Simplified, automated methods for assessing pixel intensities of fluorescently-tagged drugs in cells

Allan Kachelmeier, Tsering Shola, William B. Meier, Anastasiya Johnson, Meiyan Jiang, Peter Steyger

Research output: Contribution to journalArticle

Abstract

Assessing the cytoplasmic uptake of fluorescently-tagged drugs in heterogeneous cell types currently involves time-consuming manual segmentation of confocal microscopy images. We developed a set of methods that incorporate map algebra techniques to facilitate and expedite image segmentation, particularly of the parenchyma of intermediate cells in the stria vascularis of the inner ear. Map algebra is used to apply a convolution kernel to pixel neighborhoods to create a masking image to select pixels in the original image for further operations. Here, we describe the utility of integrated intensity-based, percentile-based, and local autocorrelation-based methods to automate segmentation of images into putative morphological regions for pixel intensity analysis. Integrated intensity-based methods are variants of watershed segmentation tools that determine morphological boundaries from rates of change in integrated pixel intensity. Percentile-and local autocorrelation-based methods evolved out of the process of developing map algebra-and integrated intensitybased tools. We identified several simplifications that are surprisingly effective for image segmentation and pixel intensity analysis. These methods were empirically validated on three levels: first, the algorithms were developed based on iterations of inspected results; second, algorithms were tested for various types of robustness; and third, developed algorithms were validated against results from manually-segmented images. We conclude the key to automated segmentation is supervision of output data.

Original languageEnglish (US)
Article numbere0206628
JournalPLoS One
Volume13
Issue number11
DOIs
StatePublished - Nov 1 2018

Fingerprint

Pixels
drugs
Algebra
Pharmaceutical Preparations
Image segmentation
Autocorrelation
cells
autocorrelation
Stria Vascularis
methodology
Confocal microscopy
Inner Ear
Watersheds
Convolution
Confocal Microscopy
ears
seeds
algebra

ASJC Scopus subject areas

  • Biochemistry, Genetics and Molecular Biology(all)
  • Agricultural and Biological Sciences(all)

Cite this

Simplified, automated methods for assessing pixel intensities of fluorescently-tagged drugs in cells. / Kachelmeier, Allan; Shola, Tsering; Meier, William B.; Johnson, Anastasiya; Jiang, Meiyan; Steyger, Peter.

In: PLoS One, Vol. 13, No. 11, e0206628, 01.11.2018.

Research output: Contribution to journalArticle

Kachelmeier, Allan ; Shola, Tsering ; Meier, William B. ; Johnson, Anastasiya ; Jiang, Meiyan ; Steyger, Peter. / Simplified, automated methods for assessing pixel intensities of fluorescently-tagged drugs in cells. In: PLoS One. 2018 ; Vol. 13, No. 11.
@article{9ad4a36cc37a4bd8b398d5886a13eeb9,
title = "Simplified, automated methods for assessing pixel intensities of fluorescently-tagged drugs in cells",
abstract = "Assessing the cytoplasmic uptake of fluorescently-tagged drugs in heterogeneous cell types currently involves time-consuming manual segmentation of confocal microscopy images. We developed a set of methods that incorporate map algebra techniques to facilitate and expedite image segmentation, particularly of the parenchyma of intermediate cells in the stria vascularis of the inner ear. Map algebra is used to apply a convolution kernel to pixel neighborhoods to create a masking image to select pixels in the original image for further operations. Here, we describe the utility of integrated intensity-based, percentile-based, and local autocorrelation-based methods to automate segmentation of images into putative morphological regions for pixel intensity analysis. Integrated intensity-based methods are variants of watershed segmentation tools that determine morphological boundaries from rates of change in integrated pixel intensity. Percentile-and local autocorrelation-based methods evolved out of the process of developing map algebra-and integrated intensitybased tools. We identified several simplifications that are surprisingly effective for image segmentation and pixel intensity analysis. These methods were empirically validated on three levels: first, the algorithms were developed based on iterations of inspected results; second, algorithms were tested for various types of robustness; and third, developed algorithms were validated against results from manually-segmented images. We conclude the key to automated segmentation is supervision of output data.",
author = "Allan Kachelmeier and Tsering Shola and Meier, {William B.} and Anastasiya Johnson and Meiyan Jiang and Peter Steyger",
year = "2018",
month = "11",
day = "1",
doi = "10.1371/journal.pone.0206628",
language = "English (US)",
volume = "13",
journal = "PLoS One",
issn = "1932-6203",
publisher = "Public Library of Science",
number = "11",

}

TY - JOUR

T1 - Simplified, automated methods for assessing pixel intensities of fluorescently-tagged drugs in cells

AU - Kachelmeier, Allan

AU - Shola, Tsering

AU - Meier, William B.

AU - Johnson, Anastasiya

AU - Jiang, Meiyan

AU - Steyger, Peter

PY - 2018/11/1

Y1 - 2018/11/1

N2 - Assessing the cytoplasmic uptake of fluorescently-tagged drugs in heterogeneous cell types currently involves time-consuming manual segmentation of confocal microscopy images. We developed a set of methods that incorporate map algebra techniques to facilitate and expedite image segmentation, particularly of the parenchyma of intermediate cells in the stria vascularis of the inner ear. Map algebra is used to apply a convolution kernel to pixel neighborhoods to create a masking image to select pixels in the original image for further operations. Here, we describe the utility of integrated intensity-based, percentile-based, and local autocorrelation-based methods to automate segmentation of images into putative morphological regions for pixel intensity analysis. Integrated intensity-based methods are variants of watershed segmentation tools that determine morphological boundaries from rates of change in integrated pixel intensity. Percentile-and local autocorrelation-based methods evolved out of the process of developing map algebra-and integrated intensitybased tools. We identified several simplifications that are surprisingly effective for image segmentation and pixel intensity analysis. These methods were empirically validated on three levels: first, the algorithms were developed based on iterations of inspected results; second, algorithms were tested for various types of robustness; and third, developed algorithms were validated against results from manually-segmented images. We conclude the key to automated segmentation is supervision of output data.

AB - Assessing the cytoplasmic uptake of fluorescently-tagged drugs in heterogeneous cell types currently involves time-consuming manual segmentation of confocal microscopy images. We developed a set of methods that incorporate map algebra techniques to facilitate and expedite image segmentation, particularly of the parenchyma of intermediate cells in the stria vascularis of the inner ear. Map algebra is used to apply a convolution kernel to pixel neighborhoods to create a masking image to select pixels in the original image for further operations. Here, we describe the utility of integrated intensity-based, percentile-based, and local autocorrelation-based methods to automate segmentation of images into putative morphological regions for pixel intensity analysis. Integrated intensity-based methods are variants of watershed segmentation tools that determine morphological boundaries from rates of change in integrated pixel intensity. Percentile-and local autocorrelation-based methods evolved out of the process of developing map algebra-and integrated intensitybased tools. We identified several simplifications that are surprisingly effective for image segmentation and pixel intensity analysis. These methods were empirically validated on three levels: first, the algorithms were developed based on iterations of inspected results; second, algorithms were tested for various types of robustness; and third, developed algorithms were validated against results from manually-segmented images. We conclude the key to automated segmentation is supervision of output data.

UR - http://www.scopus.com/inward/record.url?scp=85055905950&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85055905950&partnerID=8YFLogxK

U2 - 10.1371/journal.pone.0206628

DO - 10.1371/journal.pone.0206628

M3 - Article

C2 - 30383813

AN - SCOPUS:85055905950

VL - 13

JO - PLoS One

JF - PLoS One

SN - 1932-6203

IS - 11

M1 - e0206628

ER -