Generative modeling of living cells with SO(3)-equivariant
implicit neural representations

J 2024

Generative modeling of living cells with SO(3)-equivariant implicit neural representations

WIESNER, David, Julian SUK, Sven DUMMER, Tereza NEČASOVÁ, Ulman VLADIMÍR et. al.

Basic information

Original name

Generative modeling of living cells with SO(3)-equivariant implicit neural representations

Authors

WIESNER, David (203 Czech Republic, belonging to the institution), Julian SUK (528 Netherlands), Sven DUMMER (528 Netherlands), Tereza NEČASOVÁ (203 Czech Republic, belonging to the institution), Ulman VLADIMÍR (203 Czech Republic), David SVOBODA (203 Czech Republic, guarantor, belonging to the institution) and Jelmer WOLTERINK (528 Netherlands)

Edition

Medical Image Analysis, Netherlands, Elsevier, 2024, 1361-8415

Other information

Language

English

Type of outcome

Článek v odborném periodiku

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

Switzerland

Confidentiality degree

není předmětem státního či obchodního tajemství

References:

URL

Impact factor

Impact factor: 10.900 in 2022

Organization unit

Faculty of Informatics

DOI

http://dx.doi.org/10.1016/j.media.2023.102991

UT WoS

001171218800001

Keywords in English

cell shape modeling; neural networks; implicit neural representations; signed distance function; generative model; interpolation

Abstract

V originále

Data-driven cell tracking and segmentation methods in biomedical imaging require diverse and information-rich training data. In cases where the number of training samples is limited, synthetic computer-generated data sets can be used to improve these methods. This requires the synthesis of cell shapes as well as corresponding microscopy images using generative models. To synthesize realistic living cell shapes, the shape representation used by the generative model should be able to accurately represent fine details and changes in topology, which are common in cells. These requirements are not met by 3D voxel masks, which are restricted in resolution, and polygon meshes, which do not easily model processes like cell growth and mitosis. In this work, we propose to represent living cell shapes as level sets of signed distance functions (SDFs) which are estimated by neural networks. We optimize a fully-connected neural network to provide an implicit representation of the SDF value at any point in a 3D+time domain, conditioned on a learned latent code that is disentangled from the rotation of the cell shape. We demonstrate the effectiveness of this approach on cells that exhibit rapid deformations (Platynereis dumerilii), cells that grow and divide (C. elegans), and cells that have growing and branching filopodial protrusions (A549 human lung carcinoma cells). A quantitative evaluation using shape features and Dice similarity coefficients of real and synthetic cell shapes shows that our model can generate topologically plausible complex cell shapes in 3D+time with high similarity to real living cell shapes. Finally, we show how microscopy images of living cells that correspond to our generated cell shapes can be synthesized using an image-to-image model.

Links

LM2023050, research and development project

Name: Národní infrastruktura pro biologické a medicínské zobrazování

Investor: Ministry of Education, Youth and Sports of the CR, Czech BioImaging: National research infrastructure for biological and medical imaging

MUNI/A/1081/2022, interní kód MU

Name: Modelování, analýza a verifikace (2023)

Investor: Masaryk University

Citovat

WIESNER, David, Julian SUK, Sven DUMMER, Tereza NEČASOVÁ, Ulman VLADIMÍR, David SVOBODA and Jelmer WOLTERINK. Generative modeling of living cells with SO(3)-equivariant implicit neural representations. Medical Image Analysis. Netherlands: Elsevier, 2024, vol. 2024, No 91, p. 102991-103008. ISSN 1361-8415. Available from: https://dx.doi.org/10.1016/j.media.2023.102991.

@article{2324238,
   author = {Wiesner, David and Suk, Julian and Dummer, Sven and Nečasová, Tereza and Vladimír, Ulman and Svoboda, David and Wolterink, Jelmer},
   article_location = {Netherlands},
   article_number = {91},
   doi = {http://dx.doi.org/10.1016/j.media.2023.102991},
   keywords = {cell shape modeling; neural networks; implicit neural representations; signed distance function; generative model; interpolation},
   language = {eng},
   issn = {1361-8415},
   journal = {Medical Image Analysis},
   title = {Generative modeling of living cells with SO(3)-equivariant implicit neural representations},
   url = {https://doi.org/10.1016/j.media.2023.102991},
   volume = {2024},
   year = {2024}
}

TY  - JOUR
ID  - 2324238
AU  - Wiesner, David - Suk, Julian - Dummer, Sven - Nečasová, Tereza - Vladimír, Ulman - Svoboda, David - Wolterink, Jelmer
PY  - 2024
TI  - Generative modeling of living cells with SO(3)-equivariant implicit neural representations
JF  - Medical Image Analysis
VL  - 2024
IS  - 91
SP  - 102991-103008
EP  - 102991-103008
PB  - Elsevier
SN  - 13618415
KW  - cell shape modeling
KW  - neural networks
KW  - implicit neural representations
KW  - signed distance function
KW  - generative model
KW  - interpolation
UR  - https://doi.org/10.1016/j.media.2023.102991
N2  - Data-driven cell tracking and segmentation methods in biomedical imaging require diverse and information-rich training data. In cases where the number of training samples is limited, synthetic computer-generated data sets can be used to improve these methods. This requires the synthesis of cell shapes as well as corresponding microscopy images using generative models. To synthesize realistic living cell shapes, the shape representation used by the generative model should be able to accurately represent fine details and changes in topology, which are common in cells. These requirements are not met by 3D voxel masks, which are restricted in resolution, and polygon meshes, which do not easily model processes like cell growth and mitosis. In this work, we propose to represent living cell shapes as level sets of signed distance functions (SDFs) which are estimated by neural networks. We optimize a fully-connected neural network to provide an implicit representation of the SDF value at any point in a 3D+time domain, conditioned on a learned latent code that is disentangled from the rotation of the cell shape. We demonstrate the effectiveness of this approach on cells that exhibit rapid deformations (Platynereis dumerilii), cells that grow and divide (C. elegans), and cells that have growing and branching filopodial protrusions (A549 human lung carcinoma cells). A quantitative evaluation using shape features and Dice similarity coefficients of real and synthetic cell shapes shows that our model can generate topologically plausible complex cell shapes in 3D+time with high similarity to real living cell shapes. Finally, we show how microscopy images of living cells that correspond to our generated cell shapes can be synthesized using an image-to-image model.
ER  -

WIESNER, David, Julian SUK, Sven DUMMER, Tereza NEČASOVÁ, Ulman VLADIMÍR, David SVOBODA and Jelmer WOLTERINK. Generative modeling of living cells with SO(3)-equivariant implicit neural representations. \textit{Medical Image Analysis}. Netherlands: Elsevier, 2024, vol.~2024, No~91, p.~102991-103008. ISSN~1361-8415. Available from: https://dx.doi.org/10.1016/j.media.2023.102991.

Detailed Information on Publication Record