Gradient correction and classification of CT lung images for the automated quantification of mosaic attenuation pattern.

Abstract

The detection of density differences, or "mosaic attenuation pattern," on CT images may be difficult when the regional inhomogeneity of the density of the lung parenchyma is subtle. The purpose of this work was to develop a fully automated method for the reproducible quantification of the underattenuated areas of the lung parenchyma. This technique may be useful in increasing the precision of investigation of structure/function relationships. Anatomical segmentation was achieved by a structure-filtering operator based on mathematical morphology. To compensate for the density gradient visible on lung CT scans, a model-based iterative deconvolution filter and an adaptive clustering algorithm were developed. Validation was performed with CT images from a lung phantom, 15 patients with constrictive obliterative bronchiolitis, and 8 normal subjects. The accuracy of the estimate of the density gradient on phantom studies was 93.3%. The automated quantification of the areas of decreased attenuation on scans of constrictive obliterative bronchiolitis was within 8.2% from the average scoring of two experienced observers. The proposed technique is fully automated and can accurately correct for density gradient and classify areas of decreased attenuation on lung CT images.