Contributions
Abstract: EP1520
Type: Poster Sessions
Abstract Category: Pathology and pathogenesis of MS - MRI and PET
Introduction: Deriving biomarkers for cognitive impairment based on single neuroimaging modality analysis is limiting due to the complexity and heterogeneity of multiple sclerosis (MS) disease. Hence, a method which simultaneously analyzes many variables complementing each other is needed to better understand cognitive decline in MS.
Objectives: In this study we applied machine learning method to cognition in MS to determine relationships among different variables of disease beyond their individual values and find inherent patterns in data. We aimed to test if our algorithm classifies memory-impaired and cognitively preserved patients into different groups accurately.
Methods: 36 relapsing remitting relapsing remitting MS (RRMS) patients (16 with isolated memory impairment, 20 cognitively preserved) from MEM CONNECT cohort were included in this study. We trained a linear discriminant analysis based pattern classifier using MRI based neuroimaging data (cortical thickness/volume, global brain and subcortical gray matter structure volumes, T2 lesion loads, diffusion tensor imaging based white matter integrity measures and resting state fMRI) and other patient characteristics (disease duration, IQ, years of education, age, sex, depression and fatigue levels) as the features.
Results: Our results revealed that, when all above mentioned features were combined, our algorithm produced a classification with 70 % accuracy, 67% sensitivity and 65 % specificity for isolated memory impaired-cognitively preserved MS patients comparison.
Conclusions: These results show that isolated memory impairment in MS can be predicted using MRI based neuroimaging data and other patient characteristics with machine learning method. This pattern can be used as a biomarker and would help to develop personalized therapies specific to memory impairment in MS. Future studies with larger sample sizes should address impairments in the other domains (e.g., processing speed, attention, executive functions, language). In addition to supervised method, application of unsupervised pattern classification methods should be used to find new cognitive phenotypes which have not yet been detected.
Disclosure: Nothing to disclose.
Abstract: EP1520
Type: Poster Sessions
Abstract Category: Pathology and pathogenesis of MS - MRI and PET
Introduction: Deriving biomarkers for cognitive impairment based on single neuroimaging modality analysis is limiting due to the complexity and heterogeneity of multiple sclerosis (MS) disease. Hence, a method which simultaneously analyzes many variables complementing each other is needed to better understand cognitive decline in MS.
Objectives: In this study we applied machine learning method to cognition in MS to determine relationships among different variables of disease beyond their individual values and find inherent patterns in data. We aimed to test if our algorithm classifies memory-impaired and cognitively preserved patients into different groups accurately.
Methods: 36 relapsing remitting relapsing remitting MS (RRMS) patients (16 with isolated memory impairment, 20 cognitively preserved) from MEM CONNECT cohort were included in this study. We trained a linear discriminant analysis based pattern classifier using MRI based neuroimaging data (cortical thickness/volume, global brain and subcortical gray matter structure volumes, T2 lesion loads, diffusion tensor imaging based white matter integrity measures and resting state fMRI) and other patient characteristics (disease duration, IQ, years of education, age, sex, depression and fatigue levels) as the features.
Results: Our results revealed that, when all above mentioned features were combined, our algorithm produced a classification with 70 % accuracy, 67% sensitivity and 65 % specificity for isolated memory impaired-cognitively preserved MS patients comparison.
Conclusions: These results show that isolated memory impairment in MS can be predicted using MRI based neuroimaging data and other patient characteristics with machine learning method. This pattern can be used as a biomarker and would help to develop personalized therapies specific to memory impairment in MS. Future studies with larger sample sizes should address impairments in the other domains (e.g., processing speed, attention, executive functions, language). In addition to supervised method, application of unsupervised pattern classification methods should be used to find new cognitive phenotypes which have not yet been detected.
Disclosure: Nothing to disclose.