Human Epithelial Cell Image Analysis and Segmentation using Threshold Based Fusion Technique

The most demanding aspect of digital image processing is segmenting an image efficiently. Cell segmentation or classifying cells in an image is essential while analyzing cell images in medical research, especially in spot diagnosis, cancer cell detection, and live-cell imaging segmentation forms a crucial component. This research examines existing segmentation algorithms and suggests a new segmentation technique that employs image filtering and thresholding. Thresholding is an essential part of image analysis and segmentation. Finally, the segmented image and the FCM (fuzzy C-means) based clustered image are merged. In terms of accuracy, sensitivity, dice-coefficient, and Jaccard-coefficient, the simulation coupled with ground truth data is proven to produce better segmentation outcomes.

T h e r a p i d d e v e l o p m e n t o f n e w technologies in the field of image processing has helped in discovering human abnormalities at an early stage and is faster than manual methods.Image analysis and segmentation have an essential role in the medical field and help in the early detection of diseases in critical health conditions.Due to the structural complexity and type of the cell image, radiologists may face difficulties in extracting the necessary features of the image, 1 which are crucial in treating the patient.Therefore, correct and meaningful segmentation is a challenging task and is required for further process.
The most used substrate for finding the antinuclear antibody (ANA) is the human epithelial type2 (HEp-2) cell samples, which are a natural protein array with plenty of antigens.The discovery of ANA in human serum is a vital transmission instrument for connective tissue syndromes, and immune ûuorescent (IIF) is the reference method for ANA testing.Indirect immune ûuorescent (IIF-ANA) test is becoming increasingly crucial for analyzing the cell image because of its ease and inexpensiveness 1,2 .The attributes of epithelial type-2 (HEp-2) cells are an oval shape with merged properties of connected tissues, which help in microscopic observation and provide better result in detecting the number of antibodies with better resolution.Therefore, cell image segmentation is an essential task that examines the appropriate biological features such as cell type, shape, group, etc.The experimental studies are focused mainly on (HEp-2) cell image and aim to identify and segment the cells based on ground truth data.
Several computer vision methods have been developed for a variety of applications in image processing.The first phase of computer vision includes pre-processing and image segmentation. 3In this stage dissimilar objects are identified and separated from the background.The next part is feature extraction, where object are measured.The computing of image features to quantitatively evaluate some significant features of object.These clusters are combined to create a feature vector during feature mining. 4,5 he next level is classification.At this point, the output is only a decision to conclude the class of each object.The structural study is an explanation of images to properly recognize and judge the vital details of images. 6 -15

Thresholding
Thresholding is the fundamental way to do image segmentation.The grayscale image is taken as input data, and the output image will be in the form of binary.The simplest thresholding 3 method substitutes every pixel in an image with a dark pixel if the image intensity {\display style I {i, j}} Ii, j is less than definite static constant T (that is, Ii, j < T {\display style I {i, j} < T}), or a bright pixel if the image intensity is more than that constant.

Global threshold
The histogram is a graphical way of representing the image intensity.It reveals two points for an indication from the background and the foreground, respectively.Manual thresholding 3 involves intensity value (threshold) such that total pixels having an intensity value lesser than the threshold fit to one segment, and the rest of the part fits the other.Global thresholding provides better results only when the degree of intensity separation between the two points in the image.It is an unsophisticated segmentation selection 4 .

Otsu's threshold
Image segmentation splits an image into its subareas and is a crucial task in mining image features, especially in medical image analysis and segmentation approaches.Due to its understanding of perception and ease, the thresholding technique is important and essential part of diagnosing medical images. 5,6 sually, the histogram-dependent threshold choice technique, and "the maximum between-class variance method proposed by the Japanese scholar Otsu used".The selection of principle threshold criteria technique reflects the largest interclass difference between the object and background.Similarly, by varying the Otsu's threshold level is either 0 or 1 to get an FCM threshold.Thus, in the analysis of general as well as medical images, this technique provides better results compared to some of the popular methods.

