Wottschel, V;
Alexander, DC;
Chard, DT;
Ciccarelli, O;
Miller, DH;
(2014)
Prediction of time between CIS onset and clinical conversion to MS using Random Forests.
In:
Proceedings of the Joint Annual Meeting ISMRM-ESMRMB 2014.
International Society for Magnetic Resonance in Medicine (ISMRM): Milan, Italy.
Preview |
Text
Viktor.TTEprediction.main.3436_new.pdf - Accepted Version Download (99kB) | Preview |
Abstract
CIS is diagnosed after a first neurological attack and can be considered an early stage of MS as ~80% of all CIS patients will have a second relapse within 20 years. The prediction of this second clinical relapse which marks the clinical conversion to MS (i.e., clinically-definite MS, CDMS) is very challenging, and many clinical and radiological predictors of CDMS have been identified. Machine learning techniques such as support vector machines (SVMs) have been widely applied to neuroimaging data in order to associate MRI features with binary clinical outcomes. A single-centre study has shown that it is possible to predict short-time conversion after 1 and 3 years with an accuracy of ~75 % using a priori defined features from baseline MRI measures and clinical characteristics, which were applied to support vector machines (SVMs). Random forests are another type of machine learning techniques that can easily be applied to regression problems, and consist of an ensemble of decision trees for regression where each tree is created from independent bootstraps from the input data. The present study shows the feasibility of using random forests with European multi-centre MRI data (obtained at CIS onset) to predict the actual date of conversion to CDMS rather than just a binary outcome at a fixed time point.
Archive Staff Only
 |
View Item |
