Multiscale Deep Learning Convolutional Neural Network for ADHD Detection using EEG

Background: Attention-Deficit Hyperactivity Disorder (ADHD) is one of the mostcommon neuro-developmental disorders diagnosed in childhood. The medical diagnosis of ADHDtypically involves extensive time-consuming interviews and tests. Electroencephalography (EEG) is amethodology to record brain activity. In recent years, several Machine Learning (ML) models havebeen developed for diagnosing ADHD using EEG data. However, they are unreliable for two reasons: 1)Performing manual pre-processing and feature selection are error-prone, and 2) training and testingthe model using the same subjects is defective. The main goal of this study is to develop a morerobust and reliable learning model for detecting ADHD. Methods: To achieve this objective, we developed an effective Deep-Learning MultiscaleConvolutional Neural Network called EEG-MSCNet. We used a Leave-One-Out subject-cross-validation approach validation to obtain more reliable results. Results: Our EEG-MSCNet model achieved an accuracy of 88.42%, surpassing the previous state-of-the-art models, including EEGNet with an accuracy of 83%. It features robust, automatic feature extraction directly from raw EEG signals. To support transparency and reproducibility, we have made the source code, databases, and logs publicly available, facilitating validation and further research. Conclusion: The proposed model achieved the best f1-score and accuracy compared withstate-of-the-art DL models. Second, an automatic end-to-end classification process could beperformed using our proposal. Consequently, it can be applied on an outpatient basis, allowingphysicians to follow their patients more closely and, as a result, speed up their treatment time.