bahau88/breastcancer-predictor

Web app to demo predicting breast cancer using BMI, Glucose, Insulin, HOMA and Resistin.

Breast Cancer Prediction Using Artificial Intelligence 🤖

Live at: https://breastcancer-predictor.herokuapp.com/ (heroku couldn't install numpy)

             × Encountered error while trying to install package.
             ╰─> numpy

Live at: https://breastcancer-predictor.streamlit.app

This web app uses machine learning to predict whether a person has breast cancer using some of their clinical data.

It takes the following input:

• BMI (kg/m2)
• Glucose (mg/dL)
• Insulin (µU/mL)
• HOMA
• Resistin

And returns either "no breast cancer" or "breast cancer present" along with their probabilities.

Dataset used: Breast Cancer Coimbra

Details described in: Patrício, M., Pereira, J., Crisóstomo, J. et al. Using Resistin, glucose, age and BMI to predict the presence of breast cancer. BMC Cancer 18, 29 (2018)

Demo

How to Run Locally

Clone this repository, create a new environment, and enter the following in your terminal:

streamlit run app.py

This will create a local web server which should open in your default browser. If not just use one of the links returned in your terminal.

Additional Info

• Web app built with Streamlit (which is an amazing tool, if you haven't heard of it already).
• pipeline.py is the code used to generate model.pkl. TPOT was used to create this pipeline.

Possible Issues

If your cloned version gives you an error, please generate a new model.pkl by running pipeline.py and try again. This will overwrite the existing model.pkl.

I ran into ValueError: Buffer dtype mismatch, expected 'SIZE_t' but got 'int' when trying to deploy on Heroku despite it working in my local env. Turns out you can't pickle on one architecture and unpickle on a different one (32/64-bit). Thanks to this answer for helping me understand what the issue was.