Novel Adaptive Histogram Binning-Based Lesion Segmentation for Discerning Severity in COVID-19 Chest CT Scan Images

Coronavirus sickness (COVID-19) recently adversely disrupted the medical care system and the entire economy. Doctors, researchers, and specialists are working on new-fangled methods to detect COVID-19 relatively efficiently, such as constructing computerized COVID-19 detection systems. Medical imaging, such as Computed Tomography (CT), has a lot of opportunity as a solution to RT-PCR approaches for quantitative assessment and disease monitoring. COVID-19 diagnosis based on CT images can provide speedy and accurate results. A quantitative criterion for diagnosis is provided by an automated segmentation method of infection areas in the lungs. As an outcome, automatic image segmentation is in high demand as a clinical decision aid tool. To detect COVID-19, Computed Tomography images might be employed instead of the time-consuming RT-PCR assay. In this research, a unique technique is provided for segmenting infection areas in the lungs using CT scan images from COVID-19 patients. “Ground Glass Opacity (GGO)” regions were detected using Novel Adaptive Histogram Binning Based Lesion Segmentation (NAHBLS) method. Many metrics were also employed to evaluate the proposed method, including “Sorensen–Dice similarity”, “Sensitivity”, “Specificity”, “Precision”, and “Accuracy” measures. Experiments have shown that the proposed method can effectively separate the lung infections with good accuracy. The results show that the proposed Novel Adaptive Histogram Binning Based Lesion Segmentation based on automatic approach is effective at segmenting the lesion region of the image and calculated the Infection Rate (IR) over the lung region in Computed Tomography scan.


INTRODUCTION
COVID-19 is a widespread illness which destroys thousands of individuals every day.COVID-19 is a pandemic that is spreading over the world and wreaking havoc on global public health.Coronavirus , which was originally discovered in Wuhan, China, in December 2019, has spread to over 200 nations and territories.It could be spread via respiratory droplets and direct touch.The infection produces inflammation in the alveoli, which fills the lungs with fluid or pus, making it harder for the sufferer to breathe.One of the most effective methods for infected tree pruning was early detection of COVID-19 condition (Chen et al., 2020a).Many countries health-care systems are being swamped by the increasing number of COVID-19 patients.
Early diagnosis and consistent sternness evaluation of "Unilateral", "multifocal", and "peripherally based Ground Glass Opacities (GGO)", "interlobular septal thickening", "thickening of the adjacent pleura", "presence of pulmonary nodules", "round cystic changes", "bronchiectasis", "pleural effusion", and "lymphadenopathy" are all common signs of COVID-19 on CT images (Ding et al., 2020;Meng et al., 2004).COVID-19 infection requires accurate and quick identification of these diseased cell alterations.Human Labeling by well-experienced expert clinicians is protracted, vulnerable to "inter-and intra-observer variability", and reduces dejected CT processing due to the large number of afflicted patients.Due to the severity of the disease and the huge workload and experience required for doctors, adopting chest CT scans for COVID-19 detection has become extremely problematic due to an important surge in the plenty of individuals.Hence, having a reliable computerized way for identifying and quantifying diseased lung parts exert be enormously valuable.
Chest Computed Tomography (CT) can detect specific characteristic symptoms in the lung related with COVID19 in clinical practice such as "Ground Glass Opacities (GGO)" and "irregular consolidation plaques" are the utmost related imaging characteristics in pneumonia associated with SARSCoV2 infection (Lei et al., 2020;Ng et al., 2020;Chung, et al., 2020).As a result, a low-cost, accurate, and effective approach investigative aid for initial analyze and diagnosticate of COVID-19 is recommended (Li et al., 2020a,b;Pan et al., 2020a,b;Ye et al., 2020) .In Computed Tomography (CT) lung imaging, radiologists have known three forms of abnormalities connected to COVID-19: (1) "Ground Glass Opacification (GGO)", (2) "Consolidation", and (3) "Pleural Effusion" (Ye et al., 2020;Shi et al., 2021).The term GGO denotes to a trivial growth in lung weakening that allows the primary veins will be seen (Hansell et al., 2008)."Consolidation" is defined as an increase in lung intenseness that obscures the primary arteries (Hansell et al., 2008)."Pleural effusion" is the collection of supererogatory liquid amid the layers of the pleura external the lungs, also identified as water on the lungs.Fig. 1 shows the most prevalent abnormalities in CT scan.Various scanning image types, as well as overall location and dispersion, can be regarded of as unique indications of COVID-19 and offer helpful insight for figuring out the phase and seriousness of the condition (Yuan et al., 2020;Pan, 1996;Bernheim et al., 2020;Song et al., 2020).For COVID-19 patients, radiologists must perform two tasks: localization and severity assessment.The goal of identification is to find COVID-19 individuals amid other patients so that they can be isolated as soon as feasible.Medical workers can use severity quantification to prioritize people that who will need critical health care.Both tasks require a significant amount of estimation period on the part of radiologists.In order to

