Automated abnormal asymmetry detection in digital posteroanterior chest radiographs

Abnormal asymmetry in posteroanterior (PA) chest radiographs results from gross pathologic or physical abnormalities that substantially reduce the projected aerated lung area in one hemithorax relative to the other. We have developed an automated method to identify such asymmetry in digital chest radiographs. An iterative global gray-level thresholding technique is used to construct an initial set of contours encompassing the aerated lung regions. Local gray-level thresholding yields contours that more accurately capture the lungs. The respective areas included within the right and left lung segmentation contours were compared for a 600-image database. Deviations from the distribution of the right-lung-area-to-left-lung-area ratio for known normal cases indicated the presence of abnormal asymmetry. The ability of the method to distinguish between normal cases and cases with abnormal asymmetry was evaluated through receiver operating characteristic (ROC) analysis. The performance of the method attained an area under the ROC curve of 0.84. This automated method demonstrates promising performance in its ability to detect abnormal asymmetry in PA chest images. We believe that this method could be used in a picture archiving and communications (PACS) environment in chest radiography to prioritize grossly abnormal cases that require immediate attention.