Mirza Faisal Beg

Ph.D.

Simon Fraser University

Professor

NLST (Learn more about this study)

NLST-444

Sep 28, 2018

Whole chest segmentation: automatic labeling of anatomical and pathological regions in a chest CT scan

The goal of this project is to develop a detailed and comprehensive semantic segmentation pipeline for the labeling of the various anatomical and pathological regions of interest (ROIs) in a three-dimensional (3D) computed tomography (CT) scan of the chest. In general, a deep learning-based approach will be followed, wherein convolution neural nets (CNNs) are designed for addressing each of the component ROI segmentation tasks in the proposed whole chest CT labeling framework. The proposed pipeline will be designed to produce hierarchical segmentation labels that represent the underlying anatomical and pathological structures at various scales. The segmentations from the proposed pipeline will span the whole spectrum, ranging from the gross tissue level labels to the sub-structures within each of the individual organs. Algorithms for identifying and segmenting the pathological ROIs such as lung nodules and tumors will also be developed as part of the proposed whole chest CT segmentation framework.

1. Train and validate CNN models for the hierarchical labeling of chest CT scans.

2. Build a web-based graphical user interface (GUI) for providing easy access to the whole chest CT segmentation pipeline.