Skip to Main Content

An official website of the United States government

Government Funding Lapse

Because of a lapse in government funding, the information on this website may not be up to date, transactions submitted via the website may not be processed, and the agency may not be able to respond to inquiries until appropriations are enacted. The NIH Clinical Center (the research hospital of NIH) is open. For more details about its operating status, please visit  cc.nih.gov. Updates regarding government operating status and resumption of normal operations can be found at OPM.gov.

Principal Investigator
Name
Derek Merck
Degrees
PhD
Institution
Rhode Island Hospital
Position Title
Assistant Professor
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-147
Initial CDAS Request Approval
Jul 30, 2015
Title
Machine Learning for Event Detection in Medical Images
Summary
Image event detection is a sub-field of computer vision concerned with identifying objects in natural images. This has applications in scene recognition, object detection, and content-based image retrieval. Currently the most powerful event detection frameworks are based on machine learning, where very large collections of images are analyzed to identify how patches of features like edges, colors, and textures are related to particular image events. Machine learning with medical image features could have a huge impact on the related problems of anatomic assessment, pathology detection, and treatment planning in medical imaging. However, due to the complexity of the data, costs, and privacy concerns associated with collecting, annotating, manipulating the requisite large amounts of volumetric medical images, machine learning frameworks for medical images are evolving much more slowly than their natural image counterparts. Moreover, even with a sufficiently large dataset, as the dataset size increases obtaining a spatial labelling of various image features to use for training presents an ever-larger hurdle. The objective of this project is to address the problem of working with large collections of unlabelled data by extending methods for bootstrapping novel classifiers from unlabeled 2D images to 3D medical image analysis.
Aims

Aim 1. Curate a large dataset of volumetric neuroimaging studies suitable for deep learning image feature analysis.
Aim 2. Demonstrate how novel classifiers can be created and refined through active learning with a medical image dataset.