Abstract
Pulmonary radiographs are essential tools to the evaluation and diagnosis of suspected infections of the lower respiratory system. Interpretation of a radiograph in the clinical context is a valuable diagnostic adjunct to the selection and the management of a specific clinical protocol for therapy. The key element in the proper diagnosis of a bacterial pulmonary infection is the analysis of the radiographic data accumulated over time. A dynamic consultation system that captures the progress of a disease over time can prove a valuable means to patients' monitoring and follow-up. The aim of this work is to provide an initial framework which can be used to describe the progress of a bacterial pulmonary infection based on the spatial variation of its radiographic manifestation in temporal image sequences. This is realized by the unsupervised discrimination of inflammatory areas from normal lung parenchyma in chest radiographs and their quantitative evaluation over time. Inflammatory areas, which are visually discriminated by their relative opacity within the lung fields, are identified by using an hierarchical cluster merging scheme based on successive non-negative matrix factorizations (NMF) of radiographic patterns of intensity and texture. The experimentation results validate the effectiveness of the proposed methodology along with its advantage over standard supervised methodologies where the need for feature normalization between the diverse images is prevalent.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have