Researchers have developed an explainable artificial intelligence (AI) model capable of predicting future cognitive decline in people with multiple sclerosis (MS) with around 90% accuracy, offering a potential new tool to support earlier intervention and more personalised patient care.
Published in the European Journal of Neurology, the study combined MRI brain scans with clinical and demographic data from 224 people with MS and 115 healthy controls. Participants underwent detailed neurological, cognitive and MRI assessments and were followed for a median of 3.4 years.
The researchers used the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) criteria to classify patients with Mild or Major Neurocognitive Disorder (NCD), a framework that considers both cognitive performance and its impact on everyday functioning.
At the start of the study, 4% of participants met the criteria for Mild NCD and 11% for Major NCD. During follow-up, 12% experienced cognitive deterioration, with some progressing from normal cognition to Mild or Major NCD.
The team developed a hybrid deep learning model that analysed structural MRI scans alongside information including age, disability level, disease duration, cognitive reserve and brain volumetric measurements. The model achieved a validation accuracy of 90% and an area under the curve of 0.89, indicating strong predictive performance.
Unlike many previous AI systems, the model was designed to explain how it reached its predictions. The most influential factors associated with future cognitive decline were cortical grey matter volume, age, thalamic volume, hippocampal volume, T2 lesion volume and cognitive reserve.
Analysis of MRI data also highlighted the importance of frontal brain regions, together with temporal, parietal, occipital and cerebellar structures that are known to support memory, attention, processing speed and executive function.
The authors say explainable AI could help neurologists identify patients at greatest risk of cognitive deterioration before symptoms become more severe, allowing closer monitoring, earlier cognitive rehabilitation and potentially more personalised treatment strategies.
| Sign up to the ACNR email newsletter here |
Although the findings are promising, the researchers note that the model requires validation in larger, independent patient populations before it can be adopted in routine clinical practice.
This news item has been summarised using AI and checked by humans before publication.
Sources
| Sign up for our online community for neurology specialists |
