Program Information
Liver Auto-Segmentation Based On Improved Three-Dimensional Dynamic Region Growing Algorithm
Q Qiu1,2*, J Duan1 , G Gong1 , D Li2 , Y Yin1,2 , (1) Shandong Cancer Hospital Affiliated to Shandong University, Jinan, Shandong province,China, (2) Shandong Normal University, Jinan, Shandong Province, China
Presentations
WE-RAM2-GePD-I-6 (Wednesday, August 2, 2017) 10:00 AM - 10:30 AM Room: Imaging ePoster Lounge
Purpose: To study the feasibility of liver auto-segmentation based on improved three-dimensional dynamic region growing algorithm and assess the accuracy of this method.
Methods: In this study, we proposed an improved three-dimensional dynamic region growing algorithm. The algorithm utilized one random seed point as the beginning of region growing. In order to eliminate the influences causing by the mistaken of the first seed point in traditional region growing algorithm, the algorithm combined the mean value of 26 adjacent voxels of the seed point. Meanwhile, double growing rules and dynamic thresholds were used to ensure the accuracy of segmentation. For comparison, the Dice’s Similarity Coefficient(DSC) was calculated with manually contour to assess the accuracy of this algorithm.
Results: 13 abdominal CT scans was selected to verify the proposed method. The values of DSC from automatic segmentation results and manual contour reached highest 0.982. The results showed that the accuracy of automatic liver segmentation using this algorithm was close to the results which was contoured manually. The segmentation results in this study are almost consistent with the manual contour by different experts (the mean value of DSC was 0.934), and the speed (3 seconds/per image) and the coherent in 3D directions was also improved.
Conclusion: The three-dimensional dynamic region growing algorithm which were proposed in this study performed better accuracy and efficiency in liver segmentation than manual delineation, and it could be used as an automatic liver segmentation method in radiation therapy.
Contact Email: