Meeting Banner
Abstract #2057

Prediction of Time Between CIS Onset and Clinical Conversion to MS using Random Forests

Viktor Wottschel 1,2 , Daniel C. Alexander 2 , Declan T. Chard 3 , Christian Enzinger 4 , Massimo Filippi 5 , Jette Frederiksen 6 , Claudio Gasperini 7 , Antonio Giorgio 8 , Maria A. Rocca 5 , Alex Rovira 9 , Nicola De Stefano 8 , Mar Tintor 9 , David H. Miller 3 , and Olga Ciccarelli 1

1 NMR Research Unit, Department of Brain Repair and Rehabilitation, Queen Square MS Centre, UCL Institute of Neurology, London, London, United Kingdom, 2 Microstructure Imaging Group, Centre for Medical Image Computing, Department for Computer Science, UCL, London, London, United Kingdom, 3 NMR Research Unit, Department of Neuroimflammation, Queen Square MS Centre, UCL Institute of Neurology, London, United Kingdom, 4 Department of Neurology and Section of Neuroradiology, Medical Unversity of Graz, Graz, Graz, Austria, 5 Neuroimaging Research Unit, Vita-Salute San Raffaele University, Milan, Milan, Italy, 6 Department of Neurology, Glostrup Hospital and University of Copenhagen, Copenhagen, Copenhagen, Denmark, 7 Neurology Unit, San Camillo-Forlanini Hospital, Rome, Rome, Italy, 8 Department of Neurological and Behavioral Sciences, University of Siena, Siena, Siena, Italy, 9 Department of Radiology and Neuroimmunology Unit, Hospital Vall d'Hebron, Barcelona, Barcelona, Spain

We present a feasibility study predicting the time-to-conversion (in days) from clinically isolated syndrome (CIS) to clinically definite multiple sclerosis (CDMS) using the machine learning technique random forests. T1 weighted baseline MRI data of 203 CIS patients from multiple European centres was spatially normalised and subdivided in 100 independent training and testing sets. From every training set an individual random forests was created consisting of 100 trees. The median error over all 100 bootstraps was 0.7 (range 0.57-1.16). Considering the slightly skewed data set and high similarity in T1 signal in the patient cohort, this is a very promising result.

This abstract and the presentation materials are available to members only; a login is required.

Join Here