A neural network model explaining autism

A neural network model developed to understand how the brain controls normal behaviours has been adapted to explain how brain mechanisms may break down to give rise to the behavioural symptoms of autism.

Autism is a neurodevelopmental disorder which occurs in about 5 in 10,000 live births. The key symptoms of the condition are impaired  social interactions, deficiencies in verbal and non-verbal communication and inconsistency in cognitive abilities. 

Although the cause of autism is unclear, the cerebellar, limbic and cortical systems have been implicated in the condition, and it has been proposed that the condition occurs as a result of disturbances in such a distributed circuit. A small number of post-mortem studies have revealed abnormal cerebellar cytoarchitechtonics, with decreased numbers of atypically shaped and abnormally connected Purkinje cells, but neroimaging studies have provided conflicting evidence.   

In a paper published in the July issue of Psychological Review, entitled Neural Dynamics of Autistic Behaviours: Cognitive, Emotional and Timing Substrates, Professor Stephen Grossberg, of the Department of Cognitive and Neural Systems at Boston University, and Dr. Don Seidman, a paediatrician in the DuPage Medical Group at Elmhurst, Illinois,  propose the Imbalanced Spectrally Timed Adaptive Resonance Theory (iSTART) to explain the cognitive, emotional and motor symptoms of autism.The iSTART model is actually a combination of three models. One of these, the Adaptive Resonance Theory (ART), proposes that object recognition occurs when percepts entering the sensory systems match learned expectations of objects.

“When a match occurs,” explains Grossberg, “the system locks into a resonant state that drives how we learn to recognize things; hence the term adaptive resonance.”

According to ART, the degree of matching required for resonance, and hence attention, to take place, depends on what the researchers have called vigilance parameters. Low vigilance allows for a loose match and enables the learning of broad and abstract recognition, such as a face, whereas high vigilance requires closer matches, allowing the learning of more specific concrete categories, such as, for example, a profile of a specific face. 

“The iSTART model proposes that autistics have their vigilance set at very high levels, and that this hypervigilance is what causes the problems in cognition, attention and learning symptomatic of the condition,” says Grossberg. “It is suggested that interactions in the thalamo-cortical-hippocampal system, and others, which are normally responsible for these processes, may be defective in autism.” 

The second component of iSTART is the Cognitive-Emotional-Motor (CogEM) model, which extends the ART model to the learning of associations between external objects and the emotional states which give them value. The emotions associated with objects are satisfied by the activation of motivational pathways (the thalamo-cortical-amygdala system)  underlying the actions for acquisition and manipulation of the objects.

Under normal circumstances, the arousal levels required for the activation of these circutis is set at an intermediate level. Abnormal emotional reactions are produced when there is over- or under-arousal of this system. The iSTART model proposes that individuals with autism experience under-aroused emotional depression which helps explain symptoms like reduced emotional expression as well as emotional outbursts.”

The third component of the Grossberg-Seidman model is Spectral Timing, which is used to explain how the brain times its responses to obtain rewards. Spectral Timing predicts how the brain distinguishes between expected and non-expected non-occurences of rewards. Unexpected non-occurrence of reward can trigger shifts in attention; iSTART proposes that autistics experience failures in spectral timing, leading to context-inappropriate behaviours which prevent reward.

“iSTART depicts how autistic symptoms may arise from breakdowns in normal brain processes,” says Grossberg. “[It] opens up a wide range of possible new experiments…[and] make it easier for scientists studying normal behavior to connect their work to autism research.”