Pulmonary Nodule Detection in Lung CT Images using Deep Learning Networks
Principal Investigator
Name
Tsvi Lev
Degrees
MSc (in Theoretical Physics)
Institution
NEC Corporation of America (Israeli Research Labs)
Position Title
General Manager
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-400
Initial CDAS Request Approval
Apr 17, 2018
Title
Pulmonary Nodule Detection in Lung CT Images using Deep Learning Networks
Summary
The project (ongoing for 1 year now) aims to develop a computer aided diagnostics (CAD) tool for the automatic detection of Pulmonary Nodules in Lung CT images. As is common in recent years, we are using Deep Learning to develop this model. We have internally a team of 4 researchers/developers that deal with: adapting existing leading edge DL models, improving them and using ensemble techniques to combine them, create tools for annotation of images so that Radiologists can annotate CT scans and give feedback on our model's output, a novel UI to accelerate the process of 'ruling out' false alarms by the Radiologist, and special modules for false alarm (blood vessel thickening) removal.
The project has so far operated on public open source datasets such as Luna16 and the Kaggle data set, as well as on data provided to us by Assuta Medical Center and Kupat Holim Meuhedet (an Israeli HMO). We also have a Radiologist consultant on our team for day to day annotation and feedback.
We wish to increase our dataset as we have seen clearly the results improve when we add more data and label/annotate it.
The project has so far operated on public open source datasets such as Luna16 and the Kaggle data set, as well as on data provided to us by Assuta Medical Center and Kupat Holim Meuhedet (an Israeli HMO). We also have a Radiologist consultant on our team for day to day annotation and feedback.
We wish to increase our dataset as we have seen clearly the results improve when we add more data and label/annotate it.
Aims
Using additional Lung CT images, improve the performance of the existing Deep Learning model to improve sensitivity/accuracy (percent of true positives) and dramatically reduce the amount of False Positives.
Collaborators
Radiology Institute of Assuta Medical Center, Led by Dr. Michal Guindy
https://www.assuta.co.il/en/?catid=%7B9CD5913A-B2B5-4476-9400-11FD6EF695CB%7D
https://www.linkedin.com/in/michal-guindy-830b00a6/