Department of Mechanical and Mechatronic Engineering

Permanent URI for this community

https://scholar.sun.ac.za/handle/10019.1/109

Browse

Browsing Department of Mechanical and Mechatronic Engineering by browse.metadata.advisor "Aldrich, C."

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Early detection of risk of autism spectrum disorder based on recurrence quantification analysis of electroencephalographic signals
    (Stellenbosch : Stellenbosch University, 2016-03) Heunis, Tosca-Marie; Nieuwoudt, Martin; De Vries, Petrus J.; Aldrich, C.; Stellenbosch University. Faculty of Engineering. Dept. of Mechanical and Mechatronic Engineering.
    ENGLISH ABSTRACT: Background: Autism spectrum disorder (ASD) is a neurodevelopmental disorder (NDD) with a prevalence of 1–2%. In low resource environments, in particular, early identification and diagnosis is a significant challenge. There is a great demand for ‘language-free, culturally fair’ low-cost screening tools for ASD risk that do not require highly-trained professionals. There has been growing interest in electroencephalography (EEG) as an investigational tool for biomarker development in ASD and NDDs. One of the key challenges lies in the identification of appropriate multivariate, next-generation analytical methodologies that can characterise the complex, nonlinear dynamics of neural networks in the brain. Objectives: The primary objective was to develop a biomarker to differentiate ASD from typically developing (TD) individuals. The secondary objectives included evaluating the ability of the biomarker to discriminate between ASD and non-ASD within Tuberous Sclerosis Complex (TSC), a genetic disorder, and to distinguish non-syndromal and syndromal ASD. Method: Recurrence quantification analysis (RQA) of rsEEG was explored as a potential biomarker for ASD. The final methodology comprised the analysis of continuous five second segments of rsEEG with independent component analysis ocular artefact correction, multivariate time series embedding, principal component analysis dimensionality reduction, RQA feature extraction, and classification of statistically significant features using linear discriminant analysis (LDA), multilayer perceptron (MLP) neural network and support vector machine (SVM) classifiers. A clinical scenario of diagnosing an unseen subject was simulated using a leave-one-subject-out classification approach. Results: In a sample of 7 ASD and 5 TD subjects, aged 8 – 17 years, with analysis of 12 best segments, the proof-of-principle study showed 83.3% accuracy, 85.7% sensitivity and 80.0% specificity with a LDA classifier. A replication study (study 1) was performed in an age-matched sample of 7 ASD and 7 TD subjects, aged 2 – 6 years, comprising a total of 666 segments. Classification with a nonlinear SVM classifier showed 92.9% accuracy, 100.0% sensitivity and 85.7% specificity. The biomarker was explored further in an age and intellectual ability matched sample of 5 TSC+ASD and 5 TSC-ASD subjects, aged 0 – 12 years, comprising 1 202 segments (study 2). The nonlinear MLP classifier achieved 90.0% accuracy, 80.0% sensitivity and 100.0% specificity. The ability of the biomarker to distinguish non-syndromal and syndromal ASD was evaluated in an age and intellectual ability matched sample of 6 ASD and 6 TSC+ASD subjects, aged 2 – 6 years, with analysis of 832 segments (study 3). Using the MLP and SVM classifiers, 100% accuracy, sensitivity and specificity was achieved. Conclusions: RQA of rsEEG was an accurate classifier of ASD under a range of clinical conditions, suggesting potential of this approach for global screening in ASD. Validation of this biomarker in a large and well-matched sample is required. Age, gender, intellectual ability, socio-economic status, comorbidity, medication use, eyes-open versus eyes-closed condition, the number and location of electrodes, and test-retest reliability are all important factors to consider in the evaluation and development of biomarkers for ASD and related neurodevelopmental disabilities.