Lungs cancer detection using deep learning
No Thumbnail Available
Date
2019
Journal Title
Journal ISSN
Volume Title
Publisher
UMT, Lahore
Abstract
Lungs cancer is one of the leading cause of deaths with the number of deaths exceeding, caused by colorectal cancer, breast cancer and prostate cancer collectively. Major cause of lungs cancer is smoking, but the non-smokers are not safe either because of passive smoking or air pollution.
The study we conducted is based on the data provided by Kaggle. The first step was to prepare the data, which for obvious reasons is called “Pre Processing”. All the data from the patients is then resized from 512x512x (approximately) 200. The size of the slices varied in patients data. After resizing, the dataset was 50x50x20 in size. A testing data set is separated and a validation test set is provided along with the dataset.
A 3DCNN is run throughout the data. The 3D CNN consists of 2 layers of ReLU and MaxPool followed by a DropOut (removing unnecessary values) and a SoftMax at the end which ultimately returns our binary value. We have trained it on all the dataset.
Previously, diagnosis was done by the pulmonary radiologists based on lungs CT images, but with the improvements in the processing power and lower somewhat lower costs of the hardware, and thanks to the field of deep learning and the vast community available online we have the capability of fast tracking the disease detection and even in some cases the prediction too. Helping save the valuable time for the patients and reducing costs of the procedure. Our tests resulted with an accuracy of 68%, based on the validation dataset.