TY - GEN
T1 - Hyperparameter Algorithms in Electrical Impedance Tomography for Rotational Data
AU - Winkler, Simon
AU - Lehti-Polojärvi, Mari
AU - Hyttinen, Jari
N1 - JUFOID=58152
Funding Information:
M. L-P. was funded by Emil Aaltonen Foundation and Academy of Finland Center of Excellence in Body on Chip. J. H. was funded by Academy of Finland Center of Excellence in Body on Chip. Equipment used were funded by the Tekes Human Spare Parts program.
Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
Copyright:
Copyright 2020 Elsevier B.V., All rights reserved.
PY - 2021
Y1 - 2021
N2 - Rotational electrical impedance tomography provides novel possibilities for multimodal imaging. This could be especially useful in tissue engineering studies where non-destructive and label-free imaging is needed. In difference electrical impedance tomography, the change in conductivity distribution between two samples or states is reconstructed from boundary measurements. Typically, regularization is employed in the solution to tackle the ill-posedness of the problem. The amount of regularization is controlled by a hyperparameter value that is commonly found by subjective and time consuming heuristic selection. In order to find an automatized method that works with rotational data, three state-of-the-art methods for hyperparameter selection were investigated: BestRes, L-Curve and the averaged signal-to-noise ratio (SNR¯ ) as noise performance metric. These were tested with conventional and rotational experimental data. The results show that SNR¯ was the only method that provided good image quality with rotational data.
AB - Rotational electrical impedance tomography provides novel possibilities for multimodal imaging. This could be especially useful in tissue engineering studies where non-destructive and label-free imaging is needed. In difference electrical impedance tomography, the change in conductivity distribution between two samples or states is reconstructed from boundary measurements. Typically, regularization is employed in the solution to tackle the ill-posedness of the problem. The amount of regularization is controlled by a hyperparameter value that is commonly found by subjective and time consuming heuristic selection. In order to find an automatized method that works with rotational data, three state-of-the-art methods for hyperparameter selection were investigated: BestRes, L-Curve and the averaged signal-to-noise ratio (SNR¯ ) as noise performance metric. These were tested with conventional and rotational experimental data. The results show that SNR¯ was the only method that provided good image quality with rotational data.
KW - Electrical impedance tomography
KW - Hyperparameter
KW - Regularization parameter
KW - Rotational EIT
U2 - 10.1007/978-3-030-64610-3_71
DO - 10.1007/978-3-030-64610-3_71
M3 - Conference contribution
AN - SCOPUS:85097612383
SN - 978-3-030-64609-7
T3 - IFMBE Proceedings
SP - 631
EP - 643
BT - 8th European Medical and Biological Engineering Conference
A2 - Jarm, Tomaz
A2 - Cvetkoska, Aleksandra
A2 - Mahnič-Kalamiza, Samo
A2 - Miklavcic, Damijan
PB - Springer
CY - Cham
T2 - European Medical and Biological Engineering Conference
Y2 - 29 November 2020 through 3 December 2020
ER -