Advancing ADHD Diagnosis via LSTM Models and Multimodal EEG Feature Optimization

The present study investigates the application of long short-term memory (LSTM) networks in electroencephalography (EEG)-based classification of attention deficit hyperactivity disorder (ADHD) under a novel paradigm. Attention deficit hyperactivity disorder (ADHD) is a prevalent neurodevelopmental disorder, characterised by inattention, hyperactivity and impulsivity. Conventional diagnostic methodologies are predicated on subjective clinical evaluations and are inherently constrained. At present, the diagnosis of ADHD relies heavily on physicians requesting that parents rate symptoms, a process that is too subjective and may lead to inaccurate diagnoses. The present study utilises objective neurobiological features in electroencephalography (EEG) signals with a view to enhancing diagnostic accuracy. A model based on long short-term memory (LSTM) networks and electroencephalogram (EEG) signals has been shown to provide a more objective and accurate method for the diagnosis of attention deficit hyperactivity disorder (ADHD). The experimental setup consists of a customised electroencephalogram (EEG) device and a video-guided task to elicit focused and unfocused states. Electroencephalogram (EEG) signals were processed to extract time- and frequency-domain features associated with attention and cognitive states, such as activity in the theta, alpha, and beta wave bands. Long Short-Term Memory (LSTM) networks have been shown to process time-series data and capture temporal correlations, rendering them well-suited for the analysis of electroencephalogram (EEG) signals over time. Kernel Principal Component Analysis (Kernel PCA) is an effective method of reducing the dimensionality of high-dimensional electroencephalogram (EEG) data. It extracts the nonlinear features of the data and is better able to maintain the global structure of the data. This improves the efficiency and accuracy of the model. Lasso regression (Lasso regression) and Kernel Principal Component Analysis (Kernel PCA) are used for feature selection and dimensionality reduction. The trained Long Short-Term Memory (LSTM) network model exhibits an Area Under the Curve (AUC) value of 0.96, signifying its high accuracy in differentiating attentional states. The findings of this study indicate that the integration of feature-extracted electroencephalogram (EEG) signals with deep learning models, particularly the long short-term memory network (LSTM), can substantially enhance the precision and dependability of attention deficit hyperactivity disorder (ADHD) diagnosis. Subsequent research endeavours will concentrate on augmenting the dataset, investigating enhanced feature extraction and dimensionality reduction techniques, and optimising the utilisation of deep learning and multimodal fusion to enhance classification performance.