Bimodal Feature Analysis with Deep Learning for Autism Spectrum Disorder Detection

Area of Science:

• Neurodevelopmental Disorders
• Artificial Intelligence in Healthcare
• Biomechanical Analysis

Background:

• Autism Spectrum Disorder (ASD) affects approximately 1 in 100 children globally.
• Current diagnostic methods for ASD can be invasive and subjective.
• Objective and non-invasive tools are needed for early ASD detection.

Purpose of the Study:

• To introduce a novel Deep Neural Network (DNN) for identifying ASD in children.
• To utilize gait analysis from video recordings as a non-invasive assessment method.
• To achieve high accuracy in ASD detection through advanced AI techniques.

Main Methods:

• A bimodal deep learning approach was developed, combining two Convolutional Neural Networks (CNNs).
• Features were extracted from video frames capturing children's walking patterns.
• A novel method concatenates features from two CNNs for input into fully connected layers for classification.

Main Results:

• The study presents a highly accurate, non-invasive method for ASD detection.
• The combined gait analysis and deep learning approach shows significant potential.
• The bimodal DNN effectively processes gait features for binary classification.

Conclusions:

• Gait analysis combined with deep learning offers a promising avenue for objective ASD assessment.
• The developed DNN model demonstrates efficacy in identifying ASD in children.
• This innovative approach could enhance early diagnosis and intervention for ASD.