AI Research
AI-Powered Autism Detection System
Deep Learning Research Prototype
Overview
A deep-learning system exploring Autism Spectrum Disorder (ASD) classification from facial images. Compared CNN, VGG16, VGG19, EfficientNetB3 and EfficientNetB4 architectures, with EfficientNet achieving the strongest results in the research evaluation.
Problem or Opportunity
Research contexts require rigorous comparison of deep learning architectures for image classification tasks, with reproducible evaluation and responsible presentation of results.
Proposed Solution
The project implements multiple model architectures, evaluation pipelines and a full-stack web interface for demonstration, with clear limitations around clinical applicability.
Key Capabilities
- Multiple CNN architecture comparison
- Image preprocessing with OpenCV
- Model training and evaluation pipelines
- PDF report generation
- Full-stack demonstration interface
Architecture and Workflow
TensorFlow/Keras models are trained offline with evaluated checkpoints. A React frontend communicates with backend services for inference demonstration and report generation.
Security Considerations
- Research data handled with appropriate access controls
- Clear disclaimers on non-clinical use
- No presentation as validated medical software
Challenges and Decisions
- Responsible communication of research limitations
- Managing class imbalance and dataset constraints
- Comparing architectures fairly under consistent conditions
Want to Build Something Similar?
Tell us about your project and we will outline a practical delivery approach.