[en] Recently, it has become even more important to understand the normal anatomical structures of the human body in the field of radiological image anatomy. The recognition of body cavities is useful when organ extraction is performed. However, it is difficult to perform organ extraction and to obtain a clear understanding of anatomical structures based on CT values. For example, the distributions of the CT numbers of muscle and organs overlap each other. Therefore, the recognition of the body cavity domain can reduce the range of internal organ extraction and simplify organ segmentation. The purpose of the present study was to develop an automated method for recognizing the body cavities in torso X-ray CT images. Our body cavity recognition method is based on skeletal positions and is performed by extracting the borders of the thoracic cavity, abdominal cavity, and pelvic cavity. This method detects the initial and final points of such borders and searches for the shortest path on the surfaces of the skeletal structures. It then generates the boundary surfaces of the thoracic, abdominal, and pelvic cavities by drawing straight lines between the shortest paths and the centers of the paths. The body cavities located between the superior thoracic aperture and the diaphragm, the diaphragm and the pelvic inlet, and the pelvic inlet and the pelvic outlet are regarded as the thoracic cavity, the abdominal cavity, and the pelvic cavity, respectively. The method was applied to 20 cases in which torso X-ray CT images were obtained. The results showed that the body cavities were extracted correctly in most cases: the thoracic cavity in 17 cases, the abdominal cavity in 19 cases, and the pelvic cavity in 18 cases. We confirmed that our proposed method is effective for recognizing these body cavities. As one of the applications of our study, segmentation of muscle, fat, and the rectum in CT images was performed using the information obtained for body cavity structures. The results confirmed that the body cavity structures were helpful for recognizing the muscle, fat, and rectum regions in CT images. (author)