Click here to


Are you sure ?

Yes, do it No, cancel

A Population Based Statistical Model for Dose Distribution in Nasopharynx Cancer

Gang Liu1,2,3 , Xuanfeng Ding2, Zhiyong Yang1, Zhiwen Liang1, Jing Yang1, Xin Nie1, Jun Han1, Hongyuan Liu1, Mi chen1,Ting Cao1, Xiaohui Zhu1, Hong Quan3, Qin Li1* 1. Cancer Center, Union Hospital, Huazhong University of Science and Technology, Tongji Medical College, Wuhan, 430023, China; 2. Beaumont Health System, Royal Oak, MI, 48304, USA. 3. School of Physics and Technology, Wuhan University, Hubei, Wuhan, 430072, China;


(Sunday, 7/29/2018) 3:30 PM - 4:00 PM

Room: Exhibit Hall | Forum 9

Purpose: To develop a population based statistical model of the dose distribution among nasopharynx cancer(NPC) patients.

Methods: Dose distributions of 54 NPC patients were adopted and preprocessed using Velocity AI(Varian Medical System, USA)to generate the dose template library. Subsequently, the dominant modes of dose distribution were extracted using Principal Component Analysis (PCA). Dose matrix of 4 patients outside the template library were utilized to do evaluation. Dominant modes representing at least 90% of the total variability were used to reconstruct the actual dose distribution. Residual reconstruction error between the model reconstructed and actual dose distribution was calculated to compare the accuracy of the models. And the error for each voxel in region of interesting (ROI) were descried as absolute dose and the percentage relative to actual dose respectively.

Results: The first more than seven PCA components mainly described more than 90% dose distribution variations. Voxel in the region with the threshold of 40Gy, 50 Gy and 60Gy were selected as ROIs respectively. The corresponding residual reconstruction error for each voxel in the ROIs were less than 13Gy vs 31%, 12Gy vs 22% Gy and 7Gy vs 15% respectively.

Conclusion: A PCA based model for dose distribution variations in the NPC patient was developed, and its accuracy determined. Such a model can serve as a basis for probability based treatment planning in NPC cancer patients. This method can also be developed and applied to other parts.

Funding Support, Disclosures, and Conflict of Interest: This study was supported by the Natural Science Foundation of China (No. 10875092 and 31271511), the Natural Science Foundation of Hubei Province of China (No. 2012KB04449), and the Natural Science Foundation of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (No.02.03.2017-289). There are no conflicts of interest


Dose, Reconstruction


IM- Dataset analysis/biomathematics: Machine learning

Contact Email