Classification of Autism Spectrum Disorder Using a 3D-CNN Ensemble Model and Regional Homogeneity Data from the ABIDE I Dataset

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by deficits in communication and social interaction, and presence of repetitive behaviors, and restricted interests. Many structural and functional brain alterations can be observed in individuals with ASD. Therefore, functional magnetic resonance imaging (fMRI) has been used to uncover the neurobiology of this disorder. Numerous studies have used machine learning models to classify ASD using the Autism Brain Imaging Data Exchange fMRI dataset. Currently, the state-of-the-art accuracy is close to 70% when using multi-site samples. Here, we propose a 3D version of the SqueezeNet resource-efficient architecture. We used regional homogeneity data available from the preprocessed version of the ABIDE I dataset. A 3D-CNN ensemble model was trained using a 5-fold cross-validation procedure. Our model achieved 68.93% accuracy and 0.72 Area Under the Receiver Operating Characteristic Curve (AUC) value. Despite not achieving the highest overall accuracy, our SqueezeNet-based 3D-CNN can still outperform another 3D-CNN model based on the same dataset. Using a resource-efficient architecture, we developed a model capable of processing complex 3D medical imaging data. Moreover, we achieved comparable state-of-the-art results.