Attention Deficit Hyperactivity Disorder (ADHD) is a type of mental health disorder that can be seen from children to adults. Accurate diagnosis of ADHD as early as possible is very important for the treatment of patients in clinical applications. In this paper, we propose two novel deep learning approaches for ADHD classification based on functional magnetic resonance imaging (fMRI). The first method incorporates independent component analysis with convolutional neural network. It first extracts independent components from each subject. The independent components are then fed into a convolutional neural network as input features to classify the ADHD patients from typical controls. The second method, called the correlation autoencoder method, uses correlations between regions of interest of the brain as the input of an autoencoder to learn latent features, which are then used in the classification task by a new neural network. These two methods use different ways to extract the inter-voxel information from fMRI, but both use convolutional neural networks to further extract predictive features for the classification task. Empirical experiments show that both methods are able to outperform the classical methods such as logistic regression, support vector machines, and some other methods used in previous studies.

Autism spectrum disorder (ASD) is a type of mental health disorder, and its prevalence worldwide is estimated at about one in 100 children. Accurate diagnosis of ASD as early as possible is very important for the treatment of patients in clinical applications. ABIDE Ⅰ dataset as a repository of ASD is used much for developing classifiers for ASD from typical controls. In this paper, we mainly consider three types of correlations including Pearson correlation, partial correlation, and tangent correlation together based on different numbers of regions of interest (ROIs) from only one atlas, and then twelve deep neural network models are used to train 884 subjects with 5, 10, 15, 20-fold cross-validation on two types of split methods including stratified and non-stratified methods. We first consider six metrics to compare the model performance among the split methods. The six metrics are F1-Score, precision, recall, accuracy, and specificity, area under the precision-recall curve (PRAUC), and area under the Receiver Characteristic Operator curve (ROCAUC). The study achieved the highest accuracy rate of 71.94% for 5-fold cross-validation, 72.64% for 10-fold cross-validation, 72.96% for 15-fold cross-validation, and 73.43% for 20-fold cross-validation.

The prediction interval is an important guide for financial organizations to make decisions for pricing loan rates. In this paper, we considered four models with bootstrap technique to calculate prediction intervals. Two datasets are used for the study and $ 5 $-fold cross validation is used to estimate performance. The classical regression and Huber regression models have similar performance, both of them have narrow intervals. Although the RANSAC model has a slightly higher coverage rate, it has the widest interval. When the coverage rates are similar, the model with a narrower interval is recommended. Therefore, the classical and Huber regression models with bootstrap method are recommended to calculate the prediction interval.

This study employed Graph Neural Networks (GNNs), specifically an unsupervised GNN, to extract node embeddings from brain regions in both Autism Spectrum Disorder (ASD) and control groups. The objective was to identify potential biomarkers by analyzing node embeddings extracted from a graph model based on functional Magnetic Resonance Imaging (fMRI) data. Permutation tests were conducted to identify regions with significant differences in their embeddings between the two groups. Our results revealed several regions exhibiting significant differences, including the cerebellum, temporal lobe, and occipital lobe. These findings align with previous studies on ASD. Moreover, novel regions such as Vermis_3, Vermis_4_5, Fusiform areas, Parietal, and Cuneus were identified, emphasizing the need for further investigation. This study underscores the potential of GNNs in analyzing brain networks for ASD biomarker discovery. The identified regions warrant additional validation and exploration to understand their association with specific domains of ASD symptoms. Our approach presents a promising avenue to advance the diagnosis of ASD and to improve our understanding of its underlying neural basis.