Multiple sclerosis (MS) is the most common demyelinating inflammatory disease of the central nervous system. Cognitive impairment (CI) is a common symptom in MS patients with a prevalence of 50–75 %, which produces a negative influence on their normal life (Deloire et al., 2006; Gois et al., 2021; Meca-Lallana et al., 2021; Rao et al., 1991). Early detection of CI has become an important aspect to be considered for clinical treatment. In addressing cognitive screening tools, the neuropsychological tests are the most commonly used methods in CI detection in both clinical and research fields, especially for the Montreal Cognitive Assessment (MoCA) due to the comprehensive assessments across cognitive domains (Hawkins et al., 2014; Jia et al., 2021; Nasreddine et al., 2005). However, these methods exhibit several limiting factors including limited availability of trained personnel, lack of validated tests and normative data (Meca-Lallana et al., 2021).

Conventional magnetic resonance imaging (MRI) is an indispensable clinical tool for the brain damages detection of MS. Although MS is widely known to be associated with white-matter (WM) disease, progressive gray-matter (GM) damages can be extensive and may occur partially independent of WM lesion formation in early disease stage of all MS phenotypes (Kawachi and Nishizawa, 2015; Tsouki and Williams, 2021; van Munster et al., 2015). In the 2017 revised McDonald diagnostic criteria, cortical lesions were added to fulfil MRI criteria, underscoring the importance of GM damages in MS (Thompson et al., 2018). Various mechanisms via its disease duration, mainly regarding cortical lesions and cortex atrophy in whole brain, can be more closely associated with cognitive network dysfunction and clinically significant CI than WM damages in MS (Calabrese et al., 2009; Nasios et al., 2020; Parra Corral et al., 2019; Rocca et al., 2021). Neuropsychological deficits were demonstrated in a percentage ranging from 40 to 65 % of MS patients, and have been proved to be associated with cortical demyelination and atrophy (Rinaldi et al., 2010). Therefore, cortical damages detection can be used as a noninvasive biomarker to perform cognitive assessment and might explain the CI mechanisms.

In order to extract MRI features to classify CI in individual patients, advanced statistical approaches are required. The radiomics method can convert medical images into highdimensional, mineable data via high-throughput extraction of quantitative features, followed by subsequent data analysis (Gillies et al., 2016). Several works have applied the radiomics method to perform diagnosis and prognosis discrimination in MS (Aslam et al., 2022; Liu et al., 2019; Peng et al., 2021). To our knowledge, only two recent works used this property to assessed the relationship between the cognitive status of MS patients based on the Symbol Digit Modalities Test (SDMT) and structural MRI markers through machine learning techniques (Buyukturkoglu et al., 2021; Marzi et al., 2023). However, the SDMT has been recommended to screen the information processing speed (IPS) deficits in MS, which represents only one aspect of CI. Furthermore, the structural MRI markers they enrolled were only the atrophy characteristics without the lesion features. So, this is the first study that offers a quantitative and comprehensive analysis of cortical damages based on radiomics models to classify the MS patients with CI based on the MoCA.

This study aims to investigate clinical data and the cortical damages radiomic features including cortical lesions and cortex atrophy based on conventional structural MRI in MS and to develop a discrimination model to efficiently identify the patients with CI, as defined by MoCA.