VisinEx

• 

  Fundus camera view 1

• 

  Fundus camera view 2

• 

  Fundus camera view 3

Inspiration

Diabetes is one of the most prevalent chronic disorders, affecting millions in the United States alone. Diabetes is characterized with impaired control of blood sugar levels. If sugar levels are poorly managed, blood vessels in the eyes can be damaged, affecting vision and can even cause blindness in a condition called diabetic retinopathy. Rapid detection in early stages of diabetic retinopathy can help slow or even avert the progression.

What it does

VisionEx allows you to either upload or take a retinal fundal image of your eyes. Then a machine learning model predicts the severity of diabetic retinopathy, in the following levels: none, mild, moderate, severe, and proliferative. It then provides explanations and advice on managing the condition based on your level.

How we built it

We divided the project into three sections: an ML model that predicts severity based on retinal images, Gemini text generation for explanations and advice in patient-friendly language, and an app for a user-friendly interface. We also made a fundus camera attachment for a phone camera.

Accomplishments that we're proud of

We are proud of applying transfer learning to quickly deploy an accurate image classifier based on ResNet18. The accuracy is high, at 94% for training data and 60% for validation data (note there are 5 classes). We are also proud of our user-friendly interface and features, including text generation.

What's next for VisionEx

Next steps for VisionEx is improving the accuracy of the model, improving the design of the fundus camera attachment (provides higher quality images with clinical magnification), and perhaps add classifications for other diseases, such as hypertensive retiopathy and cataracts.

Built With

• gemini
• python
• pytorch
• streamlit