Srinath Somasundaram

-

The Schmahl Science Workshops

Student

NLST (Learn more about this study)

NLST-442

Sep 10, 2018

Experimentally Testing the Efficacy of Convolutional Neural Networks and Other Machine Learning Techniques on Staging of Lung Cancer CTs

According to the American Cancer Association, there have been 234,030 new cases of lung cancer so far this year and is the most common cancer among men and women. A CAD (Computer Aided Design) oriented approach to diagnosis and more specifically the staging, of lung cancer in CT scans, which are the most commonly used modality for diagnosis and initial staging, is important to create a broadly standard and reliable option for diagnosis. A computer-aided design that can identify key nodule and staging details such as size and density through a single scan of the lungs would aid doctors and limit the amount of time and individual attention needed during the diagnosis process ultimately also limiting the use of invasive procedures for staging. While the surface level diagnosis of lung cancer through the use of computer vision has been widely researched, there is little effort focused on the next step of the process, staging. For this, we propose using convolutional neural networks, a machine learning technique used to simulate the mammalian visual cortex, to efficiently stage lung cancer. Convolutional networks have been mainly developed and used for object recognition especially as of late. Thus, our goal is to create and optimize a supervised convolutional neural network through the use of machine learning libraries such as keras and tensorflow, to accurately diagnose and stage lung cancer comparing our approach to others surveyed in literature.