Automatic Early Lung Cancer Detection with Deep Neural Networks
Principal Investigator
Name
Jackie Jiang
Degrees
Ph.D.
Institution
Wingspan Technology
Position Title
Chief Data Scientist
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-287
Initial CDAS Request Approval
Feb 22, 2017
Title
Automatic Early Lung Cancer Detection with Deep Neural Networks
Summary
With the rapid development of deep learning techniques in computer vision, neural network algorithms based on large amount of labeled data achieve better-than-human performances in 2-dimentional image classification and segmentation. While human being is more comfortable to deal with 2-dimentional images than 3-dimentional images, we expect deep learning algorithms to surpass human experts in certain aspects of 3-dimentional medical image analysis, if exposed with sufficient labeled data.
The main objective of this project is to develop deep learning algorithms for the detection, feature extraction and characterization of lung nodules in (low-dose) Computed Tomography (CT) scans. Lung nodules are widely recognized as the characterization of the early stage of lung cancer. The early diagnosis of malignant nodules is a crucial issue for reducing morbidity and mortality of lung cancer, which causes 1,378,000 deaths each year.
We propose 3D convolutional network architectures for automatic CT lung scans image processes, which includes extraction of nodule candidates, reduction of false positives and final classification to identify malignant and benign nodules.
The main objective of this project is to develop deep learning algorithms for the detection, feature extraction and characterization of lung nodules in (low-dose) Computed Tomography (CT) scans. Lung nodules are widely recognized as the characterization of the early stage of lung cancer. The early diagnosis of malignant nodules is a crucial issue for reducing morbidity and mortality of lung cancer, which causes 1,378,000 deaths each year.
We propose 3D convolutional network architectures for automatic CT lung scans image processes, which includes extraction of nodule candidates, reduction of false positives and final classification to identify malignant and benign nodules.
Aims
1. Develop deep learning algorithms for automatic nodule segmentation from lung CT scan images.
2. Engineer with and evaluate novel 3D convolution neural networks architectures and training schemes.
3. Build classifier to identify malignant and benign nodules with features extracted from lung CT scans and other patient-specific clinical factors.
4. Develop diagnostic tools that help to detect lung cancer in the early stage and understand the underlying mechanics of diseases by localizing discriminative and non-informative tissues/feature factors.
Collaborators
Haifeng Bian, CTO, Wingspan Technology
Fengbo Xie, Ph.D. Wingspan Technology