Improving accuracy in nodule detection and cancer prediction using data-driven deep learning approach
Principal Investigator
Name
Yuan Feng
Degrees
PhD
Institution
Shenzhen Vega Medical Technology Limited
Position Title
Director of AI Lab
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-342
Initial CDAS Request Approval
Sep 18, 2017
Title
Improving accuracy in nodule detection and cancer prediction using data-driven deep learning approach
Summary
Lung cancer is the leading cause of cancer-related death for male in China, according to the 2017 official report by National Cancer Center in China. Lung cancer in its earliest stage can appear as a small nodule, thus early detection by imaging offers an opportunity to catch a tumor before it grows and spreads. It is an enormous burden for radiologists, however, to analyze many millions of CT scans in lung cancer screening. A well-developed computer algorithm is therefore a very interesting topic for both computer scientists and radiologists.
In recent years, deep learning was introduced to the Computer Aided Diagnosis(CAD) area and is making progress in diseases detection. However most of the existing nodule detection and cancer prediction models were built on very limited datasets. To obtain convolutional neural network (CNN) models with high generality, we need a large amount of data for both train and validation. NLST histopathology images and NLST chest CT images will be a great help in designing a effective nodule detection and cancer prediction system.
In recent years, deep learning was introduced to the Computer Aided Diagnosis(CAD) area and is making progress in diseases detection. However most of the existing nodule detection and cancer prediction models were built on very limited datasets. To obtain convolutional neural network (CNN) models with high generality, we need a large amount of data for both train and validation. NLST histopathology images and NLST chest CT images will be a great help in designing a effective nodule detection and cancer prediction system.
Aims
To improve the accuracy of our nodule detection and cancer prediction system, which contains the following procedures:
1. lung segmentation using supervised deep learning technique
2. suspicious nodule detection within the segmented lung area
3. false positive reduction to reduce a large number of false positive candidates while maintain a high sensitivity
4. malignancy assessment for the detected nodules
Our system can help the doctor and hospital to:
1. improve the nodule detection and cancer prediction efficiency
2. reduce the corresponding diagnosis error
Collaborators
NA