RESNET34 with Synchrosqueezing Transform for ADHD Disorder Detection Using EEG Signals

Abstract

It is vital for the development of attention deficit hyperactivity disorders (ADHDs) diagnostic tools that help healthcare practitioners to have automated recognition of ADHDs. Recently, the detection of ADHD using EEG data has attracted a lot of interest due to its apparent rapidity in diagnosis and treatment. This is because the disease seems to be caused by hyperactivity and inattention. This paper presents the method for diagnosing ADHDs based on EEG data, with the nonlinear characteristics being retrieved via the Synchrosqueezing Transform (ST). The ST filter banks are used to break down the 16 channels of EEG signal information into the ideal number of time-frequency subbands. At each stage of the decomposition process, the distinctive feature vectors are assessed using CNN as well as the proposed nonlinear fractal dimensions approach. Using ResNet34, it was discovered that a modified CNN classifier that used a 10-fold cross-validation approach was an efficient classifier for differentiating between normal people and those with ADHD. The suggested method was able to successfully categorize ADHD and normal participants with the best accuracy, as shown by several performance indicators. The statistical study demonstrated that the CNN and the proposed nonlinear feature estimation techniques produce possible characteristics that are acceptable for the automated identification of ADHD. These features may be categorized with good accuracy, sensitivity and specificity. The technique that has been presented is capable of correctly differentiating between participants who have ADHD and those who do not have ADHD with an ultimate precision of 99.5%.