Using Contrastive Masked Video Autoencoders to detect lung cancer in high-risk individuals with low-dose CT scans
Principal Investigator
Name
Bryan Jiang
Degrees
GED
Institution
STEM ADEMIA LLC
Position Title
Machine Learning Engineer
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1183
Initial CDAS Request Approval
Jan 9, 2024
Title
Using Contrastive Masked Video Autoencoders to detect lung cancer in high-risk individuals with low-dose CT scans
Summary
Interpretation of medical imagery is essential in the triaging and diagnosing cancer. Lung cancer remains the foremost cause of cancer death around the world. Low-dose computed tomographic screening is a tool that can reduce mortality by up to 20%. This project aims to develop a deep learning system that can provide efficient and accurate lung cancer risk scores through computer tomography (CT) images. We will experiment with contrastive masked video autoencoders pre-trained on video data. We believe these new architectures can yield improvements in this space.
Research in the area of contrastive learning and masked auto-encoders has made significant improvements recently. Using a Video Transformers backbone, we hope to combine the advantages of contrastive learning with that of masked autoencoders to produce state-of-the-art classification results.
Research in the area of contrastive learning and masked auto-encoders has made significant improvements recently. Using a Video Transformers backbone, we hope to combine the advantages of contrastive learning with that of masked autoencoders to produce state-of-the-art classification results.
Aims
1) Develop a reliable AI model which can accurately predict lung cancer risk from low-dose CT
2) Run simulations on lung cancer CT data to test and validate our results
3) Extend the foundation video model to other data
Collaborators
none