Structure of this manuscript
• Chapter 2 and 3 present the context of our studies.
• Chapter 4 summarizes the contributions of Chapters 5, 6 and 7. Because this manuscrit is intended to be read also by medical practitioners and biologists, chapter 4 describes the essential elements presented in these chapters.
• Chapter 5 is a presentation of the image processing, image analysis and computer vision tools and methods we use in the manuscript.
• Chapter 6 details the procedures developed for the analysis of simple motion. Chapter 7 proposes a methodology for the analysis of more complex motions.
• Chapter 8 presents the pipelines of our cilia analysis procedures. The work presented in this chapter was published at ISBI 2015  and selected for oral presentation at ICIP 2016.
• Chapter 9 is a feasability study for assessing cilia motility in vivo. It was published at the medical conference ERS 2016 .
• Chapter 10 refers to our fish embryo mortality assay, and was published at ISMM 2015  and in the journal CBM 
• Chapter 11 presents the pipeline of our heart rate estimation on fish embryo’s heart and tail. It was presented at an oral session at IPTA 2016 .
• Chapter 12 concludes and proposes avenues for further work resulting from this thesis.
As outlined in the next section, all the applicative chapters have been published in some fashion.
Publications associated with this manuscript
International Journal Papers
 E. Puybareau, D. Genest, E.Barbeau, M. Léonard, and H. Talbot. An automated assay for the evaluation of mortality in fish embryo. In print in Computers in Bio-Medicine International Conferences  E. Puybareau, H. Talbot, G. Pelle, B. Louis, J-F. Papon, A. Coste, and L. Najman. Automating the measurement of physiological parameters: a case study in the image analysis of cilia motion. In Image Processing (ICIP), IEEE International Conference on, Phoenix, September 2016  E. Puybareau, H. Talbot, and M. Leonard. Automated heart rate estimation in fish embryo. In Image Processing Theory, Tools and Applications (IPTA), International Conference on, pages 379–384, Orleans, November 2015  E. Puybareau, M. Léonard, and H. Talbot. An automated assay for the evaluation of mortality in fish embryo. In Mathematical Morphology and Its Applications to Signal and Image Processing, volume 9082 of Lecture Notes in Computer Science, pages 110–121. Springer, Reykjavik, May 2015  E. Puybareau, H. Talbot, G. Pelle, B. Louis, J-F. Papon, A. Coste, and L. Najman. A regionalized automated measurement of ciliary beating frequency. In Biomedical Imaging (ISBI), IEEE 12th International Symposium on, pages 528–531, New-York, April 2015 Medical Conferences Abstracts  E. Puybareau, E. Bequignon, M. Bottier, G. Pelle, B. Louis, E. Escudier, J.-F. Papon, L. Najman, H. Talbot, and A. Coste. Frequency-based region identification and delimitation for cilia beating pattern analysis. In Cilia 2016, Amsterdam,
Estimating cilia beating frequencies
The estimation of ciliary beating frequency has been a research topic since the middle of the 20th century. One of the first methods of reference for the measurement of cilary beating frequency was proposed in 1962 and used a photo-sensitive cell . Stroboscopic methods have been replaced by more accurate techniques that use photomultiplier, photodiode and high-speed imaging. Those methods are described and compared in .
