A convolutional neural network based approach to detect small size tumor tissue at early stage
Principal Investigator
Name
Xinjun Zhang
Degrees
Ph.D
Institution
Indiana University Bloomington
Position Title
Bioinformatics Scientist
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-227
Initial CDAS Request Approval
Sep 30, 2016
Title
A convolutional neural network based approach to detect small size tumor tissue at early stage
Summary
Lung cancer is the top one cause of cancer-related deaths for both men and women in USA and many other countries. The overall 5-year survival rate for all stages patients is as low as 15%. As revealed by previous studies, low-dose CT (LDCT) scan can help lower the risk of dying from this disease. However, only experienced doctors are able to accurately identify extra small size cancer tissue and nodules from radiology image data after many years training and practice. Thus it has imposed a challenge on young doctors who generally lacks long term trainings. In recent years, rapid development of computer vision especially image recognition algorithm such as convolutional neural network, has greatly promoted precise object identification and segmentation of image data. This advancement has shed light on its potential application in analyzing radiology image data. Our target in this proposed study is to help young doctors to interpret radiology image data as well as experienced doctors without many years training. Specifically, our goal is to develop a convolutional neural network based approach to detect small size tumor tissue at early stage from radiology image data for four types of cancers. We will first ask help from experienced radiologist to manually label tumor tissue and normal tissue in image data which will be used for model training. Then we will test our model on independent test dataset and calculate measurement such as accuracy and precision. We will further compare the performance of our model with experience radiologist to identify how to further improve model accuracy.
Aims
Aims:
1. Develop a fast and accurate computational approach using convolutional neural network to classify small size cancerous tissue from the surrounding normal tissues.
2. Our algorithm will be able to differentiate disease subtypes and stages based on image data.
Collaborators
Fei Xiong, Southwest Forestry University China