MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. Daily Edge Deformation Prediction Using Artificial Neural Network Regression for Prior Contour Based Total Variation Reconstruction (PCTV-ANN) for Low Dose CBCT

Daily Edge Deformation Prediction Using Artificial Neural Network Regression for Prior Contour Based Total Variation Reconstruction (PCTV-ANN) for Low Dose CBCT

Y Chen1*, K Lafata2 , F Yin3 , L Ren4 , (1) Duke University, Durham, NC, (2) Duke University, Durham, NC, (3) Duke University Medical Center, Durham, NC, (4) Duke University Medical Center, Durham, NC

Presentations

(Monday, 7/30/2018) 1:15 PM - 1:45 PM

Room: Exhibit Hall | Forum 5

Purpose: The image quality of low-dose CBCT can be improved using prior contour based total variation (PCTV) method. However, deformation of on-board edge information from prior contour is time consuming. This study aims to develop a novel method for predicting daily on-board edge deformation using artificial neural network (ANN) to improve the PCTV reconstruction efficiency.

Methods: The new method involved patch-based ANN deformation prediction and PCTV reconstruction. Full-sampling CBCTs and retrospectively under-sampled low-dose CBCT are acquired in the first few days to train the model. Then the model predicts deformation vector field (DVF) for low-dose CBCT acquired in the following days to generate on-board contours for PCTV reconstruction. Specifically, in ANN training, patches are extracted from the same location in the planning-CT and low-dose CBCT. Only voxels at the edge region of planning-CT are selected for the prediction model and a 3D patch is applied. Patch similarity based on normalized cross correlation, image intensity, and six histogram features of each patch are the inputs to training model. DVF registered between planning-CT and full-sampling CBCT is used as output. The ANN is developed in MATLAB using 4 hidden layers with 20 neurons each, and weights are optimized based on Levenberg-Marquardt minimization. The method is evaluated using lung SBRT patient data. The first two days’ CBCTs are used for ANN training. 45 half-fan projections covering 360˚ from 3rd day CBCT is used for reconstruction. Results from Edge-preserving TV (EPTV), PCTV and PCTV-ANN are compared.

Results: The cross-correlation between edge map predicted by ANN and reference is 0.984. PCTV-ANN enhanced bone edges in CBCT compared to EPTV and achieved comparable image quality as PCTV while avoiding time-consuming deformable registration process.

Conclusion: Preliminary results demonstrated the daily deformation of on-board edge information could potentially be predicted using ANN for PCTV based low-dose CBCT reconstruction.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by NIH grant R01 CA-184173.

Keywords

Reconstruction, Cone-beam CT, Deformation

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links