AI Learning Model Predicted Cognitive Status in Patients With MS

Follow-up cognitive status was predicted with 90% accuracy when using artificial intelligence (AI) to predict the worsening of cognitive function in patients with multiple sclerosis (MS), according to a new study published in the European Journal of Neurology.¹ The results show promise in giving patients with MS a personalized cognitive assessment.

Patients with MS are often affected by cognitive deficits due to the disease, including trouble walking, vision changes, and weakness that can lead to the loss of the ability to walk.² AI has been used to help capture complex patterns in testing and imaging, with past studies finding it capable of identifying cognitive impairment in patients with MS.¹

This study aimed to assess how prevalent mild and major neurocognitive disorders (NCDs) were in a cohort of patients with MS based on a neurophysical evaluation before testing a developed multimodal AI model to predict the development of NCDs in the future. Key predictive factors were also identified with the use of AI.

Participants were chosen retrospectively from a data repository of the Neuroimaging Research Unit from the IRCCS San Raffaele Scientific Institute in Milan, Italy. Participants were aged 18 years or older; had no systemic, neurologic disease, or psychiatric disease other than MS; and were native speakers of Italian.

Patients needed to have 2 evaluations for their clinical and neuropsychological health, with a minimum of 1 year of follow-up time in between the 2 evaluations. All participants needed to be on stable disease-modifying treatment for at least 3 months before the assessments and have been relapse-free and steroid-free.

All patients had a neurological examination and an Expanded Disability Status Scale score. Patients with MS were given the Brief Repeatable Battery of Neuropsychological Tests at baseline and follow-up. Depressive symptoms, fatigue, and health-related quality of life were also assessed at baseline. Brain scans were collected for all patients. Changes in diagnostic status were used to separate patients into 2 groups.

There were 224 patients with MS and 115 controls who were included in the study. A total of 4% of the patients with MS met criteria for mild NCD, and 11% met criteria for major NCD. Patients with major NCD were older, had a longer duration of disease, had lower cognitive reserve, had fewer years of education, and reported a lower quality of life.

Of the patients with MS, 12% had cognitive worsening after a median (IQR) follow-up of 3.4 (2.0-6.1) years. A total of 190 patients were cognitively preserved at baseline, of which 6% developed a mild NCD and 6% developed a major NCD; at follow-up, 4 of the 19 patients with mild NCD at baseline had worsened NCD.

When using the AI model, the mean validation accuracy was 90%, with a mean F1-score of 81%. A mean precision of 95% was found during the validation phase, and the recall score was a mean of 82%. The mean (SD) area under the curve was 0.91 (0.06) and 0.89 (0.09) for the training and validation phases, respectively. There was a mean uncertainty of 3.42% (0.70%).

There were some limitations to this study. There was no balance between cognitively stable vs worsened patients, as only 12% showed cognitive decline. Statistical power may have been reduced due to the smaller proportion of patients with a mild or major NCD. Additionally, leisure activities were excluded from the measurement of the cognitive reserve index, and anxiety was not assessed in the cognitive evaluation. The researchers emphasized that the model’s generalizability will need to be tested in the future.

“The multimodal AI implemented in this study demonstrated high accuracy and reliability with explainability analyses confirming the involvement of brain regions linked to cognitive impairment,” they concluded. “These results support the potential for AI-based tools in personalized cognitive assessment and monitoring in MS.”

References

1. Storelli L, Mistri D, Mastropasqua A, et al. Explainable artificial intelligence to predict neurocognitive disorder progression in multiple sclerosis using MRI and clinical data. Euro J Neurol. 2026;33:e70568. doi:10.1111/ene.70568
2. Mayo Clinic Staff. Multiple sclerosis. Mayo Clinic. November 1, 2024. Accessed April 27, 2026. https://www.mayoclinic.org/diseases-conditions/multiple-sclerosis/symptoms-causes/syc-20350269