MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Predicting Real-Time 3D Deformation Field Maps (DFM) Based On Volumetric Cine MRI (VC-MRI) and Artificial Neural Networks for On-Board 4D Target Tracking/Gating

W Harris1*, W Sun2 , J Pham1 , Z Yang1 , F Yin1 , L Ren1 , (1) Duke University Medical Center, Durham, NC, (2) School of Information Science and Engineering, Jinan

Presentations

(Wednesday, 8/1/2018) 1:45 PM - 3:45 PM

Room: Karl Dean Ballroom B1

Purpose: To predict real time 3D deformation field maps (DFMs) ahead of time using Volumetric Cine MRI (VC-MRI) and adaptive boosting and multi-layer perceptron neural network (ADMLP-NN) for 4D target tracking.

Methods: One phase of a prior 4D-MRI is set as MRIprior. Principal component analysis (PCA) is used to extract 3 major respiratory deformation patterns from the DFMs generated between the MRIprior and all other phases. VC-MRI at each time-step is considered a deformation of MRIprior, and the DFM is represented as a weighted linear combination of the PCA components. The PCA coefficients are solved by minimizing the differences between the corresponding slice of the VC-MRI and on-board 2D cine MRI acquired. PCA coefficients solved during first 90s are then inputted into an ADMLP-NN algorithm to predict 1s ahead over the next 30 seconds of PCA coefficients. The ADMLP-NN uses several identical multi-layer perception neural networks with an adaptive boosting decision algorithm. The predicted PCA curves are used to build 3D DFM to predict the VC-MRI ahead of time. The method was evaluated using a 4D computerized phantom (XCAT) with a patient-specific RPM breathing curve. The accuracy of the PCA curve prediction in each direction was evaluated. The predicted real-time 3D tumor volume was evaluated against the ground-truth tumor volume using Volume Percent Difference (VPD) and Center of Mass Shift (COMS).

Results: The average VPD/COMS for the predicted tumor volume in VC-MRI based on ADMLP-NN were 15.81±7.14%/1.20±0.65mm, respectively, across all time steps. The correlation coefficients between the predicted PCA curves and the actual PCA curves generated through VC-MRI estimation for the 1st/2nd principal components were 0.99/0.85, 0.98/0.94, and 0.88/0.89 in the SI, AP and lateral directions, respectively.

Conclusion: Preliminary studies showed the feasibility to use artificial neural networks to predict 3D DFMs to generate predicted VC-MRI for 4D target tracking/gating.

Funding Support, Disclosures, and Conflict of Interest: NIH R01-184173

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email