Dealing with confounders and outliers in classification medical studies: The Autism Spectrum Disorders case study

作者：

Highlights：

• A new workflow to face the unrepeatability problem in machine learning studies.

• It uses an autoencoder to detect multivariate outliers with unknown distributions.

• It uses the Confounding Index to find possibly misleading variables.

• A case study on Autism Spectrum Disorders shows good results holding on new data.

• Acquisition modalities turn out to be severely misleading in a neuroimaging study.

摘要

•A new workflow to face the unrepeatability problem in machine learning studies.•It uses an autoencoder to detect multivariate outliers with unknown distributions.•It uses the Confounding Index to find possibly misleading variables.•A case study on Autism Spectrum Disorders shows good results holding on new data.•Acquisition modalities turn out to be severely misleading in a neuroimaging study.

论文关键词：Confounders,Outliers,Autoencoder,Confounding Index,Machine learning,MRI,Autism Spectrum Disorders,Reproducibility

论文评审过程：Received 19 July 2019, Revised 13 December 2019, Accepted 2 July 2020, Available online 6 July 2020, Version of Record 7 August 2020.

论文官网地址：https://doi.org/10.1016/j.artmed.2020.101926