Brain age prediction model based on electroencephalogram signal and its application in children with autism spectrum disorders.

BACKGROUND

There is a lack of objective biomarkers for brain developmental abnormalities of autism spectrum disorder (ASD). We used EEG and deep learning to conduct a brain aging study in ASD.

METHODS

(1) A total of 659 healthy children and 98 ASD patients were retrospectively recruited. (2) An Auto-EEG-Brain AGE prediction model based on the Gate Recurrent Unit (GRU) neural network method was constructed. (3) Using the constructed model, we evaluated the difference between the brain age of ASD and that of healthy controls, and assessed the feasibility in the clinical assessment of ASD.

RESULTS

(1) The correlation coefficient (-value) of the model exceeded 0.8 at the whole-brain level, with the highest value reaching 0.91. (2) -values of the ASD group amounted to 0.76 at the level of the whole brain and ranged from 0.66 to 0.7 at the level of the sub-brain regions. The mean value of the brain age gap estimate (Brain AGE) in the whole brain is 0.76 years; in the sub-brain model, was 0.64-1.18 years.

CONCLUSION

We constructed the EEG-Brain AGE prediction model, which can identify an individual's brain development and be used as a biomarker for the brain development assessment in ASD.