Automatic Delineation of Gross Tumor Volume (GTV) for Non-Small Cell Lung Cancer (NSCLC) On PET/CT Images Using Fuzzy Markov Random Field (MRF) Model
Y Guo1*, J Sun2, W Lin3, P Wang4, Y Feng5, (1) Tianjin University, Tianjin, (2) Tianjin Medical University Cancer Institute and Hospital, Tianjin, (3) Tianjin University, Tianjin, (4) Tianjin Medical University Cancer Institute and Hospital, Tianjin, (5) Tianjin University, Tianjin Medical University Cancer Institute and Hospital, Tianjin, East Carolina University, Greenville, NCSU-E-J-111 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: To develop a robust method for automatic delineation of GTV for the lung tumors on fused PET/CT images.
Methods: The new method is based on fuzzy Markov random field model. The combination of PET and CT image information is achieved by using a proper joint posterior probability distribution of observed features in the fuzzy MRF model which performs better than the commonly used Gaussian joint distribution. The parameters of the joint posterior distribution are obtained by fitting the histogram of the region obtained from C-means clustering and the final membership degree of each voxel to tumor class is estimated with a gradient decent method.
The PET and CT simulation images of 3 NSCLC patients were used. Fusion results using MIM 5.2 were exported to an in-house-developed software for the study. GTV delineation with the proposed new method and manual method by an experienced radiation oncologist on the fused images were performed, respectively. The manually contoured GTVs were checked and confirmed by another experienced radiation oncologist. The robustness of the new method was evaluated by comparing the overlap of the two delineations using Dice similarity coefficient (DSC) expressed as 2x(Intersection of V1 and V2)/(V1+V2) where V1 is GTVmanual, V2 is GTVauto.
Results: GTVs obtained with the two methods were similar and DSC was 0.85±0.013, which showed the advantage of the new method that utilized the information extracted effectively from both PET and CT images in the algorithm.
Conclusion: It has been shown that effective and automatic GTV delineation can be achieved with this method for lung tumors located near other organs with similar intensities in PET and CT images (such as when the tumors extends into the chest wall or the mediastinum). The reliability of this method will be further tested and more data will be presented.
Funding Support, Disclosures, and Conflict of Interest: National Science Foundation of China (NSFC-81201148,81171342,31000784)