Skip to main content

Table 2 Literature review for automated species identification for phytoplankton species identification with machine learning (neural networks), with flow cytometric data or images

From: Combining high-throughput imaging flow cytometry and deep learning for efficient species and life-cycle stage identification of phytoplankton

 

Reference

Network

Species/classes

Species size

Parameters included

Type of parameter

Instrument

Accuracy

Flow cytometric data (scatter and fluorescence values)

Frankel et al. [19]

ANN (Kohonen network/Back-propagation neural networks)

5

Picoplankton, large phytoplankton

5

FSC

488 nm Ex./540–630 nm

514 nm Ex./540–630 nm

488 nm Ex./660–700 nm

514 nm Ex./660–700 nm

EPICS V

92–100%

Balfoort et al. [1]

Multilayer feedforward network (NWorks, ANNET, 8)

8

3–3500 µm

6

FSC, SSC, TOF

488 nm Ex./515–600 nm Em

488 nm Ex./650–750 nm Em

633 nm Ex./650–750 nm Em

Optical Plankton Analyzer

90–98%

Boddy et al. [6]

Back-propagation neural networks, hierarchical approach

40

3–40 µm

6

FSC (horizontal, vertical), SSC, TOF

488 nm Ex./> 660 nm

488 nm Ex./530–590 nm

EPICS 741

> 70%

Wilkins et al. [51]

MLP, RBF

42

Boddy et al. [6]

68–74%

Wilkins et al. [52]

RBF ANN

34

3–1000 µm

11

FSC, SSC, TOF

488 nm Ex./Red Em

488 nm Ex./Orange Em

488 nm Ex./Green Em

630 nm Ex./Red Em

Diffraction module: Vertical bar, horizontal bar, outer ring, inner ring

EurOPA

92%

Boddy et al. [5]

RBF NN

72

1–45 µm

7

FSC-H, SSC-H, FL1-H, FL2-H, FL3-H, FL3-A, TOF

FACSort™

70–77%

Pulse shape

Malkassian et al. [33]

 

20

n. a.

8

FSC, SSC

FLR (668–734 nm Em.)

FLO (601–668 nm Em.)

FLY (536–601 nm Em.)

Pulse shape descriptors (shape, length and area under the pulse)

CytoSub

78%

Images

Gorsky et al. [23]

 

3

3–43 µm

5

Area, Circularity, Convexity, Length, Perimeter

Autonomous Image Analyzer/HIAC

Embleton et al. [17]

MLP

4

10–390 µm

74

Area, Circularity, Diameter, Fibrelength, Grey level values (SD, Skewness, Kurtosis), Perimeter

Microscope camera (Sony DXC-930P)

67–93%

Sosik and Olson [45]

SVM

15 (natural samples)

10–100 µm

22 categories/210 elements

Size, shape, symmetry, Texture characteristics, Diffraction, Co-occurence

FlowCytobot

68–99%

Blaschko et al. [4]

SVM

13 classes

n. a.

780 features

Simple shape, moments, contour, differential, texture

FlowCam

71%

Correa et al. [9]

CNN

19 classes

n.a.

FlowCam

89%

Rodenacker et al. [42]

DT, LDA

23

n. a.

5

Shape, significant points, principal components, contour, fourier descriptor, extinction, Shape moments, colorimetry, fluorimetry

Inverse microscope

76%

Chen et al. [8]

CNN, PCA + SVM

1

n. a.

16

Diameter, tight area, perimeter, circularity, major axis, orientation, loose area, median radius, opd, refractive index, absorption, scattering

TS-QPI

< 85%

Li et al. [31]

CNN

9

n. a.

Mueller matrix microscope

97%

Pedraza et al. [37]

CNN

80

n. a.

Microscopy

99%

This study

CNN

9

1–90 µm

ImageStream®X MK II

97%

  1. The respective network, species or classes, species size, parameter number and parameter type, instruments or technique and the respective accuracy is provided. For some studies it was not possible to get details from the text about investigated cell size range
  2. TOF Time of flight, MLP Multilayer Perceptron/backpropagation network, RBF radial basis function, SVM support vector machine, TS-QPI Time stretch quantitative phase imaging, OPD optical path length difference, n. a. not available, Picoplankton 0.2–2 µm, Nanoplankton 2–20 µm, Microplankton 20–200 µm, Macroplankton 200–2000 µm