Finding the rib cage in chest radiographs

-We describe a system of computer algorithms that finds the rib cage in chest radiographs. Our algorithms link the dorsal and ventral portions of the rib contours such that each rib is identified and can be displayed individually. In a preliminary application of our system to five adult chest radiographs, about 90)o of the ribs were found, and ca. 15% of the computed rib contours were false. Pattern recognition Heuristic search Image enhancement Edge detection Contour following Representation Search strategy I N T R O D U C T I O N We describe a system of computer algorithms that currently finds ca. 90% of the rib cage in chest radiographs. As a component of a computer-assisted system for diagnosing chest radiographs, our rib cage finder is expected to have several important uses, including the following. 1. Rib images contribute false positives in tumordetection algorithms. Hence, detecting the ribs should facilitate removal of these false positives. 2. The ribs give the radiologist a convenient frame of reference for the description and location of ribs within the thoracic cavity. 3. The outer envelope of the rib cage provides a more accurate basis for computing the boundary of the chest cavity than other available techniques. Earlier work on automatic rib detection has been reported by Toriwaki et al. m Their technique, however, has the following shortcomings: (a) It is restricted to a vertical strip occupying approximately the middle third of each lung, and within that strip is effective mainly on the middle subset of ribs. (b) It seems to be seriously degraded by tumors and lesions that overlap the rib images. (c) It does not detect the lower contours of the ventral portions of the ribs. Our system seems to have none of these weaknesses. In particular, our system has successfully detected rib contours that are partially obscured by a large tumor, as illustrated in Fig. 4. O V E R V I E W O F T H E RIB C A G E F I N D E R Our approach to finding the rib cage is to find and identify the upper and lower contours of each rib. Each of these pairs of contours constitutes our description of a rib. We found that a detector of the rib contours must overcome the following difficulties: 1. The characteristics of a contour often change significantly as one progresses along the contour and, in some instances, parts of a contour disappear. 2. The images of the ribs may be partially obscured by the images of other ribs, tumors, and fibrous or vascular tissue. Our rib cage finder consists of the following sequence of major Steps, as illustrated by the block diagram in Fig. 1. Step 1 scans, digitizes and filters the input radiograph. We refer to this step as the Preprocessor. The scanning is carried out by a Technical Operations, Inc. rotating-drum digital scanning microdensitometer Model 651. Because our scanner can accept photographs no larger than $ in. x 4 in., we reduced our 17in. x 14in. chest radiographs by a factor of approx. 4. The scanning is followed by a filtering of the digitized picture. A digitized input radiograph is illustrated in Fig. 2. Figure 3 shows this radiograph after it has been filtered. It can be observed in Figs. 2 and 3 that the dorsal portions of the ribs are more distinct than the ventral portions. This is a result of the attenuation and scattering of X-rays by the lung tissue lying between the CHEST RADIOGRAPH • P R E P R O C E S S O R ~ STEP I STEP 2 STEP 3 STEP 4 Fig. 1. Block diagram of rib cage finder.