BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scattering Networks and Singular Values Decomposition: different m
 ethods to remove background in Single-Molecule Localization Microscopy ima
 ge - Lisa Cuneo\, University of Genova and Istituto Italiano di Tecnologia
  
DTSTART:20230721T160000Z
DTEND:20230721T170000Z
UID:TALK203422@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:In optical image formation\, a major challenge in Single-Molec
 ule Localization Microscopy (SMLM) is\nthe presence of background noise\, 
 which degrades image quality and contrast [1]. This arises from an\noverla
 p of sparse\, localized molecules with a fixed background. To address this
  issue\, we explore two\nmethods: the Scattering Network and Singular Valu
 e Decomposition (SVD).\nThe Scattering Network offers a translation-invari
 ant image representation that is stable to deformations\,\nachieved throug
 h fixed wavelet filters in a deep Convolutional Neural Network (CNN) archi
 tecture [2].\nThis representation has several advantages\, such as low com
 putational requirements and interpretability\,\nmaking it ideal for SMLM. 
 However\, it cannot take into account the temporal information present in\
 nSMLM datasets.\nTo include dynamic information\, we propose SVD as a spat
 ial-temporal representation. SVD decomposes\nthe images into temporal and 
 spatial components\, which are combined and weighted by singular values.\n
 By focusing on components associated with smaller singular values\, known 
 to be related to molecules [3]\,\nwe effectively filter out background noi
 se.\nWith the goal of removing the background from SMLM images\, we propos
 e two methods: Scattering\nNetwork and SVD. The former exploits the repres
 entation of wavelet filters\, incorporating predefined\ngeometric priors. 
 We combine the scattering networks with CNNs to separate the signal and ba
 ckground\nin the scattering representation domain and reconstruct the imag
 e. The latter exploits the spatial-temporal\nrepresentation\, incorporatin
 g also the temporal dynamics of an SMLM dataset.\nWe conducted a comprehen
 sive comparison between these two methods and state-of-the-art techniques\
 nto evaluate their performance. Overall\, our work focuses on enhancing im
 age quality and contrast in\nSMLM by addressing the background noise probl
 em using the Scattering Network and SVD. Our results\ndemonstrate the pote
 ntial of these techniques for improving molecule localization precision an
 d spatial\nresolution in SMLM images.\nReferences\n[1] H. Deschout\, F. Za
 nacchi\, M. Mlodzianoski et al\, Precisely and accurately localizing singl
 e emitters\nin fluorescence microscopy\, Nat. Methods 11\, 253–266 (2014
 ).\n[2] J. Bruna and S. Mallat\, Invariant Scattering Convolution Networks
 \, IEEE Transactions on Pattern\nAnalysis and Machine Intelligence 35 (8)\
 , 2013.\n[3] G. S. Alberti\, H. Ammari\, F. Romero and T. Wintz\, Mathemat
 ical Analysis of Ultrafast Ultrasound\nImaging\, SIAM Journal on Applied M
 athematics 77\, 1-25 (2017).\n\nhttps://cl-cam-ac-uk.zoom.us/j/94991672751
 ?pwd=eUR6TEsyazhsK3VzL2NlOWxCc3BlQT09\n
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR