Postdoctoral Scholar,
Dept. of Scientific Computing,
Florida State University
"A Deep-CGO Method for Electrical Impedance Tomography"
Sep 1, 2021 Schedule:
- Tea Time - F2F ( 417 DSL) / Virtual ( Zoom)
- 03:00 to 03:30 PM Eastern Time (US and Canada)
- Colloquium - F2F ( 499 DSL) / Virtual ( Zoom)
- 03:30 to 04:30 PM Eastern Time (US and Canada)
Meeting # 942 7359 5552
Abstract:
Complex Geometrical Optics (CGO) methods are a class of reconstruction methods for the noninvasive imaging modality Electrical Impedance Tomography (EIT), of which the D-bar method is best known. In 2D, D-bar reconstructions for EIT are known to be robust to modeling errors but inherently blurry/smooth. Post-processing the images via a trained Convolutional Neural Network (CNN) shows great promise. Re- cently, the 3D conductivity reconstructions from linearized CGO methods (Calderon and t-exp) were shown to be much faster (approx 5 seconds) versus traditional optimization-based approaches (approx 12 hrs). The speed of the linearized CGO methods holds great promise for real-time imaging but suffers again from smoothed reconstruction. This talk extends the CNN post-processing approach from 2D Deep-D-bar to the 3D setting by training a CNN to learn the blurring inherent in the “Born approximation” (t-exp) linearized CGO method for 3D EIT. This approach, relying on the U-Net architecture, produces significant improve- ments in image quality.
