Image feature analysis for computer-aided diagnosis: detection of right and left hemidiaphragm edges and delineation of lung field in chest radiographs.

Diaphragm edges, together with ribcage edges, in chest radiographs provide useful information on the location, shape, and size of the lung fields that are required by computer-aided diagnosis (CAD) schemes for automated detection of various abnormalities. In this continued study, we developed a computerized method for detection of the right and left hemidiaphragm edges. First, the right hemidiaphragm edges in a PA (postero-anterior) chest image were determined by edge gradient analysis. An initial vertical ROI was then placed at the middle of the left hemidiaphragm, based on a "standard rule" for determination of the starting points to search for the left hemidiaphragm edges. Seven categories were used to assess the accuracy of the placement of the initial ROI and the selection of the primary left starting point within the initial ROI. For some categories, it was necessary to select a second left starting point besides the primary one. Therefore, for these categories, two sets of "detected left hemidiaphragm edges" resulted from the two left starting points. Two parameters were used as measures to eliminate the false left hemidiaphragm edges which were due to an incorrect left starting point. Two polynomial functions were applied separately which produced smooth curves for the right and left hemidiaphragm edges. Finally, the delineation of the lung field in a chest image was obtained by connecting the right and left hemidiaphragm edge curves with the corresponding ribcage edge curves. The subjective evaluation results indicated that the accuracy for the determination of the right and left hemidiaphragm edges was approximately 97% and 90%, respectively.