Room: AAPM ePoster Library
Purpose: CT reconstruction kernels impact the sharpness of CT images. The purpose of this study was to establish the impact of CT reconstruction kernels on atlas-based segmentation of normal tissue.
Methods: CT scans acquired of 5 head-and-neck cancer patients were reconstructed using the B30 (smooth), B60 (sharp), and B80 (ultra-sharp) kernels on the Siemens Biograph mCT. Atlas-based auto-segmentation was performed on each CT and compared with manually delineated contours by a single experienced radiation oncologist for the following structures: brainstem, parotids, and mandible. Quantitative comparison of the contours was performed with the Dice Index, which measures overlap. Repeat measure ANOVAs were performed on the dice indices for each of the different contours using the reconstruction kernels as the independent factor.
Results: The average dice indices for the following structures are as follows, brainstem: 0.65(B30), 0.65(B60), 0.61(B80); left parotid: 0.51(B30), 0.58(B60), 0.49(B80); right parotid 0.61(B30), 0.66(B60), 0.57(B80); and mandible: 0.8(B30), 0.76(B60), and 0.75(B80). The repeat measure ANOVA p-values were: brainstem: p=0.76, left parotid: p=0.69, right parotid: p=0.89, and mandible: p=0.88. No significant differences were detected in dice indices between the different reconstruction filters.
Conclusion: There were no significant differences detected between the repeat measure ANOVAs for any of the structures, which indicates the reconstruction kernels do not impact atlas-based auto-segmentation. However, further analysis should be performed on the required time investment for review and adjustment of the auto-segmented contours across the different reconstruction filters.
CT, Contour Extraction, Image Correlation