Extracting Weak Signal with the Help of Data Clustering: Application to Motoneuron Adhesion
Résumé
It is known that the external forces exerted on cells can have a strong influence through the mechano-transduction processes. They may drive the cell differentiation into osteocytes or neuron, for example. In this work, the cell adhesion of motoneuron cells have been investigated by picosecond acoustic experiment coupled to microscope objectives. Acoustic echoes bouncing on a metal/cell interface have been imaged at the micron scale. The echo?s temporal shape can reveal the spatial and dynamical properties of the cell adhesion. Due to the data acquisition time, a trade-off has to be found between signal to noise ratio and spatio-temporal resolution. In order to improve the data analysis, the signals recorded at different location are clustered according to their temporal variation which may be partially masked by the noise. It is based on a spectral clustering method where the similarity matrix is given by the cross-correlation of the time traces. Actin layers, thinner than the acoustic wavelength, have been located and their viscosity have been extracted. Complementary AFM images show good coorelation with the acoustic images. Since the cell viscosity may affect cell adhesion measurement as performed in picosecond acoustics experiments, the problem arising from the standard massless Hook spring used to model the adhesion layer will be discussed.