Comparative Study of Computational Methods for Classifying Red Blood Cell Elasticity
<p>Summary of the framework. Blue and magenta arrows represent actions performed with training and evaluation data, respectively. Blue boxes represent input/output; orange boxes represent actions performed with the data.</p> "> Figure 2
<p>On the left, microfluidic channel topology is shown. Only the basic part with five obstacles (depicted with blue colour) was simulated. The figure on the right shows the scheme of the simulation box with the dimensions of the individual parts.</p> "> Figure 3
<p>Time series plot of surface-area-to-volume (<math display="inline"><semantics> <mrow> <mi>S</mi> <mi>A</mi> <mo>:</mo> <mi>V</mi> </mrow> </semantics></math>) ratio for a single healthy RBC.</p> "> Figure 4
<p>Minimum, maximum, and average <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>A</mi> <mo>:</mo> <mi>V</mi> </mrow> </semantics></math> ratio for nine healthy RBCs. Cells are sorted by average.</p> "> Figure 5
<p>Average, minimum, maximum, and variance of <math display="inline"><semantics> <mrow> <mi>S</mi> <mi>A</mi> <mo>:</mo> <mi>V</mi> </mrow> </semantics></math> ratio for all 4 cell types. Cells of each type are sorted by the observed characteristic for each plot.</p> "> Figure 6
<p>Variance of surface area and volume for cell types 0 and 3.</p> "> Figure 7
<p>Dependence of classification results on <span class="html-italic">S</span> when predicting 4 classes.</p> "> Figure 8
<p>Dependence of classification results on <span class="html-italic">S</span> when predicting 2 classes.</p> "> Figure 9
<p>Dependence of classification results on predictor set when predicting 4 classes.</p> "> Figure 10
<p>Dependence of classification results on predictor set when predicting 2 classes.</p> "> Figure 11
<p>Importance of predictors from the 6th set.</p> "> Figure 12
<p>Importance of predictors from the 4th set.</p> ">
Abstract
:1. Introduction
2. Materials and Methods
2.1. Underlying Computational Model
2.2. Channel Geometry and Flow Setting
2.3. Cell Settings
3. Summary of Neural Network Results
4. Statistical Analysis of Elasticity from the Perspective of Red Blood Cell Surface Optimization
4.1. Data and Analysis Tools
4.2. Method of Analysis
4.3. Analysis Results
5. Classification Using Decision Trees
5.1. Used Predictors
- How difficult it may be to extract the predictor from real experiments;
- Why one should expect that the elasticity of RBCs affects the given predictor.
- Set 1: It consists of only two predictors—the dimensions of the rectangle in which the monitored cell is located, i.e., the dimensions of the cell in the direction of the x-axis and in the direction of the y-axis. These values can be easily obtained from a (static) snapshot of the cell. We may expect that the bounding box of an elastic RBC will vary in size more than the bounding box of a stiff RBC;
- Set 2: The additional predictors are velocities and changes in the velocity of the cell in the x- and y-directions. These data can be calculated from several consecutive images. The elasticity of an RBC affects its shape, which may affect the velocity of the cell in the flow;
- Set 3: It also includes dimensions and velocities in the z-axis direction. These values can be obtained from images taken from a different angle;
- Set 4: We added the cell axis length and the maximum and minimum diameter of the cell equator, which can potentially be determined from multi-angle images. While it is more difficult to obtain these characteristics from real experiments, one may expect that the shape of a cell is better represented by these values than just by the bounding box dimensions alone;
- Set 5: It additionally contains predictors that can be calculated from the complete cell triangulation—cell surface, cell volume and means, standard deviations and skewness coefficients calculated from the lengths of all triangulation edges, angles formed by every two triangulation triangles, and solid angles at all triangulation nodes. Cell triangulation is quite difficult to obtain from a real experiment; it would require the creation of a 3D image of the cell from the scanned flow. However, by using these complex characteristics, we may be able to distinguish more subtle changes in the shape of the RBC than with just the basic dimensions;
- Set 6: It also contains the means, standard deviations, and coefficients of the skewness of deviations and the absolute deviations of the characteristics added in the fifth set. The deviations are calculated from the cell in a relaxed state, so, for the calculation, we need to have the basic triangulation of the cell, which, in our procedure, corresponds to the cell in the first step of the simulation. One may expect that quantifying the changes of the shape represents the elasticity better than the original quantities themselves.
