Robust model-based detection of the lung field boundaries in portable chest radiographs supported by selective thresholding

Abstract

Portable chest radiography is a valuable tool for screening patients hospitalized in intensive care, providing visual cues for diagnosis and physiological measurements. However, its practicality comes at the cost of quality, which is mainly affected by misaligned body positioning, thus increasing x-ray misinterpretation rates. This paper presents a novel methodology for the detection of the lung field boundaries in portable chest radiographs of patients with bacterial pulmonary infections. Such infections are radiographically manifested as foci of consolidations which can lead to vague or invisible lung field boundaries, difficult to distinguish even by experienced physicians. Conventional and state-of-the-art approaches address mainly stationary radiographs, whereas only a few of them cope with pulmonary infections. The proposed methodology is based on an active shape model incorporating shape prior information about the lung fields. The model is initialized by a novel technique utilizing a set of salient points detected on the peripheral anatomic structures of the lungs. A selective thresholding algorithm based on a spinal cord sampling process supports both the initialization and the evolution of the model for the detection of the lung field boundaries. The experiments show that the proposed methodology outperforms state-of-the-art approaches.

Patient specific normalization of chest radiographs and hierarchical classification of bacterial infection patterns

“…The radiographs are first normalized using the spinal cord as a reference point. Afterwards, the lung fields are isolated in regions of interest (ROIs) defined either manually or with a pre-processing lung field boundary detection algorithm [9]- [10]. The proposed image analysis methodology is applied only in these ROIs.…”

“…More specifically we consider the region of the spinal cord/mediastinum as a reference region for normalization. The reasons motivated our approach include: (a) this region can be easily detected since it lies roughly somewhere in the middle of the radiograph and it is generally characterized by higher intensities than other regions in a chest radiograph [9]- [10], (b) though the deviation of the absolute intensity features may considerably vary between different radiographs, the relative intensities of the different anatomic structures within each image, considering the spinal cord/mediastinum region as a reference, are rather constant. This is explained by the nature of This work was supported in part by the European Commission's Seventh Framework Information Society Technologies (IST) Programme, Unit ICT for Health, project DEBUGIT (no.…”

Semi-automated Lung Field Segmentation in Scoliosis Radiographs: An Exploratory Study

“…As an alternative to polynomial fitting, the fitting of Bézier curves after the extraction of the control points, as similarly done in a previous study [8], could also be considered. Other variations in the representation of the curve include increasing the order of the polynomials, especially at the apex.…”

“…More recent methods are based on established deformable models [16] and process the image globally. Some methods use shape priors or shape models to derive the initial template from a training set or multiple images which are manually outlined [8,9]. This imparts the knowledge of the user to the algorithm by restricting the output to natural shapes.…”

“…Many lung segmentation methods are based on knowledge of the shape of the lung and require a training set for the construction of a prior model [8][9][10]. One such technique uses a statistical shape model created from the features of a training dataset to obtain shape priors, which are further processed via iterative thresholding [9].…”