Lung cancer is one of the deadly cancers in the U.S. Early Screening is crucial in cancer diagnosis but is error-prone to false positives, which leads to unnecessary treatments and examinations. In this paper, we propose a method to classify lung cancer subtype using deep learning models when input 2D joint histogram preprocessed from 3D CT scan images of patients. With a simple Convolutional Neural Network architecture, we showed an improvement in correctly classifying both 1st and 2nd dominant subtypes of lung cancer compared with a baseline method (which predicts with the average of training labels). This paper was accepted to IEIE 2018, a local conference in South Korea.
- Python 3.4
- TensorFlow 1.3
- Keras 2.0.8
- Python source code of cancer subtype classification pipeline
- MATLAB source code for generating 2D joint histograms from 3D raw images
- lung cancer dataset of 2D joint histograms (.csv)
- visualize the detection pipeline at every step
- Example of training on your own dataset
- demo.ipynb
- random_forest.ipynb
Check out my bachelor's thesis: Lung Cancer Subtype Deep Learning Classifier based on 2D Joint Histograms of multi-modal CTs for more details on this work.