Applying Machine Learning Methods to Improve Standard Medical Research

TY - CHAP

T1 - Applying Machine Learning Methods to Improve Standard Medical Research

AU - Stahl, Daniel Richard

AU - BASES Team

N1 - Conference code: 28

PY - 2014/12/1

Y1 - 2014/12/1

N2 - Machine learning methods are successfully applied to a variety of medical fields that deal with high-dimensional data and often small sample sizes such as genetic microarray or fMRI data. The aim of this study is to assess the usefulness of machine learning algorithms for applications in medical research as an alternative or in addition to classical statistical inference methods. The usefulness of a relatively simple regularized discriminant function analyses integrated with feature variable selection based on t-scores will be assessed by reanalyzing an event-related brain potential (ERP) dataset from infants at high or low risk of developing autism [1]. Standard analysis of ERP measurements usually involves large number of univariate mean group comparisons resulting in a multiple testing problem. Machine learning methods allow to assess the predictive performance of a model, thereby avoiding multiple testing problems. The analyses showed that both machine learning methods successfully discriminated above chance between groups of infants at high and low risk of a of autism and correlation-adjusted t-scores [2] identified key variables, which separated the two groups. The results demonstrate the usefulness to integrate machine learning methods in standard statistical analyses in standard medical research in order to approach multiple testing problems more efficiently

AB - Machine learning methods are successfully applied to a variety of medical fields that deal with high-dimensional data and often small sample sizes such as genetic microarray or fMRI data. The aim of this study is to assess the usefulness of machine learning algorithms for applications in medical research as an alternative or in addition to classical statistical inference methods. The usefulness of a relatively simple regularized discriminant function analyses integrated with feature variable selection based on t-scores will be assessed by reanalyzing an event-related brain potential (ERP) dataset from infants at high or low risk of developing autism [1]. Standard analysis of ERP measurements usually involves large number of univariate mean group comparisons resulting in a multiple testing problem. Machine learning methods allow to assess the predictive performance of a model, thereby avoiding multiple testing problems. The analyses showed that both machine learning methods successfully discriminated above chance between groups of infants at high and low risk of a of autism and correlation-adjusted t-scores [2] identified key variables, which separated the two groups. The results demonstrate the usefulness to integrate machine learning methods in standard statistical analyses in standard medical research in order to approach multiple testing problems more efficiently

KW - Machine learning

M3 - Conference paper

SP - 1

EP - 4

BT - Conference paper of 28th Neural Information Processing Systems Conference

T2 - Neural Information Processing Systems

Y2 - 8 December 2014 through 13 December 2015

ER -