Autism Spectrum Disorder detection framework for children based on federated learning integrated CNN-LSTM

Area of Science:

• Neurodevelopmental Disorders
• Machine Learning in Healthcare
• Pediatric Health Informatics

Background:

• Autism Spectrum Disorder (ASD) incidence is rising, necessitating advanced detection and monitoring tools.
• Existing machine learning frameworks for ASD detection offer limited patient health management services.
• Genetics and environmental factors contribute to ASD, impacting children's communication, behavior, and social skills.

Purpose of the Study:

• To present a complex and efficient ASD framework to enhance existing ASD detection and patient monitoring.
• To introduce the Federated Learning-enabled CNN-LSTM (FCNN-LSTM) scheme for improved ASD detection in children.
• To integrate multimodal datasets and IoT applications for optimal ASD patient learning and efficiency.

Main Methods:

• Developed a Federated Learning-enabled CNN-LSTM (FCNN-LSTM) scheme in a distributed computing environment.
• Utilized multimodal datasets (ASQ, CSBS, PEDS, M-CHAT, STAT) trained and validated across connected laboratories.
• Implemented Advanced Standard Encryption (AES) for secure data handling and integrated IoT applications for patient efficiency.

Main Results:

• The FCNN-LSTM scheme achieved approximately 99% accuracy in ASD detection.
• Demonstrated superior performance compared to existing ASD detection frameworks.
• Enabled secure, distributed training and validation of multimodal ASD datasets.

Conclusions:

• The proposed FCNN-LSTM framework significantly enhances ASD detection accuracy in children.
• The integrated approach improves patient monitoring and learning efficiency through IoT applications.
• Federated learning with robust security mechanisms provides a scalable and effective solution for ASD research.