[home] [Personal Program] [Help]

---

Session: Poster session 2 (Odd numbers)
Session starts: Friday 27 January, 10:00
Presentation starts: 10:00

---

Metin Calis (Delft University of Technology)
Alle-Jan van der Veen (Delft University of Technology)
Massimo Mischi (Eindhoven University of Technology)
Bori Hunyadi (Delft University of Technology)

Abstract:
In most ultrasound despeckling applications, the noise is considered multiplicative and Rayleigh shaped. Commonly the logarithm is taken, transforming the multiplicative noise into additive noise. According to [1], the logarithm of the noise follows Fisher-Tippet distribution, which can be approximated as white gaussian noise with outliers. In our previous work [2], we assumed that the noise was approximately white gaussian and recovered the clean ultrasound data as a low-rank approximation of the noisy observations via multilinear singular value decomposition (MLSVD). In a simulation study, we show that the true rank of the underlying signal could be approximated most of the time correctly. On the in-vivo recording of 6 patients, we showed that truncated MLSVD led to a better classification of cancerous voxels using dispersion and perfusion features. However, MLSVD is not optimal in the least-squared sense and is not robust against outliers. The estimate of the underlying low-rank tensor can be improved with prior information about the noise statistics. Therefore, in this work, we improve our signal model, assuming that the received signal is the summation of a low-rank tensor, white gaussian noise, and sparse outliers. In order to recover the clean contrast-enhanced ultrasound data based on this signal model, tensor stable component pursuit or non-convex outliner resistant tensor decompositions techniques can be used. We will compare the performance of these approaches with MLSVD and show the added benefit of the prior information about the noise statistics. [1] Michailovich, O. V., & Tannenbaum, A. (2006). Despeckling of medical ultrasound images. IEEE transactions on ultrasonics, ferroelectrics, and frequency control, 53(1), 64–78. https://doi.org/10.1109/tuffc.2006.1588392 [2] Calis, M., Mischi, M., van der Veen, A. J., & Hunyadi, B. (2022). Denoising of Dynamic Contrast-enhanced Ultrasound Sequences: A Multilinear Approach. In BIOSIGNALS (pp. 192-199).