Page 223 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 7 Early detection and diagnosis using deep learning 213

of knowledge of the mechanism of ASD, conclusive results have
yet to be made. This gap of information may be bridged by the
data-driven mind-set of DL that cultivates clear markers of
intrinsic nature from vast data fields, giving doctors the opportu-
nity predict ASD (Fig. 7.7).

3.4 Attention deficit hyperactivity disorder
Attention deficit hyperactivity disorder (ADHD) is one of the
most extensive neuropsychiatric disorders that arise during
childhood and adolescence. Its major symptoms include inat-
tention, hyperactivity, and impulsivity. Depending on the
advancement of these symptoms, the disorder itself can be cate-
gorized into two different subtypes. Patients who are diagnosed
with both subtypes are characterized by significant attention
problems. However, only patients with ADHD-C are additionally
characterized by impulsivity/hyperactivity. A variety of models
provide an understanding of the underlying etiology, and the
most widely accepted theories relate to dopaminergic process-
ing and changes in prefrontal cortex functioning. Currently,
the standard procedure for diagnosing ADHD consists of
conducting clinical interviews, symptom questionnaires with
multiple raters, neuropsychological testing, and the elimination
of other underlying causes that may also bring about the
observed symptoms. Over the past few years, a considerable
amount of effort has been put into research to assess the useful-
ness of neurophysiological and functional imaging data to aid
the existing process. Nearly, all models that have been built to
tackle this issue focus on a one-to-one relationship between
neural signals and the phenotypic expression of the disorder.
The advent of ML approaches, specifically DL, offers a great
number of new opportunities for diagnostic purposes. DL
provides a platform computational model to understand and uti-
lize the representations of data with multiple levels of abstraction
using all the information that the data set has to offer. This
method is particularly advantageous in comparison with the
conventional ML approaches, which hardly make use of the avail-
able resources. DL can also be used for the classification of EEG
(electroencephalogram) data. For this purpose, preproposed
architectures are enforced on very specific data sets, which
consequently gives birth to a new and more advanced DL archi-
tecture. A compact convolutional network for EEG-based braine
computer interfaces has the function of decoding brain states.