ADHD classification using auto-encoding neural network and binary hypothesis testing

Highlights：

• A deep-learning-based ADHD classification framework is proposed by using a binary hypothesis testing of test data.

• Under a label hypothesis, functional connectivies of test data are used in the feature selection of training data.

• The high-level features are learned from the selected features of training data via a modified auto-coding network.

• A variability score is adopted to measure the clustering performance of these learned high-level features.

• A label is given to test data by the variability score comparison on the high-level features under binary hypotheses.