liver ver sc scan ans s vi via a a re a recu current rrent - - PowerPoint PPT Presentation

Sp Spati tio-te tempo mporal ral mot

• tion
• n predic

diction tion in free-bre breathing athing liver ver sc scan ans s vi via a a re a recu current rrent multi ti-scale scale en enco coder er dec ecod

• der

er

Liset Vázquez Romaguera1, Rosalie Plantefève2, Samuel Kadoury1,2

1 Polytechnique Montreal, Montreal, Canada 2 CHUM Research Center, Montreal, Canada

Context

• Technological

advances in image-guided radiotherapy have motivated the use of image surrogates to drive motion models

• Motion extrapolation is essential to cope with

system latencies

2

Goal: To propose a classification-based multi-scale (MS) model for spatio-temporal 2D motion prediction Breathing-induced

• rgan deformation

Issue during radiation therapy Motion compensation

Overall pipeline

3

Data acquisition Image registration Motion encoding Predictive model Predicted sequence Recovered motion fields Training aining sta tage ge Testing sting sta tage ge

1 1http://cmictig.cs.ucl.ac.uk/wiki/index.php/NiftyReg

Sagittal slices 12 volunteers Spatial resolution 1.7×1.7 mm2 Temporal resolution 320 ms Learning framework Codebook

…

I0 I1 In … Zn Zn+1 … Zn+T

labels

Proposed model

Conv: Convolution BN: Batch normalization AP: Average pooling MP: Max pooling UC: Up-convolution C: Concatenation

32 | 1 64 | 2 128 | 2 256 | 2 64 | 1 64 | 2 32 | 2 32 | 1 32 | 2 32 | 1 25 | 1 64 | 1 64 | 1

Conv/BN/ReLU MS block/MP Conv LSTM Up-conv Conv/BN/Softmax

… …

I1 I2 In Zn Zn+1 Zn+T

# filters | stride # filters | stride # filters | stride # filters | stride # filters | stride Input tensor Output tensor + C C C AP AP AP AP MP MP MP Conv Conv UC UC UC

Multi-scale block

encoder decoder recurrent units

4

Results

5

Qualitative results and Conclusion

6

Ground-truth Prediction

• Limitations: inability to cope with out-of-plane motion
• Future work: regularization on predicted displacement fields

Th Than anks ks for

• r yo

your at atte tention tion

Acknowledgments to: