Role of Image Segmentation Methods in Increasing the Efficiency and Accuracy of Neural Networks in Cancer Detection

Principal Investigator

Name
Zong Zhang

Degrees
International Baccalaureate Diploma

Institution
N/A

Position Title
Student, Intern

Email
zzhang2@fredhutch.org

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-335

Initial CDAS Request Approval
Jul 28, 2017

Title
Role of Image Segmentation Methods in Increasing the Efficiency and Accuracy of Neural Networks in Cancer Detection

Summary
In previous studies, image cropping and segmentation have shown to greatly aid the creation of more accurate and perceptive neural networks. In this study, different image segmentation algorithms will be used in image prepossessing, and the effectiveness of each method will be assessed. Some methods include Chan-vese segmentation, K-means, region growing, edge detection, and so on. After the images are processed, they will be inputted into convolutional neural networks (both dense and conventional) and the logloss of the neural network over time will be an indicator for the success of both the network and the mode of image segmentation.

Aims

- Identify the most successful method of image segmentation for the NLST dataset
-Creation of a neural network with a high degree of accuracy in differentiating between high risk cancer patients and healthy patients.

Collaborators

Chen Xu, Senior at Interlake High School