Automatic X-Ray Screening For Rural Areas

AUTOMATIC X-RAY SCREENING FOR RURAL AREAS

Imaging scientists at an R&D division of the NIH National Library of Medicine have developed an algorithm to auto-detect TB in chest x-rays in order to meet radiologist shortage in rural areas. The algorithms have been validated, allowing radiologists to view and confirm only the x-rays that are identified (by the software) as exhibiting the disease. In addition to developing a user-friendly interface for their application optimized for providers at the point of care, the team proposes to leverage techniques identified through HHS Ignite to seek out new collaborations, and opportunities for pilot deployments at suitable sites domestically in the U.S., and worldwide.

PROJECT SUMMARY
The National Library of Medicine (NLM), part of the National Institutes of Health (NIH) has an ongoing collaborative project with AMPATH (Academic Model Providing Access to Healthcare), an organization that implements the largest AIDS treatment program in the world, and supported by USAID. This project exploits the convergence of imaging research and system development at NLM and NIH policy objectives in global health. The objective is to leverage NLM’s in-house expertise in image processing to clinically screen HIV-positive patients in resource-constrained rural Kenya for lung disease with a special focus on tuberculosis (TB).

Since chest radiography is important to the detection of TB and other pulmonary infections prevalent in HIV-positive patients, AMPATH aims to use lightweight digital x-ray units readily transportable in rural areas. Their staff will take chest x-rays of the population and screen them using NLM developed chest x-ray screening software for the presence of disease. The x-ray units were mounted on the truck that was acquired by AMPATH, and are being tested at two sites in a 40 kilometer vicinity of Eldoret in Western Kenya.

Since the lack of sufficient radiological services in the area suggests the need for automation to perform the screening, NLM’s in-house research effort focuses on developing software to automatically screen for disease in the chest x-ray images. NLM researchers have developed algorithms to automatically segment the lungs, detect and remove ribs, heart, aorta and other structures and then to detect texture features characteristic of abnormalities, leading to a 2-class discrimination between abnormal vs. normal cases and a measure of confidence in its determination.

This team has worked with radiologists from various institutions including the Clinical Center at NIH, Yale University, University of Missouri in Columbia, a regional TB clinic in Maryland, a clinic in India, and a hospital in China to obtain test image data, and expert annotations and radiological readings on image data sets. The screening system’s performance was similar to human experts (87% accurate) but it tended to be more sensitive leading to nearly twice as many false positives. While this may not be considered acceptable for solely automatic screening, it is a significant advance in the science and can be engineered to be useful for population screening purposes in a resource constrained setting. Further research is in progress: advancing the classifiers performance; develop algorithms for detecting rib structures; and improving quality control in image acquisition.

NLM aims to pilot the screening software through the collaboration with AMPATH, but is also seeking opportunities to collaborate with other partners domestically and worldwide.

TEAM MEMBERS 
Sameer Antani (Project Lead), National Institutes of Health
Sema Candemir, National Institutes of Health
Stefan Jaeger, National Institutes of Health
Alexandros Karargyris, National Institutes of Health
Santosh KC, National Institutes of Health
Zhiyun Xue, National Institutes of Health

PROJECT LEAD’S SUPERVISOR
George Thoma, Branch Chief, Communication Engineering Branch, National Library of Medicine, National Institutes of Health

TECHNICAL ADVISOR
Jon Payne, Senior Technical Advisor, United Nations Foundation

BACKGROUND
HHS Ignite is an “incubator for new ideas” run out of the HHS IDEA Lab. Selected teams are introduced to startup methodologies for problem identification and project implementation. In the entrepreneurial spirit, Ignite projects are iterative, their impacts measurable, and their solutions scalable. This is one of 11 projects in Ignite’s second round which began on June 9, 2014 and goes until mid-September 2014.

A federal government Website managed by the U.S. Department of Health & Human Services
200 Independence Avenue, S.W. - Washington, D.C. 20201