ryver.ai_tensorTUM

start screen
statistics
leaderboard
quiz
quiz 2

Inspiration

AI can be used in many different ways. To help future doctors to make more precise diagnoses we decided to build a precise classifier for chest CT images

What it does

the application gives the student a picture of a chest CT and lets the student select a disease. During the game, the application collects data and creates a probability distribution based on the ability to recognize the diseases.

How we built it

For the model, we used the azure multiclass classifier and trained it with CT images of different diseases. We optimized which data to use, the training time, and the distribution of the number of images to use with every disease. In total, we trained 16 different models and chose the best one for our application. We used React to build the frontend. For the backend we used Django Restframework and the blob storage of Azure

Challenges we ran into

Finding good data for certain diseases was difficult. Since we have never used Azure navigating was partially slow in the beginning. Connecting the model API with the frontend was a little struggle which was caused by some technical issues

Accomplishments that we're proud of

we basically finished the challenge. We are very proud to have built something of this size in such little time. The precision and recall of the model are in the upper 90% which we are very proud of

What we learned

how to use Azure and the prebuilt models. How to connect Django Restframework with the Azure blob storage. How to navigate in a team and deal with failure within a limited time frame

What's next for ryver.ai_tensorTUM

we would love to upgrade our model to identify more diseases. To add the option to upload own CT images for identification or learning. Add more types of image data like MRI, Ultrasound