Improvement of Cell Counting Program by using Image Processing Methods

Abstract

Cell counting is one of the most important, common tasks in experiments that deal with cell. Traditionally, manual cell counting uses hemocytometer which is the gold standard for cell counting for biomedical research. However, manual cell counting is time consumed; this may also lead to inaccurate counting results, especially if the counting operators have many samples to analyze over a short period of time. In particular, computer- assisted programs can reduce time consumption, as well as avoiding manual artifact about cell counting. Furthermore, it can diminish the variation of counting results among the counting operators which sometimes is related to personal judgment. For this study, input data are microscopic images that are stained with Trypan Blue dye. The main objectives of this study are to propose a computer-assisted cell counting approach based on image processing techniques and to improve the accuracy and reliability of cell counting, especially dead cells counting. The proposed computer-assisted cell counting algorithm works in four stages. First, image denoising is applied by using image guided filter, since original microscopic staining images have background noise and small debris. Secondly, a thresholding method is used to extract background and foreground objects, and then the third stage is to apply the morphological operations for image analysis. The final stage is to analyze and identify live and dead cells by using object analysis. There are two approaches in this study: the first cell counting algorithm by using morphological operations of digital image approaches and image segmentation and the second algorithm by using the detection of live and dead cell based on adaptive K-means clustering. The first approach gave 83+6% and 63+10% for the accuracy of live cell counting and dead cell counting respectively when compared with the experts. The results show that the performance of the second approach using an adaptive K-means clustering reaches 8914% of live cells from three experts, showing a good likelihood in clusters of live cells and 6123% for dead cells accuracy, whereas ImageJ obtained 491% for total counting cells only compared with manual counting. The correlation coefficients between the counting results from the first approach and the experts were 0.99 and 0.74 for the living cells and dead cells. The second counting algorithm also displayed the highest correlation (r = 0.99, r-0.99, r=0.99) with three manual counting for live cells while the ImageJ did not correlate well with manual counting (r-0.91, r=0.92, r=0.91). In addition, the second approach has a good correlation with manual counting for dead cells (r=0.93, r=0.80, r=0.85) which is higher than the first approach. Therefore, both proposed computer-assisted show that the reliability and accuracy of counting are increased, especially live cells when compared with the experts. Finally, Graphical user interface (GUI) is developed to make this a user-friendly application for Trypan Blue stained microscopic image cell counting.