Improved Visualization of Digital Chest Radiographs with Bayesian Image Processing*

A web based computer environment will be used to demonstrate the improved visualization of digital chest radiographs with Bayesian image processing. Bayesian image processing is a scatter reducing statistical estimation technique that produces images with increased contrast-to-noise ratios without degradation of image resolution. This technique uses spatially varying models of scatter and noise to reduce scatter while constraining image noise. A graphical user interface will be used to present the background theory used to develop the Bayesian image processing technique. Quantitative results on residual scatter fractions, contrast-to-noise ratios, and resolution will be presented. These results suggest improved image quality by Bayesian processing. Lastly, the exhibit will demonstrate improved image visualization for the digital chest radiograph by presenting pairs of original bedside patient images and Bayesian processed image. The observer will be asked to examine the image pairs for improvements to pulmonary nodule detection, as well as, general visual improvements. Comments and feedback will be collected on site via web techniques.

Through interacting with this exhibit, participants will: Understand the Bayesian approach to chest image enhancement Interact with an image processing system View the progression of image quality as the system operates View compensated and non-compensated chest images Decide if the statistical compensation technique improves the visualization of pulmonary nodules. *This work was supported in part by grant number CA60849 from the National Cancer Institute. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute.