Article Preview
TopIntroduction
At present, biomedical intelligence for clinical data analysis is readily used for health informatics research and applications. With the technology advancements, the novel computational methods support to process large amounts of health-care data. Biomedical imaging is used to derive features and identify patterns using various analysis techniques on clinical data for medical purposes. Thus, the diagnosis process of medical untidiness can be supported by an appropriate selection and processing of image features. Recently, the use of biomedical imaging for the classification of psychiatric disorders has been considered for a precise diagnosis process.
Attention deficit hyperactivity disorder (ADHD) is a prevalent neurological disorder among children. There is a prospect of carrying the associated behavioural, communication and learning issues of such children into adulthood and comorbid with other neurological disorders (Meedeniya & Rubasinghe, 2020). ADHD can be described as the tenacity pattern of inattention, hyperactive or impulsivity that is notably higher than the corresponding development groups, as defined by the DSM-5 diagnostic criteria (American Psychiatric Association, 2013). At present, ADHD covers 5% of the entire child population causing impairments of their childhood, and 70% carries them into adulthood and a higher rate of comorbid with other neurological disorders (Polanczyk, de Lima, Horta, Biederman & Rohde, 2007). This is the main motivation of this study to propose a method for the early identification of ADHD using medical imaging. Generally, psychiatric disorders such as depression, anxiety, learning disorder, obsessive-compulsive disorder, conduct disorder, and other learning-related difficulties are some of the highly comorbid disorders with ADHD (De Silva, Dayarathna, Ariyarathne, Meedeniya & Jayarathna, 2019a; Meedeniya & Rubasinghe, 2020).
Among psychophysiological measures, electroencephalography (EEG) has shown the possibility of identifying the ADHD subtypes; inattentively, impulsivity and hyperactivity efficiently (Kaur, Arun, Singh & Kaur, 2018). However, the source reconstruction is not a well-posed problem and contain inherent limitations on spatial resolution. Thus, using EEG is not alone applicable in the deification process for a larger scale and combining heterogeneous sources (Abreu, Leal & Figueiredo, 2018). On the other hand, fMRI tests have been used widely providing effective identification process, and effectivity features than EEG by capturing the brain response in the presence of neural stimuli. Therefore, this study addresses the feature extraction process of fMRI data for ADHD classification.
fMRI data are used to identify the abnormal neural activities of ADHD subjects in the resting state of the selected brain regions (Metin et al., 2015). Hence, this paper focuses on the ADHD identification using fMRI data based on features such as fractional Amplitude of Low-Frequency Fluctuation (fALFF), Regional Homogeneity (ReHo) and seed-based correlation (SBC) to extract the regional activities in the brain. The proposed solution addresses Default Mode Networks (DMN) based brain regions to study the abnormal activities and extract features using the SBC technique as the main contribution. This study evaluates various features from fMRI image data by comparing the use of preprocessed and un-processed data in the classification process. The preprocessing of raw fMRI data and extraction of optimal features also contribute to the novelty of this study. The approach uses CNN as the learning model to derive an accurate model with fMRI data. This is developed as a prototype named ADHD-Care_v2 (ADHD-care, 2019). This is an extension of our previous study of ADHD diagnosis with eye movement data using a rule-based system (De Silva et al., 2019b).
The paper is structured as follows. Section 2 explains the theoretical approaches applied in the proposed solution along with the practical application of ADHD classification. Section 3 presents the design of the solution including the system workflow. Section 4 includes the explanation of the methodology supported by the comprehensive implementation. Then Section 5 comparatively evaluates the results and Section 6 concludes the paper.