Using a Deep Learning Network to Diagnose Congestive Heart Failure.
Long H. Ngo
Medical imaging technologies such as radiography, US, CT, and, increasingly, MRI are indispensable in the screening and diagnosis of diseases of the heart, lungs, bones, and other organs. Chest radiography to detect congestive heart failure (CHF) and related complications is one of the most common radiologic procedures performed in the United States. A positive screening result for CHF at chest radiography may be followed by serum B-type natriuretic peptide (BNP) assessment to help confirm the diagnosis. The diagnostic accuracy of BNP in the detection of CHF is excellent, with a sensitivity above 95%. However, BNP testing is not performed for all patients, the laboratory test is expensive, and the final result may be delayed if an off-site laboratory is used. In the absence of BNP data, what is the loss in diagnostic accuracy of using just the chest radiograph? Is there a data-driven solution to compensate for this loss?