DIATOS

Inspiration Diabetic Retinopathy is a leading cause of blindness globally, especially in underserved regions with limited access to ophthalmologists. Manual screening is slow and prone to errors. We wanted to build an AI-powered solution that accelerates and enhances DR diagnosis to support healthcare professionals and improve early detection outcomes.

What it does DIATOS is an AI-driven diagnostic system that uses a deep learning model (ResNet50) to classify retinal images based on Diabetic Retinopathy severity. With Intel’s NPU, OpenVINO Toolkit, and Arc GPU, the model processes and delivers accurate, real-time diagnoses through a user-friendly interface, making it ideal for clinical and telemedicine use.

How we built it We trained our model using the Kaggle Diabetic Retinopathy dataset with heavy preprocessing and augmentation. ResNet50 was fine-tuned for classification, and Intel’s hardware tools (AI PC NPU, Arc GPU) optimized training and inference. We used PyTorch, OpenVINO, and Docker for deployment and scalability.

Challenges we ran into Balancing accuracy and overfitting due to image variability.

Ensuring compatibility with Intel optimization tools.

Processing high-resolution images efficiently on edge devices.

Creating a seamless multimodal interface for clinical use.

Accomplishments that we're proud of Achieved real-time diagnostic performance with Intel hardware.

Optimized a complex model pipeline with OpenVINO.

Built a multimodal user interface suitable for telemedicine deployment.

Created a scalable, AI-powered solution with potential for real-world clinical use.

What we learned The power of Intel’s AI toolkit in accelerating model performance.

The importance of preprocessing and augmentation in medical imaging.

How to integrate deep learning into real-world healthcare workflows.

Practical insights into multimodal interfaces for diagnostic tools.

What's next for DIATOS Expand the dataset for improved model generalization.

Integrate patient history and multimodal imaging for richer insights.

Conduct clinical trials to validate and refine usability.

Launch deployments in rural or underserved healthcare regions.

Built With

• arc?
• built-with-languages:-python-frameworks:-pytorch
• conda-platforms:-intel-ai-pc-(with-npu)
• containerization:
• dataset
• deployment:
• docker
• edge
• for
• gpu
• intel-extension-for-pytorch-development-tools:-jupyter-notebooks
• intel?
• kaggle
• openvino
• portability
• scikit-learn-libraries:-openvino
• source:
• tensorflow
• vs-code