University of Cambridge > > Artificial Intelligence Research Group Talks (Computer Laboratory) > Scattering Networks and Singular Values Decomposition: different methods to remove background in Single-Molecule Localization Microscopy image

Scattering Networks and Singular Values Decomposition: different methods to remove background in Single-Molecule Localization Microscopy image

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Pietro Lio.

hybrid (zoom link at the end of the abstract)

In optical image formation, a major challenge in Single-Molecule Localization Microscopy (SMLM) is the presence of background noise, which degrades image quality and contrast [1]. This arises from an overlap of sparse, localized molecules with a fixed background. To address this issue, we explore two methods: the Scattering Network and Singular Value Decomposition (SVD). The Scattering Network offers a translation-invariant image representation that is stable to deformations, achieved through fixed wavelet filters in a deep Convolutional Neural Network (CNN) architecture [2]. This representation has several advantages, such as low computational requirements and interpretability, making it ideal for SMLM . However, it cannot take into account the temporal information present in SMLM datasets. To include dynamic information, we propose SVD as a spatial-temporal representation. SVD decomposes the images into temporal and spatial components, which are combined and weighted by singular values. By focusing on components associated with smaller singular values, known to be related to molecules [3], we effectively filter out background noise. With the goal of removing the background from SMLM images, we propose two methods: Scattering Network and SVD . The former exploits the representation of wavelet filters, incorporating predefined geometric priors. We combine the scattering networks with CNNs to separate the signal and background in the scattering representation domain and reconstruct the image. The latter exploits the spatial-temporal representation, incorporating also the temporal dynamics of an SMLM dataset. We conducted a comprehensive comparison between these two methods and state-of-the-art techniques to evaluate their performance. Overall, our work focuses on enhancing image quality and contrast in SMLM by addressing the background noise problem using the Scattering Network and SVD . Our results demonstrate the potential of these techniques for improving molecule localization precision and spatial resolution in SMLM images. References [1] H. Deschout, F. Zanacchi, M. Mlodzianoski et al, Precisely and accurately localizing single emitters in fluorescence microscopy, Nat. Methods 11, 253–266 (2014). [2] J. Bruna and S. Mallat, Invariant Scattering Convolution Networks, IEEE Transactions on Pattern Analysis and Machine Intelligence 35 (8), 2013. [3] G. S. Alberti, H. Ammari, F. Romero and T. Wintz, Mathematical Analysis of Ultrafast Ultrasound Imaging, SIAM Journal on Applied Mathematics 77, 1-25 (2017).

This talk is part of the Artificial Intelligence Research Group Talks (Computer Laboratory) series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2024, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity