Click here to


Are you sure ?

Yes, do it No, cancel

Two-Step Subspace Mapping Based Diaphragm Displacement Prediction by Markerless Abdominal Surface Measurement

H Yu*1,2, E Zhang2, S Yu2, Z Yang2, L Ma2, M Chen2, X Gu2, W Lu2, (1) Xidian University, Shaanxi, China, (2) UT Southwestern Medical Center, Dallas, TX


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

To investigate the movement correlation between the internal diaphragm and the external abdominal surface without markers, and then to predict the diaphragm displacement in real-time for surface-guided gating in radiotherapy.

24 liver cancer patients are selected in this study. Each patient has a 4D CTs with 10 phases acquired during radiotherapy simulation. For each CT volume, lungs and whole body are segmented, and the diaphragm and the abdominal surface are identified. The mass centers of the surfaces are computed to quantify the surfaces’ displacement. To solve the cross-domain prediction problem, we propose a novel two-step subspace mapping algorithm (TSSM). Considering the different distribution structures of the two organs’ movement, we utilize the principal component analysis to construct the eigenspace for each group of data. Then, a subspace mapping is obtained by a linear ridge optimization process to connect the diaphragm data with the abdominal surface data. In order to investigate the non-linear correlation, TSSM is further extended to kernel TSSM (kTSSM) which includes the polynomial kernel and Gaussian kernel. In experiments, the Gaussian noise is added to test the robustness of the proposed algorithm.

The performance of the proposed algorithm are evaluated with various metrics, including MSE, R2, and MAPE. Experiments validate the effectiveness of the proposed method in term of accuracy and robustness, where the true prediction error can be reduced to 0.19±0.13mm.

The 3D image segmentation can accurately locate the diaphragm and the abdominal surface without markers. The kTSSM with polynomial kernel obtains the best prediction performance. The prediction of the linear model TSSM has low standard deviation, especially good for the data without noise. This method has potential for improving the timing accuracy of surface-guided gating in radiotherapy.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by NIH R01 CA235723 and the seed grant of Radiation Oncology Department at University of Texas Southwestern Medical Center.


Image Analysis, Optimization, Organ Motion


TH- RT Interfraction Motion Management: Development (new technology and techniques)

Contact Email