Tags

, , ,

ResearchBlogging.org

Courtesy: Wikicommons

This is a science story of an inquisitive high school student from the Philippine Science High School who by asking the help of the right people [1], may have found a route to differentiate children with mild autism using Electroencephalogram (EEG) traces.

Writing in a soon to be published article in the International Journal of Bifurcation and Chaos, Lance Co Ting Keh, Anna Marie Chupungco, and Dr. Jose Perico  Esguerra, used three nonlinear time series analysis method to distinguish  between the EEGs of normal children, children with mild autism, and children with severe autism [2]. Although this is not the first attempt to differentiate based on EEG tracings, the methods they used offer another route for quantifying the differences.

The EEG tracings were obtained from 15 children whose age ranges from 3 to 7 years old.  These tracings were clustered according to the severity of autism of the children based on the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, Text Revision (DSM IV-TR) criteria.  The children either have severe, mild, or no autism. Each cluster consisted of 5 children. An example of a digitized trace is shown in fig 1.

Fig 1. EEG trace of one cohort. Courtesy: L. Co Ting Keh, A. Chupungco, J. Esguerra, International Journal of Bifurcation and Chaos, 2012 (preprint).

The EEG tracings  were from the sleep state, and that the focus of the analysis were done on the parietal-occipital electrode pair. The researchers based their decision to concentrate on the prefrontal and the parietal part of the brain from an earlier study by the National Institutes of Health, USA.  In that study, it was found that these parts were less likely to function synchronously in children with autism.  The parietal and occipital lobes are responsible for speech, taste, reading, integrating sensory input to form a single perception, and visual perception.

The researchers used three nonlinear time series analysis –  The Lempel-Ziv complexity, covariance complexity and prediction error. All these tests measure the randomness of the data in time.

Lempel-Ziv complexity

Fig 2. Lempel-Ziv Complexity values. Points (circles, diamonds, squares) are calculated values from EEG trace of each cohort in the study. Gaussian curves are centered at the mean values of each group,  with waists corresponding to their respective variances. Big circles are guide to show that there is clustering between values and groups. Data courtesy of  L. Co Ting Keh, A. Chupungco, J. Esguerra.

The Lempel-Ziv complexity is a way to quantify the extent of randomness in different systems. Higher Lempel-Ziv complexity values correspond to a more random data set. 

According to the researchers, “Children with mild autism have tracings that are more random and disordered. On the other hand, the lower Lempel-Ziv complexity values of the normal and severe clusters suggest that these children have more orderly EEG readings.”

There is no significant difference between the normal and severe clusters. Notice from the plots that there is a large overlap between the normal and the severe clusters as shown in figure 2.

Covariance Complexity

Fig 3. Covariance Complexity values.  Points (circles, diamonds, squares) are calculated values from EEG trace of each cohort in the study. Gaussian curves are centered at the mean values of each group,  with waists corresponding to their respective variances. Big circles are guide to show that there is clustering between values and groups. Data courtesy of  L. Co Ting Keh, A. Chupungco, J. Esguerra.

The covariance complexity measures the complexity of a signal. Its value is from 0 to 1. A value closer to 1 indicates that the signal is more random.

“There is once more a large overlap between the intervals of values for children with severe autism and for normal children, but the interval of values for the children with mild autism is again distinct from the two”, according to the researchers.

From the mean and the variance values, the normal and severe cluster cannot be distinguished from each other while the mild cluster value  slightly overlaps with the two (Fig 3).

Prediction error

Fig 4. Prediction error. Points (circles, diamonds, squares) are calculated values from EEG trace of each cohort in the study. Gaussian curves are centered at the mean values of each group, with waists corresponding to twice their respective variances. Big circles are guide to show that there is clustering between values and groups. Data courtesy of L. Co Ting Keh, A. Chupungco, J. Esguerra.

The researchers found out that the EEG readings of children with mild autism are harder to predict, while those of normal and severely autistic children are more predictable (Fig 4). The Prediction error measures how future values of a sequence can be predicted from previous values accurately. Sequences that are more ordered are more predictable.

As with the previous tests, the normal and severe cluster values overlap significantly such that one cannot distinguish one cluster from the other. However for the mild cluster, the prediction error mean value with its variance only slightly overlaps with the two clusters.

Outlook

The researchers however stressed that there are grounds for caution. For one, the sample size is  small and that more EEG samples are needed.  The distinguishability between the clusters maybe larger, smaller, or even insignificant if more EEG tracings are analyzed.

Secondly,  they don’t know exactly why the methods seem to work. None of the researchers have medical expertise on autism.

One thing that this study proves is that through the use of nonlinear time series analysis, randomness in the EEG tracings can be quantified.

“A quantification that could lead to a more definitive method of diagnosis,” according to the researchers.

__________________

[1] Other than the authors, there are a few people who helped in the research: Alfonso Albano of Bry Mawr College who is an expert on time series analysis; Mayette Valencia of the University of Santo Tomas, an expert in neurology and autism,  helped start the research and provided the 15 EEG tracings;   Mr. Adrian Solis;  Ms. Rose Butaran; Mr. Jason Alcarez.

[2] Co Ting Keh, L., Chupungco, A., & Esguerra, J. (2012). NONLINEAR TIME SERIES ANALYSIS OF ELECTROENCEPHALOGRAM TRACINGS OF CHILDREN WITH AUTISM International Journal of Bifurcation and Chaos DOI: 10.1142/S0218127412500447

Advertisements