Diagnóstico do TDAH: método baseado em aprendizado de máquina e neuroimagens

Abstract

Attention deficit hyperactivity disorder (ADHD) is one of the most common and complex neurobiological disorders. Patients suffer from a persistent pattern of inattention and/or hyperactivity-impulsivity that causes harm in various environments, including academic, personal and professional life. It is also considered one of the most heterogeneous disorders and can co-occur with other disorders. There are three presentations: predominantly inattentive, predominantly hyperactive-impulsive and the combined subtype. The traditional method of diagnosis is based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM). As these are subjective measurements that are subject to inconsistencies, they raise suspicion about the credibility of the diagnosis. In addition, patients suffer from delays in the examination process. This problem has encouraged the search for faster diagnostic options with fewer measurement errors. Evidence of brain alterations has already been implicated in rs-fMRI (resting-state functional magnetic resonance imaging) and SPECT (single-photon emission computed tomography) images of individuals with the condition. The use of machine learning algorithms in conjunction with neuroimaging has proven to be a promising combination in classifying the disorder. However, there is no consensus on the best approaches to identify the features of the disorder and how all this could become a useful tool for specialists. Therefore, the main objective of this work is to propose an objective and efficient diagnostic method for ADHD. One of the distinguishing features is the unprecedented combination of brain SPECT and rs-fMRI modalities. To this end, machine learning methods, regions of interest (ROIs), brain networks and functional connectivity were analyzed in this work. It was found that the ROIs/networks of executive control, the limbic system and the cerebellar system were the most discriminable. The Support Vector Machine (SVM) and the classifier committee favored the classification of the disorder. The performance of the model created with rsfMRI (64.89%) outperformed the best result of the global ADHD-200 competition (61.54%). This is a diagnostic method that can be implemented in software and is expected to provide more reliable diagnoses by providing objective evidence together with the patient's clinical history.