To eliminate rib opacities on chest radiographs for quantitatively evaluation of lung opacities, we have developed an automatic algorithm, which sets up sub-regions of interest (sub-ROIs) for avoiding rib opacities in the analysis on chest radiographs. In the first step, a region of interest (ROI) was selected on a chest radiograph, and a 4th-order-polynomial surface was used for background density correction. In the next step, a one-directional Laplacian-Gaussian filter was used for enhancement of rib opacities. In the third step, the image was binarized on the ROI with a multiple threshold method, and then a morphological filter was used for elimination of noise components. Finally, we divided the ROI into some lattice regions and set up sub-ROIs. We calculated the area ratio of rib opacities in the sub-ROIs and judged the existence of rib opacities according to this sub-ROIs. We employed ROC analysis to evaluate the performance of this algorithm. In conclusion, this algorithm could set up sub-ROIs, which avoid rib opacities for image feature analysis of lung opacities.
View full abstract