Abstract #0225

Perils in the Use of Cross-validation for Performance Estimation in Neuroimaging-based Diagnostic Classification

Pradyumna Lanka¹, D Rangaprakash¹, and Gopikrishna Deshpande^1,2,3

¹AU MRI Research Center, Department of Electrical and Computer Engineering, Auburn University, Auburn, AL, United States, ²Department of Psychology, Auburn University, Auburn, AL, United States, ³Alabama Advanced Imaging Consortium, Auburn University and University of Alabama, Birmingham, AL, United States

In this study, we highlight the fact that cross-validation accuracy might not be a good measure of performance estimation in neuroimaging-based diagnostic classification, especially with smaller sample sizes typically encountered in neuroimaging. We trained an array of classifiers using resting state fMRI-based functional connectivity measures from subjects in a particular age group using cross-validation, and then tested on an independent set of subjects with the same diagnosis (mild cognitive impairment and Alzheimer’s disease), but from a different age group. We demonstrate that cross-validation accuracy might give us an inflated estimate of the true performance of the classifiers.

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

Join Here