Classifying cancer nodules with deep learning and Python
a talk by Daniel Kuchta

CT scan analysis is a tedious and difficult job even for highly-trained radiologists. It takes hours to go through the whole scan and not miss a single spot which could potentially be a malign cancer nodule. It takes years for novice radiologists to learn the craft and even then they tend to misclassify nodules which might later lead to incorrect treatment of the disease.

While we can not improve the radiologist’s ability to detect and classify cancer nodules directly, we can leverage the power of machine learning algorithms we have today together with the data from the past and build a medical support decision system to advise the doctors and warn them, if there is a potentially dangerous nodule detected.

We decided to get our hands dirty and build such a system. After extensive research and understanding of the problem domain we obtained a labeled dataset which would later be used for the training of our models. The next step was to understand the data we have, uncover hidden biases and handling the data from different sources and formats. Afterwards, we build several models using technologies such as genetic algorithms, TensorFlow, Keras and Python and achieved great results. Be sure to come and I’ll tell you more about this amazing journey.

This talk is suitable for both beginner and advanced Pythonistas.

Daniel Kuchta

I'm currently the Head of Machine Learning Practice and a software engineer in GlobalLogic Slovakia where I lead the machine learning activities across the company.

I'm responsible for development of the team, exploring new technologies and working on proof of concepts and projects involving machine learning. I was a speaker at Google DevFest Kosice 2017 and Machine Learning Meetups Kosice.