Detection and Diagnosis of Lung Cancer with Deep Learning
Principal Investigator
Name
Shan Li
Degrees
Ph.D
Institution
Zephex Technology
Position Title
Co-founder
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-295
Initial CDAS Request Approval
Mar 21, 2017
Title
Detection and Diagnosis of Lung Cancer with Deep Learning
Summary
The number of lung cancer patients is increasing exponentially in China due to the worsened air quality and a large base of regular smokers. Even more alarmingly, this number is expected to rocket in the next few decades. CT scans are essential in the detection of lung nodules, but the evaluation of such scans are often subject to human error, given the overwhelming volumes received by the radiologists at hospitals in China. In this project, we aim to develop deep learning algorithms for the detection, feature extraction and characterization of malignant lung nodules in CT scans to assist radiologists to identify cancer at earlier stage with higher accuracy.
In the first phase, we want to use the images from NLST to develop an image recognition AI to differentiate CT scans between cancer and cancer free patients with both high efficiency and accuracy. We already accumulate some CT images but to build a generalizable algorithm, we need more data points, and the images from NLST would definitely be crucial for our research.
After we develop the algorithm, a portion of the images from NLST will be held out for validation tests. We will evaluate the robustness of the model by leveraging the images from NLST. We will continue to accumulate our training images database in order to build a generalizable, robust model that can be implemented to assist radiologists on their work.
In the first phase, we want to use the images from NLST to develop an image recognition AI to differentiate CT scans between cancer and cancer free patients with both high efficiency and accuracy. We already accumulate some CT images but to build a generalizable algorithm, we need more data points, and the images from NLST would definitely be crucial for our research.
After we develop the algorithm, a portion of the images from NLST will be held out for validation tests. We will evaluate the robustness of the model by leveraging the images from NLST. We will continue to accumulate our training images database in order to build a generalizable, robust model that can be implemented to assist radiologists on their work.
Aims
Firstly, we want to develop deep learning algorithms to detect lung cancer nodules from CT scans, we want all patients’ CT scans who were diagnosed with cancer and CT scans from cancer free patients.
Furthermore we want to characterize malignant and benign nodules based on information extracted from CT scans combined with other patient-specific clinical results.
Ultimately, we want to develop automated diagnosis tools integrated with EMR to assist radiologists and doctors in evaluation of CT scans as well as treatment decisions.
Collaborators
Qiaoyi Li, Zephex Technology
Mo Chen, Zephex Technology