Article Preview
TopIntroduction
For diagnosing male infertility, a sperm assessment is necessary as an initial and essential step. Among the most important parameters, we can cite sperm concentration (or density), sperm motility (or vitality), and sperm morphology (Nallella, et al., 2006). According to the World Health Organization (WHO), 15% of couples in the world suffer from infertility. Males are found to be solely responsible for 20-30% of infertility cases and contribute to 50% of cases overall. The evaluation of sperm parameters in fertility clinics and laboratories is analyzed manually by andrologist technicians. Visual problems and fatigue of the human operator negatively affect the results and make them subjective, leading to considerable intra and inter laboratory variability.
Since the 1980s, these analyzes have been performed using automatic systems called “Computer Assisted Semen Analysis” (CASA) (Amann, R. P., et al., 2014). The processing microscopic images in a computer can be useful to extract quantitative information about the specimen. The poor quality of sperm microscopic videos and the collisions of spermatozoids on the “blade + lamella” walls make CASA systems suffer from certain limitations and up today, evaluation of sperm parameters are not yet repeatable and objective . For that the microscopic images used are processed by using digital image processing techniques as enhancement techniques, segmentation techniques, object measurement, object classification,…etc.
The improvement of sperm analysis by CASA systems is currently a hot topic research. Several studies have been carried out for the automated sperm detection (Syahputra et al., 2018). In 2010, authors in (Abbiramy et al., 2010) proposed a technique for sperm motility detection which is based on Laplacian filter and Median filter as preprocessing step. Then, they segmented the images using simple thresholding and morphological operation with labeling process. The differential method is applied for reducing complexity and faster tracking of spermatozoids. One year after, Mahdavi et al. (Mahdavi et al., 2011) have thought to benefit from the elliptical shape of the spermatozoid head to search on the image that blobs having this form. Simple thresholding has been done to create a binary image from the denoised image. The threshold value is chosen after several experimental tests. The elimination of badly classified pixels is performed by morphological operators. To reinforce the detection of true spermatozoa, the authors used the geometric characteristics of labeled objects as the ratio between the major axis and the minor axis of the ellipse to eliminate false sperm.
In 2017, authors in (Urbano et al., 2017), in order to detect a spermatozoid in an image, have applied a Gaussian filter for smoothing the noisy image then used Laplacien of Gaussian filter (LoG) for detecting contours. The binary segmented image was obtained by using Otsu's thresholding method (Otsu, 1979) followed by many morphological operations. At the end, each object is labeled so they can separate sperm from non-sperm with calculating the centroid coordinates and sperm dimensions. By the year 2018, the authors in (Khalifa et al., 2018) proposed a system for counting and tracking sperms in videos. First, they have used images in HSV domain. Then thresholding process is applied. After the next steps insist to grouping pixels and then accept and reject pixel groups according to a pixel group value to remove noise and an unwanted group of pixels. After that, it calculated the center of the pixel groups and counts all pixel groups which indicated the number of sperm. Drawing a rectangle around the pixel groups is done in final step for tracking purposes.
These works based on thresholding techniques for image segmentation which are efficient approaches to sperm cells image but with highly sensitive performance to threshold values. In other words, slight errors in selecting the optimal threshold may lead to either significant lack of sperms or increased false detections.