A team of researchers from Russia and Israel applied a new algorithm to classify the severity of autistic personality traits by examining the subjects’ brain activity. The article ‘Summary: Classification of Autistic Characteristics by Brain Activity Recoded by fNIRS Using ε Complexity Coefficients’ is published in the Journal of Autism and Developmental Disorders.
In addition to conventional tests and observations, doctors are increasingly using imaging techniques to diagnose autism and other mental disorders. Such diagnostic methods are not only more objective, but also often reveal the presence of a disorder in cases where the doctor does not have sufficient behavioral data, e.g. B. when the patient is young.
An important task in developing diagnostic methods is choosing an algorithm that can identify certain brain activity patterns. Since brain cells generate many electrical impulses per second, the raw data is often insufficient to draw conclusions. The data must first be processed.
A team of researchers from Russia and Israel examined one of these algorithms. The experiment included 26 healthy volunteers, although 5 were excluded from the final sample due to noisy signals. First, participants completed the autism spectrum quotient and were divided into two groups based on the test result: those with strong autistic characteristics and those with weak autistic characteristics.
The participants then performed an interpersonal synchronized movement task: each subject was asked to move their right hand in synchronization with the researcher’s for a few minutes while their brain activity was recorded. Interpersonal synchronization tasks are often used in diagnosing Autism Spectrum Disorders because people with ASD have difficulty coordinating joint actions.
Instead of conventional MRI or EEG, the researchers opted for functional near-infrared spectroscopy (fNIRS) to record the subjects’ brain activity.
To analyze the brain activity data, the researchers used ε complexity coefficients. This relatively new mathematical approach enabled researchers to extract meaningful information from complex and noisy patterns. Based on the data processed in this way, the researchers used classic classification methods to divide the subjects into groups depending on the characteristics of brain activity when performing the synchronization task.
When experimenting with these algorithms, scientists were able to achieve a prediction accuracy of greater than 90%: in 9 out of 10 cases, the assessment of the severity of autistic characteristics in subjects with neuroimaging matched the results of the questionnaire that participants completed at the beginning.
The new technology can be used as a diagnostic tool for Autism Spectrum Disorders as it is more accessible and convenient when working with people with ASD compared to fMRI.
In addition, this study has for the first time successfully applied the ε complexity theory to decode data recorded with fNIRS. This opens up the possibility of using the new algorithm in other studies with fNIRS technology.
In our study we have examined the results of Dr. Boris Darkhovsky used ε-complexity method to develop an algorithm to classify patients based on fNIRS recordings of brain activity. The resulting model-free technology for time series analysis can be used in cases in which the requirements of traditional analysis methods are violated – for example when working with significantly transient ECG and EEG signals. Therefore, this technology can be used to study other mental disorders and characteristics whose patterns appear in the data.
Yury Dubnov, lecturer at the Faculty of Computer Science
See press release