PURPOSE: The purpose of this study was to determine the effects of atlas selection based on different morphometric parameters, on the accuracy of automatic brachial plexus (BP) segmentation for radiation therapy planning. The segmentation accuracy was measured by comparing all of the generated automatic segmentations with anatomically validated gold standard atlases developed using cadavers. METHODS AND MATERIALS: Twelve cadaver computed tomography (CT) atlases (3 males, 9 females; mean age: 73 years) were included in the study. One atlas was selected to serve as a patient, and the other 11 atlases were registered separately onto this "patient" using deformable image registration. This procedure was repeated for every atlas as a patient. Next, the Dice and Jaccard similarity indices and inclusion index were calculated for every registered BP with the original gold standard BP. In parallel, differences in several morphometric parameters that may influence the BP segmentation accuracy were measured for the different atlases. Specific brachial plexus-related CT-visible bony points were used to define the morphometric parameters. Subsequently, correlations between the similarity indices and morphometric parameters were calculated. RESULTS: A clear negative correlation between difference in protraction-retraction distance and the similarity indices was observed (mean Pearson correlation coefficient = -0.546). All of the other investigated Pearson correlation coefficients were weak. CONCLUSIONS: Differences in the shoulder protraction-retraction position between the atlas and the patient during planning CT influence the BP autosegmentation accuracy. A greater difference in the protraction-retraction distance between the atlas and the patient reduces the accuracy of the BP automatic segmentation result.
PURPOSE: The purpose of this study was to determine the effects of atlas selection based on different morphometric parameters, on the accuracy of automatic brachial plexus (BP) segmentation for radiation therapy planning. The segmentation accuracy was measured by comparing all of the generated automatic segmentations with anatomically validated gold standard atlases developed using cadavers. METHODS AND MATERIALS: Twelve cadaver computed tomography (CT) atlases (3 males, 9 females; mean age: 73 years) were included in the study. One atlas was selected to serve as a patient, and the other 11 atlases were registered separately onto this "patient" using deformable image registration. This procedure was repeated for every atlas as a patient. Next, the Dice and Jaccard similarity indices and inclusion index were calculated for every registered BP with the original gold standard BP. In parallel, differences in several morphometric parameters that may influence the BP segmentation accuracy were measured for the different atlases. Specific brachial plexus-related CT-visible bony points were used to define the morphometric parameters. Subsequently, correlations between the similarity indices and morphometric parameters were calculated. RESULTS: A clear negative correlation between difference in protraction-retraction distance and the similarity indices was observed (mean Pearson correlation coefficient = -0.546). All of the other investigated Pearson correlation coefficients were weak. CONCLUSIONS: Differences in the shoulder protraction-retraction position between the atlas and the patient during planning CT influence the BP autosegmentation accuracy. A greater difference in the protraction-retraction distance between the atlas and the patient reduces the accuracy of the BP automatic segmentation result.
Authors: Eric D Morris; Ryan G Price; Joshua Kim; Lonni Schultz; M Salim Siddiqui; Indrin Chetty; Carri Glide-Hurst Journal: Pract Radiat Oncol Date: 2018-04-06
Authors: Joris Van de Velde; Johan Wouters; Tom Vercauteren; Werner De Gersem; Eric Achten; Wilfried De Neve; Tom Van Hoof Journal: Radiat Oncol Date: 2015-12-23 Impact factor: 3.481
Authors: Joris Van de Velde; Johan Wouters; Tom Vercauteren; Werner De Gersem; Eric Achten; Wilfried De Neve; Tom Van Hoof Journal: Radiat Oncol Date: 2016-01-07 Impact factor: 3.481