Predicting outcome in clinically isolated syndrome using machine learning.

V Wottschel,M.L Stromillo,P.P Kwok,D.H Miller,A.J Thompson,N De Stefano,O Ciccarelli,D.T Chard,D.C Alexander

doi:10.1016/j.nicl.2014.11.021

Abstract

We aim to determine if machine learning techniques, such as support vector machines (SVMs), can predict the occurrence of a second clinical attack, which leads to the diagnosis of clinically-definite Multiple Sclerosis (CDMS) in patients with a clinically isolated syndrome (CIS), on the basis of single patient's lesion features and clinical/demographic characteristics.Seventy-four patients at onset of CIS were scanned and clinically reviewed after one and three years. CDMS was used as the gold standard against which SVM classification accuracy was tested. Radiological features related to lesional characteristics on conventional MRI were defined a priori and used in combination with clinical/demographic features in an SVM. Forward recursive feature elimination with 100 bootstraps and a leave-one-out cross-validation was used to find the most predictive feature combinations.30 % and 44 % of patients developed CDMS within one and three years, respectively. The SVMs correctly predicted the presence (or the absence) of CDMS in 71.4 % of patients (sensitivity/specificity: 77 %/66 %) at 1 year, and in 68 % (60 %/76 %) at 3 years on average over all bootstraps. Combinations of features consistently gave a higher accuracy in predicting outcome than any single feature.Machine-learning-based classifications can be used to provide an “individualised” prediction of conversion to MS from subjects' baseline scans and clinical characteristics, with potential to be incorporated into routine clinical practice.

Highlights

Machine learning is an emerging area of computer science and artificial intelligence that provides an increasing variety of algorithms capable of learning patterns from input data to solve classification and prediction problems (Bishop, 2006)
A key question that is of direct clinical relevance, and is addressed in this study, is whether support vector machines (SVMs) can be applied to MRI scans and clinical characteristics of patients with early features of Multiple Sclerosis (MS) to predict their prognosis
The demographic and clinical characteristics of patients are summarised in Table 1. 30 % and 44 % of patients developed clinically-definite Multiple Sclerosis (CDMS) within one and three years respectively

Summary

Introduction

Machine learning is an emerging area of computer science and artificial intelligence that provides an increasing variety of algorithms capable of learning patterns from input data to solve classification and prediction problems (Bishop, 2006). In the context of medical imaging, SVMs have shown promise for binary classifications (e.g. disease vs healthy status), on the basis of imaging characteristics (Ashburner and Klöppel, 2011). In this context, SVMs first learn the characteristics of, say, MRI scans in each of two groups; they use that knowledge to assign new brain scans, which have not been used in the training procedure, to one of the two groups. A key question that is of direct clinical relevance, and is addressed in this study, is whether SVMs can be applied to MRI scans and clinical characteristics of patients with early features of Multiple Sclerosis (MS) to predict their prognosis

Objectives

Methods

Results

Conclusion