Adaptive threshold
I n a n a d a p t i v e t h r e s h o l d , t h e threshold value is calculated for a particular region, thus it varies from the threshold rate at every pixel position based on the adjacent pixel values. 6[15][16] T[(x,y),f(x,y),P(x,y)=adaptive threshold ..

.(3) Multilevel threshold
Multilevel thresholding is the most applicable and popular image segmentation technique. 7Here, thresholding is assigned to segment a gray-level image into some discrete areas.Here, more than one threshold criterion is chosen to analyze the given image. 10..( 4)

Optimal threshold
Choosing a threshold in the valley between two overlapping peaks classifies the pixels inappropriate way.The selection of optimal threshold and the optimal regulation can be done inevitably by a similar learning process and is explicitly for the optimal selection of threshold used in predications.The directions from examples and counterexamples have been projected by reducing an entropy function. 8he rules have the procedure of disjunctions of conjunctions of predicates.Thus, such a method can study commands from cases, and depends on an optimality principle. 9,10  (5)   • pb (z), p0 (z), prob.distributions of background, object pixels.
• Pb, Po : the a-priori probabilities of background, object pixel. 17

K-means clustering
This algorithm is a simple and effective unsupervised technique, which can be used for various applications like abnormality detection, pattern analysis, image scene understanding, etc.It classifies a sample set Y (y1, y2, y3….y n ) into groups with the endeavor at minimizing an objective function J is given by, 18 ... (6)   where, n represents the quantity of samples and c i is the cluster center, where, the "Euclidean distance measure between a data point y j and the cluster center c i ". 19 Algorithm steps • Compare the distributed pixel intensity of the objects.
• Initialize the centroids with K random intensities.
• Recurrence the step till each cluster get labeled.
... (7)   • Here, N is the sample number of the i th cluster.
• Cluster the points based on distance of their intensities from the centroid intensities.
• Compute the new centroid for each of the clusters. 20

Fuzzy c-means clustering
FCM algorithm for detecting optimal partition of cluster points is one of the popular techniques for analysis of both cell and MRI images.It splits the data points into subareas named cluster groups.Here the clustering points are automatically assigned after the pre-processing stage, and the associate function is the curve that defines how each point in the input space is mapped to the degree of associated value between 0 and 1.During the iteration process, at each cycle, optimized cluster points are generated. 18,21,22 Wre, A represents the image and A = {Z (A, B), 0 < A d" M, 0 < B d" N}, and M*N denotes the cluster points.
• µ c (p, q) is the associated data points of Z (A, B) at the C th cluster, • "d c (p, q) is the Euclidian distance of the data point Z (A, B) from the C th cluster". 21 Clusters number K (2 d" K < M*N) with original clustering matrix values are labeled.
The data points Y c , and fuzzy matrix X c (p, q) are fixed by utilizing equations [9], [10], correspondingly.
The procedure is recurring till the quantities varying among two iterations and is not more than e. 22,23 ... (9)   ... (10)  The rest of this paper is structured as, proposed work in Section 2, the flowchart, and algorithm of the proposed method are presented in section 3, the result are given in section 4, and the conclusions in section 5.

Proposed method
Firstly, the input cell image is transformed to grey scale, and a Gaussian filter is employed to eliminate noise contamination in the cell image. 8A denoised image is then thresholded by an automated higher threshold value in conjunction with binary smoother.However, the thresholded pixel values are fewer than the image's mean frequency.The resulting image is processed with a "remove border" operation, and the thresholded image is equalized using an adaptive histogram.The picture is simultaneously denoised and segmented using an FCM (fuzzy C-means) based clustering approach.Finally, combining these two with the fusion procedure results in a new segmented image.The resultant image is more suitable to ground truth data, and is compared with existing thresholding algorithms. 12

Gaussian filter
The Gaussian filter is a type of linear smoothing filter in which the weights are chosen according to the Gaussian function procedure for removing noise from a normal distribution.Two-dimensional discrete Gaussian zero meanfunction is employed as a smoothing filter in image processing.