5.2. Used Machine Learning Tools
5.3. Data Preparation
5.4. Classification Results
6. Discussion
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
Abbreviations
CNN | convolutional neural network |
IBM | immersed boundary method |
KS test | Kolmogorov–Smirnov test |
LBM | lattice Boltzmann method |
RBC | red blood cell |
RF | random forest classifier from the sklearn library |
XGB | gradient decision trees classifier from the xgboost library |
Appendix A. List of Predictors
Set | Predictor | Description |
---|---|---|
1 | span x | size of the bounding box of the cell in the direction of x-axis |
span y | size of the bounding box of the cell in the direction of y-axis | |
2 | rbc velocity x | velocity of the cell in the direction of x-axis |
rbc velocity y | velocity of the cell in the direction of y-axis | |
vel diff x | change of the velocity of the cell in the direction of x-axis | |
vel diff y | change of the velocity of the cell in the direction of y-axis | |
3 | rbc velocity z | velocity of the cell in the direction of z-axis |
span z | size of the bounding box of the cell in the direction of z-axis | |
vel diff z | change of the velocity of the cell in the direction of z-axis | |
4 | cell axis length | distance between the two nodes of the cell which form the central axis of the cell |
equator diameter | maximum distance between two opposite nodes on the perimeter of the cell | |
equator diameter min | minimum distance between two opposite nodes on the perimeter of the cell | |
5 | volume | cell volume |
surface | cell surface | |
edge angle mean | mean of the angles between neighbouring triangles of the cell triangulation | |
edge angle deviation | standard deviation of the angles between neighbouring triangles of the cell triangulation | |
edge angle skewness | skewness coefficient of the angles between neighbouring triangles of the cell triangulation | |
node angle mean | mean of the solid angles at the nodes of the cell triangulation | |
node angle deviation | standard deviation of the solid angles at the nodes of the cell triangulation | |
node angle skewness | skewness coefficient of the solid angles at the nodes of the cell triangulation | |
edge length mean | mean of the lengths of the edges of the cell triangulation | |
edge length deviation | standard deviation of the lengths of the edges of the cell triangulation | |
edge length skewness | skewness coefficient of the lengths of the edges of the cell triangulation | |
6 | edge angle delta mean | as “edge angle mean”, calculated based on the deviations from the relaxed state |
edge angle delta deviation | as “edge angle deviation”, calculated based on the deviations from the relaxed state | |
edge angle delta skewness | as “edge angle skewness”, calculated based on the deviations from the relaxed state | |
node angle delta mean | as `node angle mean”, calculated based on the deviations from the relaxed state | |
node angle delta deviation | as “node angle deviation”, calculated based on the deviations from the relaxed state | |
node angle delta skewness | as "node angle skewness", calculated based on the deviations from the relaxed state | |
edge length delta mean | as “edge length mean”, calculated based on the deviations from the relaxed state | |
edge length delta deviation | as “edge length deviation”, calculated based on the deviations from the relaxed state | |
edge length delta skewness | as “edge length skewness”, calculated based on the deviations from the relaxed state | |
edge angle delta abs mean | as “edge angle delta mean”, based on the absolute deviations | |
edge angle delta abs deviation | as “edge angle delta deviation”, based on the absolute deviations | |
