Advances in optic devices and microscopes enable us to visualize micro- and nano-scale range of objects. In particular, the upgradations in imaging algorithms and softwares allow to get more appropriate images or data from various experiments. Since the architecture of Langerhans islet is unique, studying pancreatic islets with microscopes must consider its three-dimensional (3D) structure1. This difference in islet structure needed an alternate quantifying method, rather than assessing islet viability with fluorescence dyes on 2D images. Recently, advanced methods have been studied for staining methods and fluorescence dyes2, 3. In addition, some experimental protocols have already been established to manually determine the scoring of islet viability.4, 5 Although several fluorescence-based staining protocols suggests their accuracy and immediacy, it is still ambiguous to quantify cell viability, especially in the 3D culture. Thus, approach to numerical assessment of islet viability with computer vision is timely required in order to properly describe the characteristics of objects.6 In this protocol, we focused on the architecture of pancreatic islet, which is rugged sphere shaped and composition of different kinds of constitutive cells including α, β, and δ cells.7
Particularly, this protocol described the pancreatic islet viability assessment based on color pixel intensity analysis. A widely used two color fluorescence dye viability kit, LIVE/DEAD™ Viability/Cytotoxicity Kit for mammalian cells, was used. This kit works via two probes, calcein acetoxymethyl (calcein AM) with esterase activity, and ethidium homodimer (EthD-1) with plasma membrane.8 After the staining process, confocal laser scanning microscopy (LSM 510 META, AxioObserver, Carl Zeiss) was carried out to qualitatively capture the images of islet. By setting the excitation wavelength at 488 nm for EthD-1 and 543 nm for calcein AM, data was collected from LSM with emission wavelength at 518 nm and 588 nm, respectively. It was followed by converting the binary images to colored images by painting live cells with green and dead cells with red, and then subsequently stacking multiple images along the z-axis to represent the 3D structure of islet. Finally, the converted image was processed to average projection of voxels (3D pixels)9, since existing methods in just 2D plane cannot represent the 3D architecture of islets.10 Contrary to the preceding protocols that quantify these images manually with naked eye4, 11-13, here we suggest a protocol and its related algorithm for quantitatively analyzing islet viability with the obtained images from LIVE/DEAD™ assay. Additionally, a prototype of islet viability operating software was developed based on computer vision library, named OpenCV. It generates the numerical values by combining the existing cell detection algorithms, color space conversion to luminance sensitive CIE L
b*, and intensity analysis with split color channel histograms.
More specifically, the program includes the following functions and processes (Figure 1). In the first step, a copy of the original image is converted to the mask (binary image) through step-by-step imaging processes containing Gaussian blur, color space conversion, threshold and morphological process. In these steps, the program removes background clearly. Two existing methods, threshold with Gaussian blur and morphological processing were adapted. Gaussian blur is essential process to remove the image noises and to smoothen the border lines. It enables users to detect cell contours easily and to remove the background fluorescence of the materials in solution. The threshold in CIE L
b* color space is similar to the high-pass luminance filter that processes the image by lower intensity boundary value given by the user. Finally, morphological process removes negligible size of islets (mostly far away from ROIs in z-axis) and impurities. After processing the mask image, the mask and the original image are bitwise operated and combined into one implying the clear removal of the background and small impurities. Following to it, implicit pixel data from the thresholded image is calculated to obtain the average intensity, contours of islets, and red/green color histogram. Finally, the numerical islet viability result is displayed on the user interface.