A face image classification method of autistic children based on the two-phase transfer learning.

Area of Science:

• Computer Science
• Medical Imaging
• Artificial Intelligence

Background:

• Autism spectrum disorder (ASD) significantly impacts children's lives, necessitating early detection and intervention.
• Facial feature analysis via mobile devices offers a potential avenue for screening potential ASD cases.
• Existing deep learning models for mobile-based ASD screening have limitations in performance (AUC) and image size suitability.

Purpose of the Study:

• To propose a deep transfer learning method for improved ASD screening using mobile phone facial images.
• To enhance the classification performance (AUC) of mobile-suitable deep learning models.
• To address the challenge of large input image sizes for mobile terminal compatibility.

Main Methods:

• A novel deep transfer learning approach employing a two-phase transfer learning mode.
• Integration of multiple classifiers (MobileNetV2, MobileNetV3-Large) using a multi-classifier integration mode.
• Development of a multi-classifier integrating calculation method for final classification.

Main Results:

• Two-phase transfer learning significantly improved the classification performance of MobileNetV2 and MobileNetV3-Large compared to one-phase.
• The integrated classifier outperformed individual participating classifiers.
• Achieved 90.5% accuracy and 96.32% AUC, a 3.51% AUC improvement over previous studies.

Conclusions:

• The proposed two-phase transfer learning and multi-classifier integration method effectively enhances ASD screening via mobile facial image analysis.
• This approach offers a promising, high-performance solution for early ASD detection suitable for mobile applications.
• The method demonstrates superior classification performance, particularly in AUC, compared to existing techniques.