edge angle delta abs skewness | as “edge angle delta skewness”, based on the absolute deviations | |
node angle delta abs mean | as “node angle delta mean”, based on the absolute deviations | |
node angle delta abs deviation | as “node angle delta deviation”, based on the absolute deviations | |
node angle delta abs skewness | as “node angle delta skewness”, based on the absolute deviations | |
edge length delta abs mean | as “edge length delta mean”, based on the absolute deviations | |
edge length delta abs deviation | as “edge length delta deviation”, based on the absolute deviations | |
edge length delta abs skewness | as “edge length delta skewness”, based on the absolute deviations |
References
- Klei, T.R.L.; Dalimot, J.; Nota, B.; Veldthuis, M.; Mul, F.P.J.; Rademakers, T.; Hoogenboezem, M.; Nagelkerke, S.Q.; van Ijcken, W.F.J.; Oole, E.; et al. Hemolysis in the spleen drives erythrocyte turnover. Blood 2020, 136, 1579–1589. [Google Scholar] [CrossRef] [PubMed]
- Duez, J.; Holleran, J.P.; Ndour, P.A.; Pionneau, C.; Diakité, S.; Roussel, C.; Dussiot, M.; Amireault, P.; Avery, V.M.; Buffet, P.A. Mechanical clearance of red blood cells by the human spleen: Potential therapeutic applications of a biomimetic RBC filtration method. Transfus. Clin. Biol. 2015, 22, 151–157. [Google Scholar] [CrossRef] [PubMed]
- Tyas, D.A.; Hartati, S.; Harjoko, A.; Ratnaningsih, T. Morphological, texture, and color feature analysis for erythrocyte classification in thalassemia cases. IEEE Access 2020, 8, 69849–69860. [Google Scholar] [CrossRef]
- Alapan, Y.; Matsuyama, Y.; Little, J.A.; Gurkan, U.A. Dynamic deformability of sickle red blood cells in microphysiological flow. Technology 2016, 4, 71–79. [Google Scholar] [CrossRef]
- Depond, M.; Henry, B.; Buffet, P.; Ndour, P.A. Methods to investigate the deformability of RBC during malaria. Front. Physiol. 2020, 10, 1613. [Google Scholar] [CrossRef]
- Islamzada, E.; Matthews, K.; Guo, Q.; Santoso, A.T.; Duffy, S.P.; Scott, M.D.; Ma, H. Deformability based sorting of stored red blood cells reveals donor-dependent aging curves. Lab Chip 2020, 20, 226–235. [Google Scholar] [CrossRef]
- Namvar, A.; Blanch, A.J.; Dixon, M.W.; Carmo, O.M.S.; Liu, B.; Tiash, S.; Looker, O.; Andrew, D.; Chan, L.; Tham, W.; et al. Surface area-to-volume Ratio, Not Cellular Viscoelasticity, Is the Major Determinant of Red Blood Cell Traversal through Small Channels. Cell. Microbiol. 2021, 23, e13270. [Google Scholar] [CrossRef]
- Renoux, C.; Faivre, M.; Bessaa, A.; Da Costa, L.; Joly, P.; Gauthier, A.; Connes, P. Impact of Surface-area-to-volume Ratio, Internal Viscosity and Membrane Viscoelasticity on Red Blood Cell Deformability Measured in Isotonic Condition. Sci. Rep. 2019, 9, 6771. [Google Scholar] [CrossRef] [PubMed]
- Park, H.; Lee, S.; Ji, M.; Kim, K.; Son, Y.; Jang, S.; Park, Y. Measuring cell surface area and deformability of individual human red blood cells over blood storage using quantitative phase imaging. Sci. Rep. 2016, 6, 34257. [Google Scholar] [CrossRef]
- Diez-Silva, M.; Dao, M.; Han, J.; Lim, C.T.; Suresh, S. Shape and biomechanical characteristics of human red blood cells in health and disease. MRS Bull. 2010, 35, 382–388. [Google Scholar] [CrossRef]
- Brun, J.F.; Varlet-Marie, E.; Myzia, J.; de Mauverger, E.R.; Pretorius, E. Metabolic influences modulating erythrocyte deformability and eryptosis. Metabolites 2022, 12, 4. [Google Scholar] [CrossRef] [PubMed]
- Lamoureux, E.S.; Cheng, Y.; Islamzada, E.; Matthews, K.; Duffy, S.P.; Ma, H. Biophysical Profiling of Red Blood Cells from Thin-film Blood Smears using Deep Learning. Heliyon 2024, 10, e35276. [Google Scholar] [CrossRef]