Some attempts to automate the measurement of CBF have been proposed in the literature. The SAVA System  estimates frequencies from small 4×4 pixels windows. Whole frequency spectra can be simultaneously estimated. This method is based on grey-level intensity variation, which has shown some limitation if the contrast is not sufficient, rendering the reliability of the technique questionable . CiliaFA  provides a frequency histogram of a large number of small regions of interest, assuming low noise and no cell proper motion. The method proposed in  uses a sparse optical flow to estimate a single frequency per image. Thus, it is not applicable when several different beating patterns are present in the sequence. Moreover, the method is very sensitive to noise and is easily perturbed by cells proper motion. A linescan-based technique is proposed in , coupled with the Fast Fourier Transform  (first developed in 1866 ), and is evaluated on slices on brain ciliated epithelium. It deals with acquisition problems: the removal of artefacts due to the camera sensor, and frame stabilization. However, the removal needs a blank acquisition sequence and thus access to the camera. More problematic for our application, the straight linescan technique needs a straight border of cells, something not always possible with harvested cells. In the field of view, multiple cell groups are often visible, and cilia on a given cell can beat at different frequencies. As a result, many frequencies can be measured in a single field of view. Such frequencies provide information on cilia synchronization, and ultimately on the status of the cells under scrutiny. We seek to segment the field of view into regions that are consistent from the point of view of the beating pattern. Nasal brushing produces significant amounts of cells with beating cilia. The diversity in sequence appearances can be appreciated on Fig. 2.7. Brushings were all recorded under a microscope minutes after the biopsy, at 358 frames per second with a high speed
camera. The spatial resolution is 0.13μm, and the resolution is 256×192 pixels. The sequence was recorded on the border of the groups.
Table of contents :
1.1 General context
1.3 Structure of this manuscript
1.4 Publications associated with this manuscript
2 Ciliated cells analysis
2.1 Ciliated cells
2.2 Context and state of the art
2.2.1 Estimating cilia beating frequencies
2.2.2 Cilia beating characterization and diagnosis
2.2.3 Estimating cilia behaviour in vivo
3 Fish embryos and eco-toxicity
3.2 The fish embryo model
3.3 Image processing and fish studies
II Technical contributions
4 Methodology essentials
4.1.1 Sensor pattern removal
4.1.2 Sequence stabilization
4.2 Simple motion analysis
4.2.1 Motion highlighting
4.2.2 Motion segmentation by temporal gradient
4.2.3 Motion segmentation by temporal variance
4.2.4 Spurious motion elimination
4.2.5 Frequency estimation
4.3 Complex motion analysis
4.3.1 Feature-based region segmentation
5 Tools for motion analysis
5.1.2 2D+t sequences
5.2 Basic tools
5.2.1 Mathematical morphology
5.2.2 Graph-based optimisation model
5.2.3 Gaussian filter
5.2.4 Bilateral filter
5.2.5 Optical flow
5.2.7 Fourier Transform
5.3.1 Definition of motion
5.3.2 Sensor pattern removal
5.3.3 Image stabilization
6 Simple motion analysis
6.1 Motion enhancement
6.1.2 Enhancement methodology
6.2 Motion segmentation by temporal gradient
6.3 Motion segmentation by temporal variance
6.4 False motion elimination
6.5 Frequency estimation
6.5.1 Semi-automatic grey-level intensity based frequency estimation
6.5.2 Automatic optical flow based frequency estimation
7 Complex motion identification
7.1 Feature-based region segmentation
7.1.1 Graph-based optimisation model
7.1.2 Descriptors and weights
7.2 Pattern extraction: Curvescan
7.2.2 Linescan definition
III Application: cilia motility evaluation
8 Cilia Beating Analysis
8.2 Details of the methodology: common parts
8.3 Methodology for frequency estimation
8.3.1 Methodology after the segmentation
8.3.2 Results and Validation.
8.4 Methodology for cilia beating characterization
8.4.1 Methodology after the segmentation
8.5.2 Comparison of the two methods
9 In vivo assessment of cilia motility evaluation
9.1 Existing tools and solutions proposed
9.2 The Cellvizio properties
9.3 Experimental runs
9.4 Results ans analysis
IV Application: fish embryo based assays
10 Fish embryo mortality evaluation
10.3 Details of the methodology
10.4 Results and validation
10.5 Pipeline improvements for enhanced automation.
10.6 Modifications : details
11 Heart frequency estimation
11.2 Details of the methodology
11.3 Results and validations
11.4 Further work
12.1 Contribution of this work
12.2 Future work