gms | German Medical Science

72. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC)
Joint Meeting mit der Polnischen Gesellschaft für Neurochirurgie

Deutsche Gesellschaft für Neurochirurgie (DGNC) e. V.

06.06. - 09.06.2021

Algorithms for perfusion measurements – trading speed for accuracy?

Algorithmen zur Perfusionsmessung – Geschwindigkeit gegen Genauigkeit?

Meeting Abstract

Search Medline for

  • presenting/speaker Igor Fischer - Heinrich-Heine-Universität, Universitätsklinikum, Klinik für Neurochirurgie, Düsseldorf, Deutschland
  • Daniel Hänggi - Heinrich-Heine-Universität, Universitätsklinikum, Klinik für Neurochirurgie, Düsseldorf, Deutschland
  • Marcel A. Kamp - Heinrich-Heine-Universität, Universitätsklinikum, Klinik für Neurochirurgie, Düsseldorf, Deutschland

Deutsche Gesellschaft für Neurochirurgie. 72. Jahrestagung der Deutschen Gesellschaft für Neurochirurgie (DGNC), Joint Meeting mit der Polnischen Gesellschaft für Neurochirurgie. sine loco [digital], 06.-09.06.2021. Düsseldorf: German Medical Science GMS Publishing House; 2021. DocP181

doi: 10.3205/21dgnc462, urn:nbn:de:0183-21dgnc4622

Published: June 4, 2021

© 2021 Fischer et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

Objective: Singular value decomposition (SVD) is a fast and widely used algorithm for computing blood perfusion images from pCT, MRT, and ICG data. However, its speed is known to come at a price: The results, while mathematically correct, are not guaranteed to be physically plausible. We investigate four alternative algorithms for speed and accuracy and compare them with the SVD algorithm.

Methods: In perfusion imaging modalities, perfusion values (Tmax, rCBF, rCBV, and MTT) are computed per pixel from the residue function. The residue function is non-negative and, for the most part, monotonously decreasing, and has to be found numerically. SVD is not guaranteed to find solutions satisfying physical plausibility. Alternatively, solution can be found by pre-imposing a certain form on the residue function and/or using constrained optimisation. These methods are computationally more intensive.

We investigated two imposed functional forms, bi-exponential (BE) and opposing logistics (OL), and applied two optimisation algorithms on each of them: unconstrained and with physically plausible constraints. We compared the performance to the SVD algorithm.

All five algorithms were tested on a random set of pixels taken from intra-operative ICG videos of 11 patients with different diagnoses (SAH, trauma etc.). We measured 1) the execution time, 2) the average number of physically implausible results, 3) the mean error in reconstructing the contrast agent function, and, 4-7, the mean error in Tmax, rCBF, rCBV, and MTT. As the true value of these parameters cannot be known, we used the values computed by the cBE algorithm as the gold standard.

Results: SVD was, with 0.0017 s per sample image, by far the fastest method, while the constrained BE was the slowest (3.74 s). Regarding result plausibility, SVD had a 72% chance of producing physically impossible results, while both constrained algorithms were guaranteed to produce realistic results. However, in terms of clinically relevant parameters (Tmax, MTT etc.), OL turned to be the most accurate.

Conclusion: Imposing a functional form on the residue function can lead to more accurate results, but at high computationals costs (a factor of 100 or more). In clinical practice, these are unlikely to be justified. For real-time applications, SVD still seems to be the best compromise solution.