Lung lesions segmentation and cancer prediction
Principal Investigator
Name
Ha Nguyen
Degrees
Ph.D.
Institution
Vingroup Big Data Institute
Position Title
Research Scientist
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-504
Initial CDAS Request Approval
Apr 29, 2019
Title
Lung lesions segmentation and cancer prediction
Summary
Lung cancer is the second most popular cancer in Vietnam (a little bit behind liver cancer). The diagnosis and treatment of lung cancer put an extremely high pressure on doctors, especially at the national hospitals. We are therefore in an urgent need for a computer-aided diagnosis (CADx) system that is able to automatically identify the lung lesions and predict lung cancer. This system will not only accelerate the doctors' diagnosis but also serve as a second opinion for doctors' reference. Furthermore, such system can also be used in large-scale screening with low cost.
In this project, we leverage recent advancements in machine learning to detect lung lesions from CT scans and predict lung cancer from both CT scans and pathological images. In particular, deep learning models will be trained on large datasets of lung CT scans to identify and segment the tumors. The cancer likelihood will be then estimated by combining the characteristics of the tumors with their corresponding biopsies. To achieve the final diagnosis, we integrate both image features and other clinical variables (like age, smoking habit and medical history), as what doctors routinely do. We believe that the fusion of heterogeneous data (CT, pathology, clinical) will significantly boost the precision of the automated diagnosis.
In this project, we leverage recent advancements in machine learning to detect lung lesions from CT scans and predict lung cancer from both CT scans and pathological images. In particular, deep learning models will be trained on large datasets of lung CT scans to identify and segment the tumors. The cancer likelihood will be then estimated by combining the characteristics of the tumors with their corresponding biopsies. To achieve the final diagnosis, we integrate both image features and other clinical variables (like age, smoking habit and medical history), as what doctors routinely do. We believe that the fusion of heterogeneous data (CT, pathology, clinical) will significantly boost the precision of the automated diagnosis.
Aims
- Improve state-of-the-art deep learning models for lung lesions segmentation
- Improve state-of-the-art algorithms for lung cancer prediction by combining CT scans with pathological images and clinical variables
- Publish the research results in prestigious conferences and journals
Collaborators
Dat Ngo - Vingroup Big Data Institute