- Molina, A.; Rodellar, J.; Boldú, L.; Acevedo, A.; Alférez, S.; Merino, A. Automatic identification of malaria and other red blood cell inclusions using convolutional neural networks. Comput. Biol. Med. 2021, 136, 104680. [Google Scholar] [CrossRef]
- Lopes, M.G.; Recktenwald, S.M.; Simionato, G.; Eichler, H.; Wagner, C.; Quint, S.; Kaestner, L. Big data in transfusion medicine and artificial intelligence analysis for red blood cell quality control. Transfus. Med. Hemother. 2023, 50, 163–173. [Google Scholar] [CrossRef]
- Aliyu, H.A.; Sudirman, R.; Razak, M.A.A.; Abd Wahab, M.A. Red blood cell classification: Deep learning architecture versus support vector machine. In Proceedings of the 2nd International Conference on Biosignal Analysis, Processing and Systems (ICBAPS), Kuching, Malaysia, 24–26 July 2018; pp. 142–147. [Google Scholar] [CrossRef]
- Singh, V.; Srivastava, V.; Mehta, D.S. Machine learning-based screening of red blood cells using quantitative phase imaging with micro-spectrocolorimetry. Opt. Laser Technol. 2020, 124, 105980. [Google Scholar] [CrossRef]
- Molčan, S.; Smiešková, M.; Bachratý, H.; Bachratá, K.; Novotný, P. Classification of red blood cells using time-distributed convolutional neural networks from simulated videos. Appl. Sci. 2023, 13, 7967. [Google Scholar] [CrossRef]
- Jančigová, I.; Kovalčíková, K.; Weeber, R.; Cimrák, I. PyOIF: Computational tool for modelling of multi-cell flows in complex geometries. PLoS Comput. Biol. 2020, 16, e1008249. [Google Scholar] [CrossRef] [PubMed]
- Weik, F.; Weeber, R.; Szuttor, K.; Breitsprecher, K.; de Graaf, J.; Kuron, M.; Landsgesell, J.; Menke, H.; Sean, D.; Holm, C. ESPResSo 4.0—An extensible software package for simulating soft matter systems. Eur. Phys. J. Spec. Top. 2019, 227, 1789–1816. [Google Scholar] [CrossRef]
- Krüger, T.; Kusumaatmaja, H.; Kuzmin, A.; Shardt, O.; Silva, G.; Viggen, E.M. The Lattice Boltzmann Method; Springer International Publishing: Cham, Switzerland, 2017. [Google Scholar] [CrossRef]
- Bušík, M.; Slavík, M.; Cimrák, I. Dissipative coupling of fluid and immersed objects for modelling of cells in flow. Comput. Math. Methods Med. 2018, 2018, 7842857. [Google Scholar] [CrossRef]
- Mills, J.P.; Qie, L.; Dao, M.; Lim, C.T.; Suresh, S. Nonlinear elastic and viscoelastic deformation of the human red blood cell with optical tweezers. MCB Mol. Cell. Biomech. 2004, 1, 169. [Google Scholar] [CrossRef]
- Jančigová, I.; Kovalčíková, K.; Bohiniková, A.; Cimrák, I. Spring-network model of red blood cell: From membrane mechanics to validation. Int. J. Numer. Methods Fluids 2020, 92, 1368–1393. [Google Scholar] [CrossRef]
- Suresh, S.; Spatz, J.; Mills, J.P.; Micoulet, A.; Dao, M.; Lim, C.T.; Beil, M.; Seufferlein, T. Connections between single-cell biomechanics and human disease states: Gastrointestinal cancer and malaria. Acta Biomater. 2005, 1, 15–30. [Google Scholar] [CrossRef] [PubMed]
- Cimrák, I.; Bachratá, K.; Bachratý, H.; Jančigová, I.; Tóthová, R.; Bušík, M.; Slavík, M.; Gusenbauer, M. Object-in-fluid framework in modeling of blood flow in microfluidic channels. Commun.-Sci. Lett. Univ. Zilina 2016, 18, 13–20. [Google Scholar] [CrossRef]
- Brunello, A.; Marzano, E.; Montanari, A.; Sciavicco, G. J48SS: A novel decision tree approach for the handling of sequential and time series data. Computers 2019, 8, 21. [Google Scholar] [CrossRef]
- Mutlag, W.K.; Ali, S.K.; Aydam, Z.M.; Taher, B.H. Feature extraction methods: A review. J. Phys. Conf. Ser. 2020, 1591, 012028. [Google Scholar] [CrossRef]
- Rady, E.H.A.; Fawzy, H.; Fattah, A.M.A. Time Series Forecasting Using Tree Based Methods. J. Stat. Appl. Probab. 2021, 10, 229–244. [Google Scholar] [CrossRef]
- Veluchamy, M.; Perumal, K.; Ponuchamy, T. Feature extraction and classification of blood cells using artificial neural network. Am. J. Appl. Sci. 2012, 9, 615. [Google Scholar] [CrossRef]
