DIATOS

Thumbnail
Result Pic
Impact by DIATOS
Result Description

Inspiration:- Diabetic Retinopathy is a leading cause of preventable blindness, especially in underserved regions with limited access to trained ophthalmologists and diagnostic tools. The need for fast, accurate, and scalable diagnostic solutions inspired DIATOS.

What it does:- DIATOS is an AI-powered system that processes retinal images to detect and classify different stages of Diabetic Retinopathy. It provides real-time diagnostic results, aiding early intervention and treatment.

How we built it:- Dataset: Used Kaggle's Diabetic Retinopathy dataset, preprocessing it with techniques like resizing, normalization, and augmentation. Model: Leveraged ResNet50 with transfer learning for accurate DR classification. Hardware: Enhanced performance with Intel AI PC’s NPU, OpenVINO Toolkit, and Intel Arc GPU for faster training and inference. Software: Utilized frameworks like PyTorch and optimization tools from Intel to streamline development.

Challenges we ran into :- Handling the large variability in retinal images across the dataset. Achieving high accuracy while preventing overfitting. Optimizing the model for real-time performance on limited hardware resources. Accomplishments that we're proud of Improved baseline model accuracy from 67% to significantly higher post-optimization. Successfully integrated Intel technologies for hardware acceleration. Developed a robust, deployable diagnostic system suitable for real-time use. What we learned :- The power of transfer learning and hardware optimization in AI-driven diagnostics. Effective preprocessing and augmentation can significantly impact model robustness. Real-time diagnostic tools are achievable even in resource-constrained environments. What's next for DIATOS:- Expand the dataset for better generalization. Integrate additional diagnostic features like patient history and multimodal imaging. Conduct real-world clinical trials to refine usability and gather feedback.