Automated diaphragm delineation tackles segmentation in CT images

Precise delineation of relevant anatomical structures is necessary for accurate treatment planning and radiation therapy. Yet segmentation in CT images of internal features like the diaphragm can be a difficult problem.

Independent research in Canada and Japan recently established methods to automatically delineate the diaphragm from 3D CT data sets.

"Delineating the diaphragm assists in controlling the scope or spatial extent of search, detection, and segmentation procedures for thoracic and abdominal organs and tumors," said Rangaraj M. Rangayyan, Ph.D., of the electrical and computer engineering department at the University of Calgary in Alberta.

The Canadian method segments the diaphragm automatically, based on the spatial proximity of the lung muscles, identification of the lower surface of the lungs as an initial representation of the diaphragm, and application of sophisticated algorithms, such as least-squares and deformable contour modeling (J Digit Imaging 2008 Jan. 29 [Epub ahead of print]).

The paper reports the application of the new automated delineation method to nine CT exams of four pediatric patients with neuroblastoma. Results were evaluated against the boundaries of the diaphragm identified independently by a radiologist.

"Good agreement was observed between the results of segmentation and reference contours drawn by the radiologist, with an average mean distance to the closest point of 5.85 mm over a total of 73 CT slices," Rangayyan said.

The cases examined showed no significant pathological abnormalities of the diaphragm, suggesting that results may be compromised by diaphragm shapes that do not agree with the surface model in cases affected by pathology of the diaphragm.

"It will be necessary to test the procedure on a larger data set and examine the results when the procedure is applied to cases with diaphragm abnormalities," Rangayyan said.

Further tests are also indicated to determine the effect of the spatial resolution of the CT image on method accuracy.

"In the present work, the methods were applied to CT images with poor interslice resolution," he said.

Rangayyan added that higher accuracy will likely result with CT images possessing higher spatial resolution and lower interslice distance.

An automated delineation study published simultaneously by Japanese researchers reported good results obtained using high-spatial resolution CT images (IEEE Trans Biomed Eng 2008 Jan;55(1):351-353). In the Japanese method, the position of the diaphragm surface can be estimated by deforming a thin-plate model to match the bottom surface of the lung.