- Shwartz-Ziv, R.; Armon, A. Tabular data: Deep learning is not all you need. Inf. Fusion 2022, 81, 84–90. [Google Scholar] [CrossRef]
- Uddin, S.; Lu, H. Confirming the statistically significant superiority of tree-based machine learning algorithms over their counterparts for tabular data. PLoS ONE 2024, 19, e0301541. [Google Scholar] [CrossRef]
- Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef]
Parameter | Value |
---|---|
stretching coefficient () | 5 × 10−6 N/m |
bending coefficient () | 3 × 10−19 m |
coefficient of local area conservation () | 2 × 10−5 N/m |
coefficient of global area conservation () | 7 × 10−4 N/m |
coefficient of volume conservation () | 900 N/m2 |
Training Number of Classes | Validation and Testing Number of Classes | Accuracy |
---|---|---|
4 | 4 | 55.48% |
2 | 2 | 61.72% |
4 | 2 | 93.91% |
p-Value | Type 0 | Type 1 | Type 2 | Type 3 | |
---|---|---|---|---|---|
D-Value | |||||
type 0 | 0 | 0.0007 | 0.0007 | ||
type 1 | 1 | 0.3517 | 0.9895 | ||
type 2 | 0.8889 | 0.4444 | 0.7301 | ||
type 3 | 0.8889 | 0.2222 | 0.3333 |
p-Value | Type 0 | Type 1 | Type 2 | Type 3 | |
---|---|---|---|---|---|
D-Value | |||||
type 0 | 0.3517 | 0.7301 | 0.3517 | ||
type 1 | 0.4444 | 0.7301 | 0.9895 | ||
type 2 | 0.3333 | 0.3333 | 0.7301 | ||
type 3 | 0.4444 | 0.2222 | 0.3333 |
p-Value | Type 0 | Type 1 | Type 2 | Type 3 | |
---|---|---|---|---|---|
D-Value | |||||
type 0 | 0.0336 | 0 | 0.0063 | ||
type 1 | 0.6667 | 0.1256 | 0.1259 | ||
type 2 | 1 | 0.5556 | 0.9895 | ||
type 3 | 0.7778 | 0.5556 | 0.2222 |
p-Value | Type 0 | Type 1 | Type 2 | Type 3 | |
---|---|---|---|---|---|
D-Value | |||||
type 0 | 0.7301 | 0.9895 | 0.9895 | ||
type 1 | 0.3333 | 0.7301 | 0.9895 | ||
type 2 | 0.2222 | 0.3333 | 0.9895 | ||
type 3 | 0.2222 | 0.2222 | 0.2222 |
Training Number of Classes | Testing Number of Classes | Model | Predictors | Accuracy |
---|---|---|---|---|
4 | 4 | CNN | video recordings | 55.48% |
XGB | set 1 | 45.98% | ||
RF | set 1 | 45.27% | ||
XGB | set 2 | 60.81% | ||
RF | set 2 | 60.39% | ||
XGB | set 3 | 63.64% | ||
RF | set 3 | 62.85% | ||
XGB | set 4 | 69.16% | ||
RF | set 4 | 67.51% | ||
XGB | set 5 | 91.14% | ||
RF | set 5 | 86.73% | ||
XGB | set 6 | 93.90% | ||
RF | set 6 | 89.96% | ||
2 | 2 | CNN | video recordings | 61.72% |
XGB | set 1 | 82.81% | ||
RF | set 1 | 82.35% | ||
XGB | set 2 | 92.00% | ||
RF | set 2 | 91.39% | ||
XGB | set 3 | 93.32% | ||
RF | set 3 | 93.27% | ||
XGB | set 4 | 96.50% | ||
RF | set 4 | 95.55% | ||
XGB | set 5 | 98.04% | ||
RF | set 5 | 98.32% | ||
XGB | set 6 | 98.97% | ||
RF | set 6 | 99.08% | ||
4 | 2 | CNN | video recordings | 93.91% |
XGB | set 1 | 81.57% | ||
RF | set 1 | 80.89% | ||
XGB | set 2 | 91.55% | ||
RF | set 2 | 90.60% | ||
XGB | set 3 | 93.10% | ||
RF | set 3 | 93.00% | ||
XGB | set 4 | 96.03% | ||
RF | set 4 | 95.23% | ||
XGB | set 5 | 97.92% | ||
RF | set 5 | 97.93% | ||
XGB | set 6 | 98.92% | ||
RF | set 6 | 98.84% |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Bachratý, H.; Novotný, P.; Smiešková, M.; Bachratá, K.; Molčan, S. Comparative Study of Computational Methods for Classifying Red Blood Cell Elasticity. Appl. Sci. 2024, 14, 9315. https://doi.org/10.3390/app14209315
Bachratý H, Novotný P, Smiešková M, Bachratá K, Molčan S. Comparative Study of Computational Methods for Classifying Red Blood Cell Elasticity. Applied Sciences. 2024; 14(20):9315. https://doi.org/10.3390/app14209315
Chicago/Turabian StyleBachratý, Hynek, Peter Novotný, Monika Smiešková, Katarína Bachratá, and Samuel Molčan. 2024. "Comparative Study of Computational Methods for Classifying Red Blood Cell Elasticity" Applied Sciences 14, no. 20: 9315. https://doi.org/10.3390/app14209315