Diagnosis of attention deficit and hyperactivity disorder (ADHD) using Hidden Markov Models

Attention deficit hyperactivity disorder (ADHD), most often present in childhood, may persist in adult life, hampering personal development. However, ADHD diagnosis is a real challenge since it highly depends on the clinical observation of the patient, the parental and scholar information, and the specialist expertise. Despite demanded objective diagnosis aids from biosignals, the physiological biomarkers lack robustness and significance under the non-stationary and non-linear electroencephalographic dynamics. Therefore, this work presents a supported diagnosis methodology for ADHD from the dynamic characterization of EEG based on hidden Markov models (HMM) and probability product kernels (PPK). Relying on the impulsivity symptom, the proposed approach trains an HMM for each subject from EEG signals at failing rewarded inhibition tasks. Then, PPK measures the similarity between subjects through the inner product between their trained HMMs. Therefore, a support vector machine supports ADHD diagnosis as a classification task using PPK as the inner product operator. Results in a real EEG dataset evidence that the proposed approach achieves an 90.0% accuracy rate, outperforming log-likelihood features as baseline HMM-based features. Besides, achieving such an accuracy at the highest reward level supports that ADHD patients seem to be particularly sensitive to the reward presence when they execute specific tasks.

Automated summary

ADHD diagnosis relies on clinical observation and related information, proposing a method to support ADHD diagnosis through EEG characterization. By training HMM and using PPK to measure similarity between patients, support vector machine is used as a diagnostic tool for classification tasks.