Computer Vision AI to Diagnose Lung Cancer from CT Images
Principal Investigator
Name
Peter Szoldan
Degrees
M.P.P.
Institution
MedInnoScan Kft.
Position Title
CEO
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-306
Initial CDAS Request Approval
May 12, 2017
Title
Computer Vision AI to Diagnose Lung Cancer from CT Images
Summary
Hungary has two sad records - it has both the highest rate of incidence and the highest rate of mortality of lung cancer in Europe. Even though smoking is slowly decreasing in the country and stricter air pollution standards are expected to be introduced, this situation is not expected to change soon. There is a plan to institute annual chest CT screening for smokers but that would put a large burden in terms of number of CTs to evaluate on radiologists.
We would like to lessen this bottleneck by developing a computer vision AI system relying on deep convolutional neural networks that would give an estimate whether a given CT shows signs of malignant tumors, and also pinpoint the most suspicious areas for final analysis by the radiologist.
For this we need as much training data as possible, that's why we apply for access to the NLST database.
We would like to lessen this bottleneck by developing a computer vision AI system relying on deep convolutional neural networks that would give an estimate whether a given CT shows signs of malignant tumors, and also pinpoint the most suspicious areas for final analysis by the radiologist.
For this we need as much training data as possible, that's why we apply for access to the NLST database.
Aims
Our aim is to develop a CAD AI system that would assist the radiologist in establishing a lung cancer diagnosis from a CT scan, or alternatively determine that the patient is cancer-free.
We aim to reach selectivity and specificity stats close to that of the human radiologists, but we do not aim to build a system that could be relied upon without human supervision at this time.
Collaborators