Artificial intelligence-based assessment of cognition using optical coherence tomography, magnetic resonance volumetric and thickness measurements in multiple sclerosis patients

Cognitive problems are seen in 40-70% of patients with multiple sclerosis (MS). In previous years, cognitive impairment was not taken into account in the follow-up of the disease process, but now its importance is recognized and it is included in treatment strategies. Even in the radiologic isolated syndrome period, which is considered to be the preclinical phase of MS, it has been shown that 20-25% have cognitive impairment at various levels. Being able to show that there is a correlation between the cognitive functions of patients with MS with measurements to be made early in the course of MS and to identify patients at risk may help us to determine strategies for follow-up and treatment in individuals before cognitive pathology occurs clinically. The aim of this study was to evaluate whether the cognitive status of patients with MS is correlated with retinal nerve fiber layer (RNFL), macular volume, total retinal thickness, superficial and deep vessel density and volumetric measurements on brain magnetic resonance imaging (MRI) and to compare neurodegeneration in MS patients with healthy individuals. In our study, 60 MS patients who were followed up in the outpatient clinics of the Department of Neurology, Başkent University Faculty of Medicine Hospital and 35 cognitively normal individuals who were admitted to the neurology outpatient clinic for any reason and who did not have any diagnosis or findings of inflammatory or demyelinating diseases of the central nervous system were included as control group. In addition to detailed cognitive tests, optical coherence tomography (OCT), OCT-A, volumetric brain MRG measurements were performed in all patients and control group. All results were analyzed by comparing them with the model that best predicts cognition with artificial intelligence-based analysis. In the evaluations made, a decrease in OCT and OCT-A parameters was observed in MS patients regardless of the history of optic neuritis (ON), and this decrease was more pronounced in those who had ON. Among the cognitive tests used, symbol digit modalities test (SDMT), paced auditory serial addition test (PASAT) and judgment of line orientation (JLO) were more successful in showing the general cognitive picture. The superficial and deep layer vascular densities measured by OCT-A were found to be lower than in the control group, although there were regional differences, and especially the measurements of the deep vascular layer were found to be lower in patients with cognitive impairment. OCT, OCT-A and cognition-related brain region atrophy measurements were lower in patients with high EDSS disability score. When tested with artificial intelligence-based machine learning (ML) modeling, our accuracy rates were between 60% and 73% with different data sets. In conclusion, in our study, we tried to identify the cognitive status of MS patients by using artificial intelligence with the data obtained from OCT, OCT-A and MR measurements used in routine follow-up. Although the accuracy rates are up to 73% in the tests performed, it is necessary to increase the success rate by considering the accuracy and sensitivity rates in ML models for clinical use.