HistoSight

• 

  HistoSight

• 

  Loss over Epochs

• 

• 

Inspiration

As a student in Anatomy and Cell Biology, one thing that is commonly taught about across all classes relating to my major is cancer. More specifically, professors always talk about the importance of staying away from cigarettes, because lung cancer is the leading cause of cancer death worldwide. As such, we felt compelled to create a program that would accelerate cancer screening by identifying if a lung tissue sample was cancerous or not.

What it does

HistoSight receives an uploaded image from the user of a histological lung slide. You can think of histology as the "in-between" of biology and anatomy; it is the essentially the study of tissues. HistoSight is then able to detect with a >95% accuracy whether the tissue sample is cancerous or not. Identifying cancerous nodes can be tricky sometimes, even to the trained eye, so our program would be able to assist and validate cancer diagnosis.

How we built it

We built our model using PyTorch with an EfficientNet-B0 architecture. We used the "Lung and Colon Cancer Histopathological Images" dataset from Kaggle to train our model over 5 epochs. The backend API was built and hosted using FastAPI and the front end used Angular.

Challenges we ran into

We definitely were not experienced in connecting the front end to the back end, so that definitely posed as a challenge for us. However, once everything was up and running, the program worked just as intended and we were very happy with ourselves.

Accomplishments that we're proud of

For some of us, this was our first hackathon or even our first time completing an actual project. So this whole event was really a great accomplishment for us because we came up with something simple, and we are just glad that we are able to have a functioning program to be able to showcase.

What we learned

We learned a LOT about APIs and how to properly utilize it to connect the front end to the back end. It is definitely something that we are glad to have gone through though, since it was truly such a great learning experience for all of us.

What's next for HistoSight

There is lots of room for HistoSight to grow. As of right now, the program can only recognize lung tissues and differentiate it between cancerous vs. non-cancerous. We would be able to expand it by training the model with ALL the different types of tissues in the human body AND recognize whether it is benign or cancerous.

Built With

• angular.js
• fastapi
• python
• pytorch
• typescript