Multidimensional Noise Reduction in CT scans via 2D iterative reconstruction and 3D CNN

[Multidimensional Noise Reduction in CT scans via 2D iterative reconstruction and 3D CNN]

BACKGROUND & OBJECTIVE

  • CT scan exposes people to an enormous amount of radiation but lowering radiation dose degrades image quality
  • Deep learning have shown outstanding performance but tends to lose fine details and blur out high frequency edge
  • To overcome those weakness, reduce noise through 3D CNN architecture (to consider adjacent voxels) and 2D Landweber-type iteration

PROCESS

1. 3D deep learning based noise reduction

2. 2D one-step Landweber-type iteration [1]

[1] Landweber, L. American journal of mathematics, 73(3), 615-624 (1951)