Automatic lung nodule detection by CNN
Principal Investigator
Name
Shuang Wu
Degrees
Ph.D.
Institution
Yitu USA
Position Title
Research Scientist
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-258
Initial CDAS Request Approval
Dec 2, 2016
Title
Automatic lung nodule detection by CNN
Summary
We'd like to apply deep convolutional neural networks to identifying medical conditions based on CT scans.
The first goal is to automatically detect lung nodules.
The second goal is diagnosis of several other types of lung/heart conditions such as Atelectasis, Cardiomegaly, Effusion, Pneumonia, etc.
We have several years' of research experience in computer vision, especially using deep learning approach. The main technique is convolutional neural networks(CNN). We have already developed a prototype, training on a few hundred CT scans, and got some promising results. We feel with more data, we can push accuracy even higher. For example, for nodule detection, we are aiming for sensitivity > 90% when false positive rate is 0.1 nodule/scan. For the lung/heart condition diagnosis task, we are aiming for accuracy > 95%.
The first goal is to automatically detect lung nodules.
The second goal is diagnosis of several other types of lung/heart conditions such as Atelectasis, Cardiomegaly, Effusion, Pneumonia, etc.
We have several years' of research experience in computer vision, especially using deep learning approach. The main technique is convolutional neural networks(CNN). We have already developed a prototype, training on a few hundred CT scans, and got some promising results. We feel with more data, we can push accuracy even higher. For example, for nodule detection, we are aiming for sensitivity > 90% when false positive rate is 0.1 nodule/scan. For the lung/heart condition diagnosis task, we are aiming for accuracy > 95%.
Aims
We'd like to increase the lung nodule detection rate to 90% with false positive rate under 1%
Collaborators
Shu Rong, Yitu Inc.