Rotation equivariant and invariant neural networks for microscopy image analysis

Benjamin Chidester, Tianming Zhou, Minh N. Do, Jian Ma

Research output: Contribution to journalArticle

Abstract

Motivation: Neural networks have been widely used to analyze high-throughput microscopy images. However, the performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Highly relevant to the goal of automated cell phenotyping from microscopy image data is rotation invariance. Here we consider the application of two schemes for encoding rotation equivariance and invariance in a convolutional neural network, namely, the group-equivariant CNN (G-CNN), and a new architecture with simple, efficient conic convolution, for classifying microscopy images. We additionally integrate the 2D-discrete-Fourier transform (2D-DFT) as an effective means for encoding global rotational invariance. We call our new method the Conic Convolution and DFT Network (CFNet). Results: We evaluated the efficacy of CFNet and G-CNN as compared to a standard CNN for several different image classification tasks, including simulated and real microscopy images of subcellular protein localization, and demonstrated improved performance. We believe CFNet has the potential to improve many high-throughput microscopy image analysis applications.

Original languageEnglish (US)
Article numberbtz353
Pages (from-to)i530-i537
JournalBioinformatics
Volume35
Issue number14
DOIs
StatePublished - Jul 15 2019

Fingerprint

Image Analysis
Microscopy
Equivariant
Image analysis
Microscopic examination
Invariance
Convolution
Discrete Fourier transforms
Neural Networks
Neural networks
Invariant
Encoding
High Throughput
Throughput
Rotation Invariance
Equivariance
Discrete Fourier transform
Image classification
Image Classification
Fourier Analysis

ASJC Scopus subject areas

  • Statistics and Probability
  • Biochemistry
  • Molecular Biology
  • Computer Science Applications
  • Computational Theory and Mathematics
  • Computational Mathematics

Cite this

Rotation equivariant and invariant neural networks for microscopy image analysis. / Chidester, Benjamin; Zhou, Tianming; Do, Minh N.; Ma, Jian.

In: Bioinformatics, Vol. 35, No. 14, btz353, 15.07.2019, p. i530-i537.

Research output: Contribution to journalArticle

Chidester, Benjamin ; Zhou, Tianming ; Do, Minh N. ; Ma, Jian. / Rotation equivariant and invariant neural networks for microscopy image analysis. In: Bioinformatics. 2019 ; Vol. 35, No. 14. pp. i530-i537.
@article{e891d2b101dd4ed5800e691c77b9e870,
title = "Rotation equivariant and invariant neural networks for microscopy image analysis",
abstract = "Motivation: Neural networks have been widely used to analyze high-throughput microscopy images. However, the performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Highly relevant to the goal of automated cell phenotyping from microscopy image data is rotation invariance. Here we consider the application of two schemes for encoding rotation equivariance and invariance in a convolutional neural network, namely, the group-equivariant CNN (G-CNN), and a new architecture with simple, efficient conic convolution, for classifying microscopy images. We additionally integrate the 2D-discrete-Fourier transform (2D-DFT) as an effective means for encoding global rotational invariance. We call our new method the Conic Convolution and DFT Network (CFNet). Results: We evaluated the efficacy of CFNet and G-CNN as compared to a standard CNN for several different image classification tasks, including simulated and real microscopy images of subcellular protein localization, and demonstrated improved performance. We believe CFNet has the potential to improve many high-throughput microscopy image analysis applications.",
author = "Benjamin Chidester and Tianming Zhou and Do, {Minh N.} and Jian Ma",
year = "2019",
month = "7",
day = "15",
doi = "10.1093/bioinformatics/btz353",
language = "English (US)",
volume = "35",
pages = "i530--i537",
journal = "Bioinformatics",
issn = "1367-4803",
publisher = "Oxford University Press",
number = "14",

}

TY - JOUR

T1 - Rotation equivariant and invariant neural networks for microscopy image analysis

AU - Chidester, Benjamin

AU - Zhou, Tianming

AU - Do, Minh N.

AU - Ma, Jian

PY - 2019/7/15

Y1 - 2019/7/15

N2 - Motivation: Neural networks have been widely used to analyze high-throughput microscopy images. However, the performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Highly relevant to the goal of automated cell phenotyping from microscopy image data is rotation invariance. Here we consider the application of two schemes for encoding rotation equivariance and invariance in a convolutional neural network, namely, the group-equivariant CNN (G-CNN), and a new architecture with simple, efficient conic convolution, for classifying microscopy images. We additionally integrate the 2D-discrete-Fourier transform (2D-DFT) as an effective means for encoding global rotational invariance. We call our new method the Conic Convolution and DFT Network (CFNet). Results: We evaluated the efficacy of CFNet and G-CNN as compared to a standard CNN for several different image classification tasks, including simulated and real microscopy images of subcellular protein localization, and demonstrated improved performance. We believe CFNet has the potential to improve many high-throughput microscopy image analysis applications.

AB - Motivation: Neural networks have been widely used to analyze high-throughput microscopy images. However, the performance of neural networks can be significantly improved by encoding known invariance for particular tasks. Highly relevant to the goal of automated cell phenotyping from microscopy image data is rotation invariance. Here we consider the application of two schemes for encoding rotation equivariance and invariance in a convolutional neural network, namely, the group-equivariant CNN (G-CNN), and a new architecture with simple, efficient conic convolution, for classifying microscopy images. We additionally integrate the 2D-discrete-Fourier transform (2D-DFT) as an effective means for encoding global rotational invariance. We call our new method the Conic Convolution and DFT Network (CFNet). Results: We evaluated the efficacy of CFNet and G-CNN as compared to a standard CNN for several different image classification tasks, including simulated and real microscopy images of subcellular protein localization, and demonstrated improved performance. We believe CFNet has the potential to improve many high-throughput microscopy image analysis applications.

UR - http://www.scopus.com/inward/record.url?scp=85068935286&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068935286&partnerID=8YFLogxK

U2 - 10.1093/bioinformatics/btz353

DO - 10.1093/bioinformatics/btz353

M3 - Article

C2 - 31510662

AN - SCOPUS:85068935286

VL - 35

SP - i530-i537

JO - Bioinformatics

JF - Bioinformatics

SN - 1367-4803

IS - 14

M1 - btz353

ER -