Electromagnetic relays used for power system .pptx
Convolutional neural network based metal artifact reduction in x ray computed tomography
1. Convolutional Neural Network Based
Metal Artifact Reduction in X-Ray
Computed Tomography
Yanbo Zhang, Hengyong Yu
Department of Electrical and Computer Engineering, University of Massachusetts at Lowell
IEEE TRANSACTIONS ON MEDIAL IMAGING, June 2018
April 11, 2019
SMC AI Research Center
Kyuri Kim
2. Introduction
- Metal Artifact
• Large number of metal artifact reduction (MAR) method have been
proposed during the past four decades, there is still no standard solution.
6. Experiments
1.3 CNN
• Training dataset
• Original + BHC + LI 3 channel Image
• 64*64 image patch (Original 512*512 size)
• 1000 patch samples
7. Experiments
1.4 Tissue Processing
After tissue processing
Ex)
• Although the metal artifacts are significantly reduced after the CNN processing,
remaining artifacts are still considerable.
• Because the water equivalent tissues have similar attenuations and are accounted for
a dominate proportion in a patient, we assign these pixels with uniform value to
remove most artifacts and obtain a CNN prior image.
• By the K-means clustering on the CNN images, two thresholds are automatically
determined and the CNN image is segmented into bone, water and air.
8. Experiments
1.5 CNN-MAR Method
• Clear that the region of water equivalent tissue are flat and the artifacts are removed.
• Simultaneously, the bony structure are preserved very well.
MAR phase
10. Result & Conclusion
- Real Data: Surgical Clip
• Clear that the region of water equivalent
tissue are flat and the artifacts are removed.
• Simultaneously, the bony structure are
preserved very well.
- Results obtained by directly adopting a CNN
image as the prior image with the tissue
processing step.
11. Result & Conclusion
- Different channel of input images
Two-channel : Original + LI
Three-channel : Original + LI + BHC
Five-channel : Original + LI + BHC + NMAR1 + NMAR2