, which automatically segments and classifies lesions according to their progression.
One study [1] involved 83 MS patients referred for suspected disease progression. Follow-up MRIs were analyzed using Pixyl.Neuro.MS® software. The results of this AI-assisted reading were compared with standard radiological reports produced in routine clinical practice. AI-assisted analysis detected significantly more new lesions than conventional reading and identified expanding lesions in nearly one-third of patients, whereas none had been reported in standard reports. Integrating AI-assisted radiological findings with clinical data led to a change in therapeutic strategy for 10% of patients.
The other study [2] included follow-up MRIs of 35 MS patients with a high lesion burden. Two radiologists performed a radiological analysis of the images, followed by AI-assisted reading using either Jazz or Pixyl.Neuro.MS®. While conventional readings detected 8 new lesions in two patients, AI-assisted readings identified at least 17 true positives in 7 patients and ruled out 61 false positives.
Both studies highlight the significant contribution of AI in detecting new and evolving lesions during follow-up MRIs in MS patients. This improvement in diagnostic accuracy can meaningfully influence therapeutic decisions, particularly in patients with a high lesion burden. However, expert validation by a radiologist remains necessary to ensure the accuracy of results and eliminate false positives.
Contact at Frédéric-Joliot Institute for Life sciences:
This text was translated with the assistance of Mistral AI.