Lung Nodule Segmentation and Cancerous Nodule Detection using A Deep Learning Approach
Principal Investigator
Name
Zhou Ren
Degrees
PhD in Computer Science (in process)
Institution
University of California, Los Angeles
Position Title
Graduate Student Researcher
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-180
Initial CDAS Request Approval
Dec 7, 2015
Title
Lung Nodule Segmentation and Cancerous Nodule Detection using A Deep Learning Approach
Summary
Accurate lung nodule segmentation is essential for cancerous nodule detection. Continuous efforts have been devoted in building automatic lung nodule segmentation systems. Prior work mainly utilize the geometric properties of the lung nodule region (e.g., shape registration based techniques) or the low-level visual properties (e.g., graph-based segmentation).
However, such geometric and low-level feature based methods are sensitive to the variations and noises in the imaging. In order to obtain accurate lung nodule segmentation and an automatic cancerous nodule detection system, we propose to utilize deep learning technique, which has shown remarkable advances in many artificial intelligence domains. Especially in the domain of computer vision that targets on natural image processing, recent algorithms built on top of the deep learning techniques have achieved state-of-the-art performance in various tasks, e.g., segmentation, classification, etc.
In this project, we will develop novel algorithms to leverage the representation and classification capacities of deep learning techniques for medical image processing, in the specific tasks of lung nodule segmentation and cancerous nodule detection. A graphical visualization system will also be developed to visualize the 3D shape of the normal nodules, cancerous nodules, and also the nodule features.
However, such geometric and low-level feature based methods are sensitive to the variations and noises in the imaging. In order to obtain accurate lung nodule segmentation and an automatic cancerous nodule detection system, we propose to utilize deep learning technique, which has shown remarkable advances in many artificial intelligence domains. Especially in the domain of computer vision that targets on natural image processing, recent algorithms built on top of the deep learning techniques have achieved state-of-the-art performance in various tasks, e.g., segmentation, classification, etc.
In this project, we will develop novel algorithms to leverage the representation and classification capacities of deep learning techniques for medical image processing, in the specific tasks of lung nodule segmentation and cancerous nodule detection. A graphical visualization system will also be developed to visualize the 3D shape of the normal nodules, cancerous nodules, and also the nodule features.
Aims
1. To design an appropriate deep learning network architecture for lung nodule segmentation.
2. To design a deep learning network for cancerous lung nodule classification using the lung nodule segmentation results in step 1.
3. To develop a graphical visualization system to visualize the 3D shape of the normal nodules, cancerous nodules, and also the nodule features.
Collaborators