A multi-task learning model with reinforcement optimization for ASD comorbidity discrimination | JoVE Visualize

Area of Science:

Neuroscience
Computational Psychiatry
Biomedical Engineering

Background:

Discriminating comorbidities in autism spectrum disorder (ASD) is challenging due to overlapping behaviors and insufficient training data.
Electroencephalogram (EEG) analysis shows promise but faces limitations in differentiating conditions like ASD, ADHD, and ASD+ADHD.
Current methods struggle with feature overlap between ASD and ASD+ADHD, and the need for extensive, reliable datasets.

Purpose of the Study:

To develop and validate a novel method for discriminating comorbidities in ASD using EEG data.
To address the limitations of existing methods, specifically feature overlap and data insufficiency.
To enhance the accuracy and reliability of diagnosing conditions such as ASD+ADHD.

Main Methods:

A multi-task learning (MLT) framework was employed, utilizing a Convolutional Neural Network (CNN) for discrimination.
An auxiliary task (ASD vs. Typically Developed (TD)) was integrated to aid the primary discrimination task (ASD vs. ASD+ADHD).
A reinforcement learning (RL) algorithm was incorporated for "reinforcement optimization" (RO) to enhance feature extraction and fusion.

Main Results:

The proposed RO-MLT method demonstrated superior performance across all indicators compared to state-of-the-art approaches.
Experimental results on resting-state EEG data from 150 children (ASD, ASD+ADHD, TD) confirmed RO-MLT's effectiveness.
Ablation studies revealed that the combined MLT and RL approach yielded an 11.07% performance boost, exceeding the sum of individual technique contributions.

Conclusions:

The RO-MLT method offers a significant advancement in the discrimination of comorbidities in ASD, particularly ASD+ADHD, using EEG.
This approach effectively mitigates challenges related to EEG feature overlap and data limitations in diagnostic modeling.
The findings suggest that RO-MLT holds substantial potential for improving clinical diagnosis and personalized treatment strategies for individuals with ASD and co-occurring conditions.