🧠 AutismLens

AutismLens is a deep learning-based project focused on improving the early diagnosis of Autism Spectrum Disorder (ASD) using facial image analysis. This project integrates advanced CNN architectures, particularly ConvNeXt, with interpretability tools like Grad-CAM to provide accurate and explainable predictions. The system is designed to assist clinicians and researchers in identifying ASD-related patterns through facial features.

🎯 Goals

• Early, explainable detection of ASD through facial image recognition.
• Improve diagnostic accuracy and speed using advanced AI.
• Provide a simple and interpretable tool for researchers and clinicians.

🧩 How It Works

1. Upload a facial image of a child.
2. The model analyzes the image and detects whether the child has autism.
3. Grad-CAM highlights key facial features used in the prediction for better interpretability.
4. Results are displayed on the web interface.

📽 Demo

Watch the demo

📂 Datasets Used

1. Autism vs. Healthy Children
   📥 Kaggle Dataset – Autism Image Data

2. Neurodevelopmental Disorders (NDD)
   📥 Roboflow Dataset – Down Syndrome Facial Images

3. Other Sources (NDD Syndromes)
   Additional facial images for the following syndromes were obtained from various open-access sources:

   • 22q11.2 Deletion Syndrome
   • 22q11.2 Duplication Syndrome
   • Fragile X Syndrome
   • Williams-Beuren Syndrome
   • Cerebral Palsy Disorder

📂 Model Weights

The pre-trained model weights for AutismLens can be accessed and downloaded from the following link:

📥 Download Model Weights

📊 Model Performance

Model	Accuracy (%)	Precision (%)	Recall (%)	F1-Score (%)
ConvNeXt	92%	92%	92%	92%
Vision Transformer (ViT)	88%	88%	88%	88%
DenseNet121	87%	87%	87%	87%
EfficientNet_b0	87%	87%	87%	87%
ResNet50	89%	89%	89%	89%

💡 Key Technologies (Tools)

• Google Colab – for training and experimenting with the model in a cloud-based environment
• PyTorch – the deep learning framework used to build and train the models
• Grad-CAM – for visual explanations to enhance model interpretability (XAI)
• FastAPI – to serve the model through a lightweight and efficient web API
• GitHub – for version control, collaboration, and sharing the project source code
• Firebase – used for managing the database and user authentication
• Cloudinary – used to store images, with URLs saved in Firebase
• Visual Studio Code – used for website development

⚙️ FastAPI Integration Steps

1. upload the [code] project folder into --> C:\Users*your account name*

2. open vs code and then from file open [autismlens-project] folder that is inside [code] folder

3. follow the steps here [https://typer.tiangolo.com/virtual-environments/#create-a-project], starting from Create a Virtual Environment command: NOTE: If PowerShell prevents script execution[Activate the Virtual Environment]: PowerShell might block the execution of scripts due to its execution policy. You can temporarily allow scripts to run by changing the policy: Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope Process Then try activating again: .\venv\Scripts\Activate.ps1

4. pip install fastapi uvicorn torch torchvision pydantic requests

5. uvicorn main:main --reload

👩‍💻 Developers

• Jumana Khawaji

  

• Atheer Al Otaibi

  

• Jana Albader

  

• Rama Alzahrani

  

• Ruba Alshehri