End to End CNN Model for Lung Cancer Prognosis Using 3D Time Serial Imaging

Abstract

Recently there has been an increase in studies that attempt to use machine learning to improve diagnosis and treatment planning, as well as assess treatment response in various cancers including lung cancer. However, most of the studies focus on the former and there is a distinct lack of studies that attempt to tackle the latter. Medical imaging provides noninvasive means for tracking tumor response and pro- gression after treatment.

However, quantitative assessment through manual mea- surements have been proven to be both time-consuming, tedious, as well as highly susceptible to variability depending on visual evaluation. Machine learning meth- ods offer automatic quantification of radiographic characteristics which may even- tually lead to better predictions of treatment response and pave the way for more individualized treatments.

The aim of this thesis is to evaluate a deep learning algorithm that predicts lung cancer treatment response by examining time serial 3D computed topography (CT) volumes. The model will attempt to predict survivability rates as well as other cancer endpoints such as progression, distant metastases, and local-regional recur- rence by tracking biomarkers in CT scans before and after radiation therapy or surgery.