Computer Vision Applications for Medical Imaging
Principal Investigator
Name
Jameel Ghata
Degrees
BSBA Economics; MA International Business
Institution
University of Florida
Position Title
Director
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-189
Initial CDAS Request Approval
Jan 29, 2016
Title
Computer Vision Applications for Medical Imaging
Summary
As computer vision and machine vision have progressed in recent years, the ability for algorithmic detection and diagnosis of disease is now possible. I hope to use the dataset to perform preliminary research to train computer vision algorithms to detect abnormalities in the dataset images.The goal will be to develop classifiers to perform the research and determine the outcomes of the test on the image set. In order to train the algorithms one requires a large image set given the nature of deep learning algorithms. This is the purpose of my research project and the nature of my request for the images.
My hypothesis is that in order to accurately classify varying types of lung cancer and lung disease using the technique I have outlined. I will require a large data set to train the algorithm on. In terms of what I will classify, the first attempt will be to classify what a healthy lung looks like, including all of the various nodes, bronchi, and healthy tissues. Once, that is accomplished, I will utilize the data set to introduce diseased lungs including various lung caners, inflammations, tumors and other abnormalities in order to research if the algorithm can detect the lungs with various conditions. My research is very dependent on the image set so the larger the breadth of images and scope of diseases the more I can work to classify. I did not see any other data sets available in the quantity necessary for my research and thus have decided on the lungs due to the image availability. If there are other large file CT image sets available that focus on the brain, including intracranial hemorrhaging, this is also an area of research I hope to review.
My hypothesis is that in order to accurately classify varying types of lung cancer and lung disease using the technique I have outlined. I will require a large data set to train the algorithm on. In terms of what I will classify, the first attempt will be to classify what a healthy lung looks like, including all of the various nodes, bronchi, and healthy tissues. Once, that is accomplished, I will utilize the data set to introduce diseased lungs including various lung caners, inflammations, tumors and other abnormalities in order to research if the algorithm can detect the lungs with various conditions. My research is very dependent on the image set so the larger the breadth of images and scope of diseases the more I can work to classify. I did not see any other data sets available in the quantity necessary for my research and thus have decided on the lungs due to the image availability. If there are other large file CT image sets available that focus on the brain, including intracranial hemorrhaging, this is also an area of research I hope to review.
Aims
I hope to prove that this approach utilizing deep learning and computer vision can enhance the tools that physicians have available to them to readily and accurately diagnose disease. Initial research has shown that the ability of deep learning and computer vision is approaching human capabilities in terms of cognitive abilities. This technology has a chance to transform the way we diagnose disease making it more accurate and efficient to perform this complex work.
Collaborators
Jameel Ghata; University of Florida