Evaluation of Deformation Image Registration Algorithms Based On Deformation Vector Field for Radiotherapy
E Li1,2, Z Zhong1,2, Y An1,2, S Huang1, W Zheng1,3, J Lian1,4, X Yang1*, (1) Sun Yat-sen University Cancer Center; State Key Laboratory of Oncology in South China; Collaborative Innovation Center for Cancer Medicine; Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangzhou, Guangdong province, 510060, China,(2) Guangzhou Xinhua College, Guangzhou, Guangdong, 510520, China,(3) Department of Radiation Oncology, Southern Theater Air Force Hospital of the People's Liberation Army, Guangzhou, Guangdong province, 510050, China,(4) Department of Radiation Oncology, The First Affiliated Hospital of Guangzhou University of Traditional Chinese Medicine, Guangzhou, Guangdong province, 510405, China
Presentations
PO-GePV-M-174 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To investigate the quality, accuracy, and plausibility of the Deformable Image Registration (DIR) algorithm by using various Deformation Vector Field (DVF) indexes.
Methods: In both lung and nasopharyngeal radiotherapy cases, DVF are constructed based on different DIR algorithms. Firstly, the DVF evaluation includes the Jacobian determinant mentioned by AAPM-TG132 and the inverse consistency error (ICE). Secondly, we extended the evaluation metrics into the image similarity index. Both mutual information (MI), mean squared error (MSE) is also introduced to assist the evaluation. Furthermore, based on the visual evaluation of difference graph in DIR, we also considered a variety of imaging results, including the visualization of the Jacobian, the 3D volume, and the error histogram of the ICE. Finally, the performance of different DIR was evaluated by measuring the DVF as a quantitative index.
Results: In the comparison between the optical flow (OF) algorithm and the Demon algorithm, the difference graph before and after registration shows that the OF algorithm is better than the Demon algorithm, the general accuracy limit of the Jacobian mean is closer to 1, which means the DVF performs better. The mean standard deviation of the global OF algorithm is about 34.5% closer to the accuracy limit than that of the Demon algorithm. What’s more, the OF algorithm has an error of less than 46% in the inverse consistency error. And the visualization and inverse consistency errors of 3D volume also show that the OF algorithm is better than the Demon algorithm.
Conclusion: The work shows the potential of DVF index in DIR algorithm evaluation. In the evaluation of registration error, Jacobian determinant and inverse consistency error have certain accuracy and reliability. The indexes quantify the accuracy of task-specific registration and recommend appropriate algorithms for doctors and radiotherapy physicists as a key part of implementing radiotherapy plans.
Keywords
Taxonomy
IM/TH- Image Registration: General (Most aspects)
Contact Email