Page 222 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

212 Chapter 7 Early detection and diagnosis using deep learning

3.3 Autism spectrum disorder
Standardized tests that give out conclusive results to a patient
for their respective treatment have been the gold standard for
curing a person. However, they fail to meet the requirements
for a vast array of mental disorders. Autism spectrum disorder
(ASD) is a neurodevelopment disorder that is near impossible
to single out in its early stages, especially in the case of children.
Unlike biological tests that are quantitative, mental disorders rely
primarily on behavioral symptomology and are hence vulnerable
to misdiagnosis. The issue lies with a multitude of symptoms
shown by mental disorders that closely or entirely resemble
symptoms of other disorders. ASD is a lifelong disorder that
causes various impediments to everyday life, and with over 1%
of children showing signs, it is imperative that it is treated head
on in its early stages.
ASD shows unique behavioral symptomology and leads to the
development of distinguishable repetitive actions. When these
are monitored along with the analysis of brain data, possible
quantitative means to identifying ASD may be discovered. This
is where DL comes in. With mass analysis of information from
population demographics and brain scans that show ASD, bio-
markers could be developed that shows researchers reliable signs
of development of mental disorders. Hence, moving forward with
this task requires advancements in scalable DL infrastructure.
The algorithms that analyze data in DL can avoid strict catego-
rization due to their unprecedented nature of unsupervised
learning. DL is able to make its own categories that rely on min-
imal human interaction, resulting in the extraction of data that
the machine feels are relevant. This allows for classification at a
more intrinsic level due to unsupervised access into clinical
data sets. This vastly differs from supervised learning. Supervised
learning methods involve associating the learner to a specific
label or tag using which the classifier sorts data. With this step
removed, unsupervised learners have the capacity to explore
demographics, brain analysis, and big data sets and form classi-
fiers of their own. This results in biomarkers for all types of
mental disorders that are less susceptible to errors.
Research into ASD, thus far using MRI, shows clear signs both
structurally and functionally in brain imaging of patients with
ASD. MRI scans reveal that individuals exhibiting ASD show an
age-related brain growth abnormality in the frontal section,
making it imperative that ASD is identified in its initial stages.
From this, we have been able to identify, for the most part,
patients showing ASD; however, due to the complexity and lack