Acute Lymphoblastic Leukemia Detector

While the State of the Art, AI models are helping us become more comfortable whether it be writing code or searching something, the need of the hour is to use this technology for disease detection.

Comment

Inspiration

From a young age, I have been passionate about Technology and Science. From the last 2-3 years, I have been a silent watcher seeing the unrealistic development of AI. How to it is so similar to the lifecycle of the human yet is able to transform our lives at a whole different scale.
Whether it is help us get information about something we want to learn about or stamping the breaks for us in case, we miss, AI has shown that it can do it all.
Now, I don't want to be a silent watcher, I want to help develop it so that it doesn't just improve our comfort but can also save lives as well. My model Acute-Lymphoblastic-Leukemia-Detector has been made for the timely identification and detection of Acute Lymphoblastic Leukemia which is a cancer affecting the bone marrow in children.

What it does

It takes in an image of the blood cells as input and then classifies it into one of the 4 possible categories: Benign, Early, Pre and Post stages. Doing this classification, can aid people with limited resources to get a timely identification and hence a probable prevention of this cancer. As the model is hosted online, it is usable free of charge.

How we built it

First, the datasetwas found. Then the probable techniques which could be used were identified. In our case, a Vision Transformer(State of The Art model for image classification )was trained for this particular classification task using pytorch. Then the model was downloaded onto the local machine. An interactable interface was created using gradio and the model was hosted on hugging face.

Challenges we ran into

Computational Resources for training- used kaggle as their GPUs boosted the training by 15-20 times the speed on our local machine
Which architecture to use
Setting files for hugging face

Accomplishments that we're proud of

Train Accuracy -100%
Test Accuracy- 99%+
A free to use model, which hopefully prompts for better models to be built

What we learned

Transfer Learning
File Structure for Hugging face
Gradio usage

What's next for Acute Lymphoblastic Leukemia Detector

To make a multimodal cancer detector which can classify different cancer types easily

Built With

  • gradio
  • hugging-face
  • kaggle
  • python
  • torch
Share this project:

Updates