Lung nodule screening and classification using convolutional neural network (CNN) algorithms
Principal Investigator
Name
Mike Liu
Degrees
MBA
Institution
Raywon Technology Inc
Position Title
CEO
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-202
Initial CDAS Request Approval
Mar 29, 2016
Title
Lung nodule screening and classification using convolutional neural network (CNN) algorithms
Summary
The field of deep learning and images recognition is experiencing rapid development due to the progresses achieved on developing CNN and computational power. Medical images screening represents the most possible field to utilize this developments. The goal of our study is to investigate and develop CNN algorithms to automatically detect and classify lung nodules.
Further more, we will use the subjects’ pathology and diagnostic data to study the clinical relevancy with medical images.
The five years mortality rate of a cancer patient in US is 25%, but in China this number is 75%. By identifying nodules and correlate them with diagnosis, we will help radiologists to detect cancer in an early stage. This will result in reducing mortality rates through early medical treatment, especially in China where the population of patient is huge.
In order to train the CNN algorithms, a large amount of image data is required for achieving accurate results. With the help of NLST’s dataset, we will be able to build the algorithms and contribute to the global trend of utilization of deep learning on medical images recognition.
Further more, we will use the subjects’ pathology and diagnostic data to study the clinical relevancy with medical images.
The five years mortality rate of a cancer patient in US is 25%, but in China this number is 75%. By identifying nodules and correlate them with diagnosis, we will help radiologists to detect cancer in an early stage. This will result in reducing mortality rates through early medical treatment, especially in China where the population of patient is huge.
In order to train the CNN algorithms, a large amount of image data is required for achieving accurate results. With the help of NLST’s dataset, we will be able to build the algorithms and contribute to the global trend of utilization of deep learning on medical images recognition.
Aims
Aim #1: lung nodules detection
Our output for this aim is to have a high accuracy detection system to identify lung nodules and suspected regions.
Aim #2: Nodule classification
The second step of our study is to classify nodules’ malignancy with a probability output.
Aim #3: Extract high degree similarity to correlate medical images and diagnosis.
Collaborators
Eric Ding, Raywon Tech
Yi Guo, Raywon Tech