A deep learning approach for prediction of ADHD using brain structure

Abstract

Attention deficit hyperactivity disorder (ADHD) is a complex condition affecting the brain’s neurodevelopmental processes. Prompt treatment and accurate diagnosis can alter neural connections and improve symptoms. This article focuses on the classification of MRI scans of individuals with ADHD and those without the disorder using deep learning algorithms. Pre-trained models like VGG-16, RegNet-50, and DenseNet-121 were used, along with non-pretrained models like CNN and convolutional LSTM. VGG16 is known for its intricate architecture, while CNNs have undergone significant advancements, resulting in improved classification accuracy. The convolutional LSTM model, a novel integration of CNNs and LSTM networks, was used to predict ADHD based on anatomical data from the brain. The results showed that only the VGG-16, CNN, and convolutional LSTM models exhibited superior accuracy. Ensemble learning was used to create an ensemble model, with convolutional LSTM having 93% accuracy, CNN having 95% accuracy, and ensemble learning having 97% accuracy. Overall, ensemble learning had the highest accuracy, indicating the need for a model that can detect ADHD with better accuracy.