Deep CNNs to detect cardiomegaly in chest X-rays
Brown Bag Lecture by Dr. Sema Candemir | 4/3/2018 11:00AM – 12PM | 7th Floor Conference Room, Bldg 38A
Cardiomegaly is heart enlargement generally due to high blood pressure or coronary artery disease. One way to detect cardiomegaly in chest X-rays is to compute radiographic indexes such as the cardiothoracic ratio, defined as the ratio between the maximum transverse cardiac diameter and the maximum thoracic diameter measured between the inner margins of ribs. We had previously used a rule-based approach to detect cardiomegaly, and now use convolutional neural networks (CNN) to perform this classification. Here we compare the two approaches and show improved performance with CNNs. Further, we use the output probability of the CNN model to estimate the severity of the disease as borderline, moderate and severe.
Dr. Candemir is a research scientist at the Communications Engineering Branch (CEB), Lister Hill National Center for Biomedical Communications. Her research interests are image processing, medical image analysis and computer vision: mainly medical image segmentation, local image descriptors, graph cut algorithms and image registration. She applies her expertise in medical image analysis related projects in CEB.