Automatic Detection and Classification of Ca2+ Release Events in Confocal Line- and Framescan Images


abstract text


Numerous methods exist for analysing sparks in confocal linescans. These employ various different approaches: noise tresholding (Ríos 2001, Picht 2007), wavelet transform (Szabó 2010), etc.

Recently, a method was developed by Tian et al., (Tian 2012), where fluorescence time trace in each pixel is fitted and provides a practically noise free approximation of the original fluorescence data. This pixel-by-pixel method had, however, several limitations which made it impractical to be used for Ca2+ release event detection.

Here we extend the method by Tian et al., in several ways. The new method allows for Ca2+ release event classification based on pixel-by-pixel denoising of the original signal.


Cell isolations


Data acquisition was performed on two confocal setups. Linescan images were obtained on a Olympus FluoView 1000 confocal microscope. Framescan images were recorded with a VTInfinity multi-beam confocal microscope recording 512x64 pixel images at 150Hz freqency.


The detection algorithm is presented in the Results section.


Pixel by pixel event classification

The algorithm is presented schematically on figure 1. Each subroutine is explained in detail below.

Image preprocessing

The only preprocessing step used is convolving the image with a \((2n+1)\times(2n+1)\) kernel where the center element is \(1/(n+1)\) and the k-th layer surrounding the center is made up of values \(1/(8k\cdot(n+1))\). For example, when n=1 the kernel would be [ \left( \begin{array}{ccc} ^1/{16} & ^1/{16} & ^1/{16}\\ ^1/{16} & ^1/{2}& ^1/{16}\\ ^1/{16} & ^1/{16} & ^1/_{16} \end{array} \right) ] Convolving the image with this kind of kernel reduces the noise while retaining more of the original signal than simple averaging. Contributions from the i-th layer around the center will have the same weight as the central pixel. In this work we use the kernel with \(n=1\).

Region detection

Before it is possible to fit the fluorescence signal from each pixel with a transient function, candidate regions containing possible events must be detected. For this we have modified a continous wavelet transform based peak detection algorithm by Du et al.(Du 2006). Whereas the the original algorithm of Du et al. provides the location of the peak, we have extended it to also yield the the width of the peak.

The original algorith works by joining the wavelet transform values along local maxima for increasing window lengths. Ridge lines obatained in this manner are taken to correspond to peaks if they satisfy certain criteria (length of the ridge, SNR, etc.) In our extension, the width of the peak is obtained from finding the first maximum of the wavelet transform values along the ridge line.

Region estimation is performed iteratively. This is done to ensure that overlapping events are correctly identified. At each pass regions which have no overlaps and regions which only overlap with regions having a lower peak SNR are fitted. After this the fit result is subtracted from the original signal and region detection is performed again, followed by fitting. This is done until no more regions are detected or successfully fitted.

Signal fitting

The function used for fitting Ca2+ release events is shown on Figure \ref{fig:f2}. The shape of the function is described by 4 parameters: amplitude (\(A\)), rise