Automatic identification of chest orientation in digital radiographic images

Abstract

Radiology departments are implementing conversion from the use of hard copy film in favor of digital imaging. New digital acquisitions are increasing the efficacy of radiological imaging. The outputs of new modalities such as magnetic resonance (MR) and computed tomography (CT) are digital. They both involve gathering information that allows reconstructing cross sectional projections of internal structures and displaying them as digital images. Other technologies, e.g., computed radiography (CR), can provide digital radiographic data that replaces analog projection radiography. To date, the processed digital data is still transferred to film to provide a typical radiographic film in appearance. The film is presented to the doctor for diagnostic review. The research in this dissertation is concerned with making a film-less department. It specifically addresses problems in presenting CR images to the physician. The goal of this research is to create a computer recognition algorithm that will automatically recognize the orientation and discriminate between the lateral and posteroanterior view of digital chest radiographs image. The algorithm maintains 91.9% accuracy rate. The recognition takes .15 second per image.