Cancer Detector

For this project, I trained a CNN based on the UNET architecture on a database of labelled cancerous cell images. The goal was to predict cancer with >85% accuracy. (Greater than 85% because that’s about the score you’d get with this dataset by predicting everything was non-cancerous.)

The code for this project is here.

Here’s an example of the dataset and the image that I compare against:

Here are some training results where I can predict with 92.5% accuracy where the cancerous cells are:

and the final resulting prediction which can be compared to the ground truth:

Not perfect, but does a decent job!