Zhimin Huo PhD, Presenter: Nothing to Disclose

Fan Xu, Abstract Co-Author: Nothing to Disclose

Hui Zhao, Abstract Co-Author: Nothing to Disclose

John C. Wandtke MD, Abstract Co-Author: Research Grant, Carestream Health, Inc

Anne-Marie Gisele Sykes MD, Abstract Co-Author: Nothing to Disclose

Susan K. Hobbs MD, PhD, Abstract Co-Author: Nothing to Disclose

Jing Zhang, Abstract Co-Author: Nothing to Disclose

David Foos MS, Abstract Co-Author: Employee, Carestream Health, Inc

Chest radiograph is the most commonly used exam by clinicians for screening and diagnosis of lung diseases. Bone structures such as ribs and clavicles on chest radiographs are the major noise contributors. Studies showed that rib suppression improved the detection rate of lung nodules. However, published techniques on rib suppression are currently limited to standard PA chest images. Portable AP chest images are rather challenge because of poorer image quality and inserted foreign objects. We develop rib suppression technique for chest radiographs including AP chest images. Our goal is to improve the detection of lung nodules and pneumothorax on chest radiographs.

Our technique includes 1) lung segmentation, 2) initial rib detection based on pixel classification, 3) rib labeling to identify individual ribs, 4) rib modeling to further refine rib detection, 5) rib edge detection based on already-refined rib detection, and 6) rib suppression. The initial rib detection requires training on a set of chest images with manually identified ribs as the truth using a set of derivative and other features extracted from each individual pixel. We have independently collected the standard PA (59 and 51) and portable AP (41 and 74) images acquired from DR and CR systems, respectively. The detected ribs from step 5 before suppression were evaluated against the hand-draw truth.  

The rib detection results yielded average sensitivities of 83.06% (+/-6.59%) and 74.51% (+/-11.49%) for CR PA and AP images, respectively. The false positive detections, defined as the ratio of area detected outside of the rib to the hand-drawn rib region, are 15.75% and 26.99% on average. The lower sensitivity for potable AP images is largely due to the poor image quality and bad patient positioning. Continue effort is made to improve the detection of ribs on the top of the lung. Overall, our initial assessment of the rib suppressed images showed that current rib detection is sufficient for our suppression techniques. The rib suppressed images will be evaluated visually in terms of an increased visibility on lung nodules and pneumothorax by a group of radiologists.

Our rib suppression technique works well for both standard chest radiographs and portable AP chest radiographs

Our method can potentially improve the detection of both lung nodules and pneumothorax, especially for potable AP images.

Huo, Z, Xu, F, Zhao, H, Wandtke, J, Sykes, A, Hobbs, S, Zhang, J, Foos, D, On Rib Suppression Technique for an Enhanced Interpretation of Chest Radiographs.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12038217.html