An intelligent pre-processing framework for standardizing medical images for CAD and other post-processing applications

There is an increasing need to provide end-users with seamless and secure access to healthcare information acquired from a diverse range of sources. This might include local and remote hospital sites equipped with different vendors and practicing varied acquisition protocols and also heterogeneous external sources such as the Internet cloud. In such scenarios, image post-processing tools such as CAD (computer-aided diagnosis) which were hitherto developed using a smaller set of images may not always work optimally on newer set of images having entirely different characteristics. In this paper, we propose a framework that assesses the quality of a given input image and automatically applies an appropriate pre-processing method in such a manner that the image characteristics are normalized regardless of its source. We focus mainly on medical images, and the objective of the said preprocessing method is to standardize the performance of various image processing and workflow applications like CAD to perform in a consistent manner. First, our system consists of an assessment step wherein an image is evaluated based on criteria such as noise, image sharpness, etc. Depending on the measured characteristic, we then apply an appropriate normalization technique thus giving way to our overall pre-processing framework. A systematic evaluation of the proposed scheme is carried out on large set of CT images acquired from various vendors including images reconstructed with next generation iterative methods. Results demonstrate that the images are normalized and thus suitable for an existing LungCAD prototype(1).