Diagnosis of autism spectrum disorder with machine learning techniques based on functional magnetic resonance images

Brain injuries are very serious health problems that affect people's lives. Some of these injuries can be completely reversed with methods such as medication. On the other hand, there is no known permanent treatment for damage caused by diseases such as Autism Spectrum Disorder (ASD). In such disorders, treatments are usually aimed at slowing down the progression of the disease. Therefore, it is important to diagnose the disease at an early stage before behavioural disorders occur. In this study, we present a study on the detection of ASD through rs-fMRI. rs-fMRI provides both spatial and temporal information about brain function. However, fMRI data are highly complex due to the heterogeneity of the samples, high variances between measurements due to the use of different scanner devices or parameters. In this study, ASD and healthy individuals were distinguished on 871 samples obtained from The Autism Brain Imaging Data Exchange I (ABIDE I) dataset. Accordingly, two different feature extraction methods were developed and classified with different classification algorithms. In the first method, long short-term memory network (LSTM), convolutional neural network (CNN) and hybrid models were used together for classification. In addition, different feature extraction methods such as multilayer perceptron (MLP), k-nearest neighbor algorithm (kNN), linear discriminant analysis (LDA), and support vector machines (SVM) were used for classification. Finally, an evaluation of which regions of the brain are more effective in feature selection is presented. The results obtained are promising for ASD detection on fMRI.