Kidney CT Analyzer

• 

  App Interface

• 

  Vision Transformer Results

• 

  Efficient Net Results

• 

  Github Repo

Inspiration

I lost my maternal grandpa last year. While his death was due to natural causes but he had become sick for the last due to cyst in the abdomen region. If that was perhaps identified timely, he would have perhaps enjoyed his time on the planet in the way he wanted rather than being stuck at home. I aim to build a proper model which can help all patients get a basic opinion, specially in unavailability of a doctor. Since, machines can investigate and understand the problem much faster thus, helping save lives.

What it does

It inputs the model one wants to use (Efficient Net or Vision Transformer) and a CT scan image of the kidney and classifies it into 4 categories Tumor, Cyst, Normal and Stone.

How we built it

Identified the task and dataset. The models were trained on kaggle as the training set was about 2 Gigabytes big and Kaggle provides limited free GPUs to train the models. Once trained, the models were saved. A user interface was built using gradio and the model was uploaded to hugging face so anyone who needs it can use it.

Challenges we ran into

Understanding the basics of machine learning and computer vision to train our model.
Setting up the interface for the models
Had to wait weekly to understand and train the model

Accomplishments that we're proud of

A training accuracy of 100% for the Vision Transformer
A training accuracy of 98.5 for the Efficient Net
An application that is available for free, which can aid doctors make informed decisions faster and give second advice to patients who might be getting scammed.

What we learned

How to use hugging face
How to use gradio as a Ui/Ux

What's next for Kidney CT Analyzer

To merge together more models by taking user feedback

Built With

• gradio
• hugging-face
• kaggle
• python
• pytorch