نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Spotting Autism Early in Young Kids Through Brain Wave Patterns and Smart Computer Tools

Subject:

Autism and Brain Mapping

Presenter Author:
Name:	Ashkan Alimohammadi

Author:
Name 1:	Ashkan Alimohammadi - Corresponding - Presenter
Affiliation 1:	Students' Scientific Research Center (SSRC), Tehran University of Medical Sciences, Tehran, Iran

Name 2:	Mohammad Erfan Mohammaddost Golrodbari
Affiliation 2:	Students' Scientific Research Center (SSRC), Tehran University of Medical Sciences, Tehran, Iran

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Autism Spectrum Disorder (ASD) affects how kids connect with people and their surroundings. It often appears as trouble with social interaction and repetitive behaviors or interests. Identifying it early increases the chances of effective support and therapies that can significantly impact a child’s life. Currently, most diagnoses come from observing a child’s behavior over time, usually after they turn three. We aimed to find clues earlier by studying the brain's tiny electrical signals using EEG recordings. Specifically, we focused on some less-common “nonlinear” ways these signals fluctuate and applied modern machine-learning techniques to detect patterns that might indicate autism in very young children.

Methods:

We used resting-state EEG recordings from children aged 2 to 6 years. Half had been diagnosed with ASD while the other half were typically developing, serving as a control group. To clean up the data, we first removed artifacts like eye blinks and small muscle movements using a method called Independent Component Analysis (ICA). We then filtered the signals to focus on frequencies between 1 and 45 Hz. From each recording, we extracted several nonlinear measures: Sample Entropy (which shows how unpredictable the signal is), Lempel-Ziv Complexity (which gauges how much new information appears), Detrended Fluctuation Analysis (which examines long-distance patterns over time), and the Hurst Exponent. We also included the traditional power levels in the usual frequency bands—delta, theta, alpha, and beta. All these measurements together created a detailed picture of each child’s brain activity. We then used three different machine-learning methods: Support Vector Machine, Random Forest, and a Convolutional Neural Network. The models learned to differentiate between the two groups, and we carefully checked their performance using 10-fold cross-validation, tracking accuracy, sensitivity, specificity, and the AUC score.

Results:

The brain waves revealed interesting patterns—especially in the frontal and temporal areas. Kids with ASD typically showed lower Sample Entropy and different long-range scaling in the Detrended Fluctuation Analysis compared to the control group (p < 0.05). Among our three models, the Convolutional Neural Network performed the best, achieving around 90% accuracy. It seemed to detect subtle patterns spread across both time and different parts of the brain—details the other models missed. When we examined which features were most significant using SHAP analysis, changes in frontal alpha and beta power, along with an overall drop in signal complexity, emerged as the clearest indicators separating autism from typical development.

Conclusion:

Our findings are promising: these nonlinear methods for analyzing EEG signals, combined with powerful machine-learning tools, can identify reliable early signs of autism that standard checks might overlook. This could result in a simple, affordable, and completely non-invasive screening method that doctors and parents could use alongside regular developmental assessments. Starting support earlier could greatly benefit families. In the next phase, we hope to track kids over longer periods to see if these patterns remain consistent, and we are also considering combining EEG with other gentle imaging techniques like near-infrared spectroscopy to strengthen our approach.

Keywords:

Autism Spectrum Disorder, EEG, Nonlinear Brain Signals, Machine Learning, Early Screening, Brain Activity Patterns