4D Semantic Cardiac Magnetic Resonance Image Synthesis on XCAT - - PowerPoint PPT Presentation

4d semantic cardiac magnetic resonance
SMART_READER_LITE
LIVE PREVIEW

4D Semantic Cardiac Magnetic Resonance Image Synthesis on XCAT - - PowerPoint PPT Presentation

4D Semantic Cardiac Magnetic Resonance Image Synthesis on XCAT Anatomical Model S. Abbasi-Sureshjani 1 , S. Amirrajab 1 , C. Lorenz 2 , J. Weese 2 , J. Pluim 1 , M. Breeuwer 1,3 1 Eindhoven University of Technology, Eindhoven, The Netherlands 2


slide-1
SLIDE 1

4D Semantic Cardiac Magnetic Resonance Image Synthesis on XCAT Anatomical Model

  • S. Abbasi-Sureshjani1, S. Amirrajab1, C. Lorenz2, J. Weese2, J. Pluim1, M. Breeuwer1,3

1Eindhoven University of Technology, Eindhoven, The Netherlands 2Philips Research Laboratories, Hamburg, Germany 3Philips Healthcare, Best, The Netherlands

slide-2
SLIDE 2

Motivation

Expensive data acquisition Rare properly annotated data Subjective annotations Restrictive sharing policy Physic-based image simulation & Data-driven image synthesis To develop new algorithms To augment the data To validate and benchmark To improve domain generalization and adaptation

2

slide-3
SLIDE 3

Controlling anatomical content and style

Limitations in Image Synthesis

3

High dimensional data reflecting motion and volumetric changes Anatomically and physiologically plausible images

slide-4
SLIDE 4

XCAT: eXtended Cardiac and Torso computerized human phantom

❖ Controllable 4D voxelized

heart model:

❖ scaling factors in 3D ❖ orientation & translation ❖ cardiac cycle timing ❖ etc.

4

Segars W. et al.: 4D XCAT phantom for multimodality imaging research. Medical physics, 37 (9), 4902–4915 (2010)

slide-5
SLIDE 5

Conditional Image Synthesis

5

Generator

Real Synthetic image Label map

Discriminator

slide-6
SLIDE 6

Conditional Image Synthesis

6

Park T. et al., Semantic image synthesis with spatially-adaptive normalization. CVPR, pages 2332–2341, 2019.

SPADE: SPatially-Adaptive (DE)normalization

slide-7
SLIDE 7

Method Overview

Training Inference 4D Rendered XCAT 4D Labeled Synthetic XCAT 4D Voxelized XCAT Label 4D Synthetic XCAT 2D Synthetic ACDC SPADE-GAN ACDC Cardiac Label ACDC Cardiac Data SPADE-GAN

7

Bernard O. et al.: Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: Is the problem solved? TMI, 37(11):2514–2525, Nov 2018.

slide-8
SLIDE 8

3D+t Image Synthesis: 25 time frames for 18 slice locations

8

Apex Base

slide-9
SLIDE 9

Stylized 4D labeled Synthetic Dataset

w/o IN w/ IN

Style Image Synthetic images

9

slide-10
SLIDE 10

Summary

❖ 4D labeled cardiac synthetic MR images ❖ Wide range of anatomical and style variations ❖ Inconsistencies in the background ❖ Future work: ❖ Improving image synthesis ❖ Quantitative evaluation ❖ Generating a large virtual population ❖ MICCAI2020:

❖ XCAT-GAN for Synthesizing 3D Consistent Labeled Cardiac MR

Images on Anatomically Variable XCAT Phantoms

10