R-AGI

Inspiration

Radiologists need to interpret hundreds of images a day (average one image gets 3–4 seconds).

What it does

Analyzes an x-ray image: anomalies are localized and diagnosed.

How we built it

We used one pre-trained model and another we trained ourselves (after training via kaggle), and mixed the results via confidence analysis to view the most appropriate results.

Challenges we ran into

We had issues running the models, getting proper results, putting everything together with the UI, and having the framework operate smoothly.

Accomplishments that we're proud of

The system's accuracy is proving 90% and above!

What we learned

Anomaly localization is not a solved problem, there is work that needs to be done there.

What's next for R-AGI

Add more defined models (to detect more diseases/issues with x-ray scans) with feedback optimization loops to provide continuous improvement.

Built With

• keras
• python
• pytoch
• tensor
• tkinter