Automated Lung Nodule Detection using Deep Neural Networks
Principal Investigator
Name
Kun-Hsing Yu
Degrees
M.D., Ph.D.
Institution
President and Fellows of Harvard College
Position Title
Assistant Professor
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-286
Initial CDAS Request Approval
Feb 22, 2017
Title
Automated Lung Nodule Detection using Deep Neural Networks
Summary
Lung CT is essential for the early diagnosis of lung cancer. However, manual evaluation of CT scan images is laborious and subject to human error. In this study, we will build a fully-automated system that analyzes CT scans from various sources, identify scans with suspicious lung nodules, and correlate CT scans with pathology image patterns.
With a better lung nodule detection system, we could point to clinicians images requiring further review and reduce the rate of human error in cancer diagnosis.
Aims
Specific Aim 1. To build a normalization method that accounts for batch effects.
To address batch effects and other artifacts, we will build a normalization method that makes the images comparable.
Specific Aim 2. To identify lung nodules systematically
We will train deep neural networks to identify possible lung nodules from the CT scans.
Specific Aim 3. To classify benign and malignant lung tumors
We will label the lung tumors by reviewing the scans manually and build an automated nodule classification system using convolutional neural networks.
Specific Aim 4. To reveal the radiology-pathology association via machine learning
Patients with different pathology subtypes may present different radiology manifestations on their CT scan images. We will develop machine learning algorithms to associate radiology and pathology imaging findings.
Collaborators
None.