Improved Transfer-Learning-Based Facial Recognition Framework to Detect Autistic Children at an Early Stage

Area of Science:

• Computer Science
• Neuroscience
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) is a neurodevelopmental condition impacting social and communication skills.
• Early identification of ASD is crucial for timely and effective intervention strategies.
• Facial features and eye contact analysis offer potential biomarkers for ASD detection.

Purpose of the Study:

• To develop an advanced transfer-learning framework for precise early-stage autism face recognition in children.
• To evaluate the efficacy of various machine learning and deep learning models for ASD identification.

Main Methods:

• Collected a dataset of children's facial images from Kaggle.
• Applied and compared multiple machine learning, deep learning, and transfer-learning models.
• Utilized an improved MobileNet-V1 model for face recognition and k-means clustering for sub-typing.

Main Results:

• The enhanced MobileNet-V1 model achieved 90.67% accuracy in ASD identification.
• This model demonstrated superior performance with the lowest fall-out and miss rates compared to other methods.
• K-means clustering with MobileNet-V1 identified autism sub-types with 92.10% accuracy for k=2.

Conclusions:

• The proposed transfer-learning framework, particularly the improved MobileNet-V1, shows significant promise for early ASD detection in children.
• This AI-driven approach can assist clinicians in more explicit and timely diagnosis of autism.
• Further research could refine the model for broader clinical application and sub-type differentiation.