Higher_threshold
Bernsen's locally adaptive binarization approach is utilized to extract the adhesive cell image feature from a grayscale picture.It is also ...( 13) If the contrast measure C (x, y) < l, the neighborhood is made up of only one class, foreground or background.Furthermore, r and l values altered based on the image regions employed.

Binary smoother
Binary smoothing is a method for smoothing data or filtering noisy data.The data is first categorized, and the sorted values are distributed across many segments or cells.The technique conducts local smoothing since they refer to a neighborhood of data.Smoothing by binary means replaces each value in a binary with the binary mean value.

Adaptive histogram
The image is splits into discrete blocks, and a histogram is computed for each segment.Thus, adaptive histogram equalization computes many histograms, each of which corresponds to a different part of the picture.Consequently, it improves the local contrast and edge definitions in all segregated sections of the image.

Algorithm steps
Input: Cell image to be segmented Output: Segmented image • A cell image is converted to a gray-scale image.
• Image denoising is performed with the Gaussian filter to diminish the noise contamination in the cell image.
• Higher_threshold value is used to extract the required features from a denoised image, with binary smoother.
• Segmented image is processed by adaptive histogram, which is more accurate and noise-free.resulTs And discussiOns

Performance matrices
The proposed algorithm is coded and executed in (MATLAB R2020a).The cell image (Hep-2) and the corresponding reference data is extracted from SNPHep-2 Dataset.The results are compared with existing methods and are found to be better.

Accuracy
It specifies the perfect class of the image pixels.The measure of accuracy 24   Proposed method represents the pixels which are not belonging to the specified class.FN is incorrectly projected pixels, which are not fitting to the specified class.sensitivity Sensitivity is calculated through positive predicted and negative predicted values, which is the ratio of true outcome of all segmented results. 24nsitivity=TP/((TP+FN)) ...( 15) Dice co-efficient Dice-coefficient examines the identical things in between the segmented data and the reference data, which is retrieved from manual segmentation.Dice-coefficient of segmented image 24, 25 is given by ... (16)   where, and cardinalities of two sets.

Jaccard-coefficient
Jaccard-coefficient is as same as the dice-coefficient, which calculates the commonness in-between ground truth data with different segmentation methods employed in this paper and 24, 25 is given by ...( 17) cOnclusiOn A unique threshold-based image segmentation approach is suggested in this paper.The proposed model's performance is compared to manual and current segmentation approaches, as well as the pre-processing procedure.However, instead of current segmentation strategies, the suggested method delivers superior segmentation by integrating the FCM-based clustering technique with image segmented by higher thresholding, which is one of the study's important advantages.Furthermore, classic threshold-based segmentation approaches such as Otsu's, multi-level, and FCM thresholding methods, suffer, loss of intrinsic information after segmenting the cells, whereas global and adaptive thresholding give superior segmentation, but the threshold criteria must be set manually.Although unsupervised clustering techniques such as K-means and FCM are popular, the findings of both approaches do not meet the acceptable level throughout the assessment, particularly for adhesive cell segmentation.As a result, the proposed method has an edge and is useful in segmenting the adhesive Hep-2 cell image without sacrificing essential information at the fusion level.While examining the biological aspects of the Hep-2 cell image, radiologists will find the entire research quite useful.The simulation output of the proposed study is more effective and efficient in terms of computing.It is discovered that performance metric's sensitivity parameter improves by around 3%, and the segmented image's effectiveness is assessed in terms of accuracy, sensitivity, dice-coefficient, and Jaccardcoefficient.The main contribution of this work is the improvement of performance measure, which is supported in all of these factors and is clear from the outcomes of the proposed algorithm.

Conflict of interest
There is no conflict of interest.

Funding source
There are no funding sources tested for various neighborhood values and contrast limitations.T(x,y)=(Z_low )+((Z_high ))/2 ...(12) (Z low ) is the lowest and (Z high ) is the highest gray level pixel value in a square r x r neighborhood centered at pixel (x, y), C(x,y)= (Z_high )-(Z_low )

Fig. 7 .
Fig. 7. Performance evaluation of suggested study with existing techniques

Table 1 .
Comparison of Hep-2 cell image segmentation result