Research Objectives
The primary goal of image segmentation is to enhance the surgeon's finding by automatically recognizing and classifying abnormal patterns.The major goal of this study is to use image processing techniques to build a diagnostic approach for identifying COVID-19 infected region (Lesion).In order to achieve this, Novel Adaptive Histogram Binning based Lesion Segmentation (NAHBLS) method is used to isolate the affected region from Computed Tomography (CT) medical image modality.Since it extracts the items of our interest for further processing such as description or recognition, segmentation is a key stage of the image recognition system.The classification of image pixels is achieved using image segmentation.To do object analysis, segmentation techniques are utilized to separate the essential region from the image.To evaluate an algorithm's efficiency, parameters such Dice Similarity Coefficient, Structural Similarity Index Method and Accuracy can be measured.
CT scanning is a common method for diagnosing lung disorders (Li et al., 2020a,b;Sluimer et al., 2006).In practice, segmenting various organs and lesions from chest CT slices can offer doctors with important information for diagnosing and quantifying lung disorders (Kamble et al., 2020).The execution of "image segmentation models" utilizing various "deep learning approaches" has improved dramatically in recent years (Gordaliza et al., 2020;Badrinarayanan et al., 2017;Chen et al., 2018).Fan et al. (2020) proposed an instinctive detect infectious sections from chest CT scans using Deep Network (Inf-Net).The experimental result for proposed Inf-Net was 0.579,0.870,0.974,0.500and 0.047, and proposed "Semi-Inf-Net" acquired 0.597,0.865,0.977,0.515and 0.033 for Dice, Sensitivity, Specificity, Precision and Mean Square Error respectively.Some of the limitations in this study are the Inf-Net mainly concentrates on segmenting lung infections for COVID-19 patients.Furthermore, they utilize the "Inf-Net' to acquire the infection regions for our multi-class infection labelling framework in order to provide guidance for the multi-class labelling of various lung infections.
A CT image collection including 46,096 images of either normal and diseased patients that has been classified by qualified clinicians have been provided by (Huang et al., 2018).It was gathered from 55 control patients and 106 patients who had been admitted with 51 proven instances of COVID-19 pneumonia.The system attained a per-patient "sensitivity of 100%", "accuracy of 92.59%", "specificity of 81.82%", "PPV of 88.89%", and "NPV of 100%" in the 27 probable people using the radiologist's findings as the gold standard.Chen et al. (2020b) presented a novel deep learning model for segmenting COVID-19 infection areas in chest CT scans quickly and accurately.The model is built on the U-Net architecture for multiple areas auto segmentation for COVID19 and extracted features from the CT image using aggregated residual transforms.For multi-class segmentation of CT images, the residual attention U-Net achieved a Dice coefficient, Accuracy, Precision of 0.94, 0.89, 0.95 for infectious lesions segmentation.Xu et al. (2020) developed "3D deep learning algorithm", the appellant debility areas be first segregated out of the "pulmonary CT image" collection.Using a location-attention classification model, these unglued images were then divided into the COVID-19, "Influenza A Viral Pneumonia (IAVP)", and "Irrelevant To Infection (ITI)" groups, along with the accompanying confidence scores.This research has several restrictions.First off, there may be some overlap between COVID-19 symptoms and those of other pneumonias as "IAVP", "organizing pneumonia", and "eosinophilic pneumonia".Solitary the CT appearance of "COVID-19 and IAVP" were compared.Second, there weren't as many model samples in this study.To increase accuracy further, there should be more training and test samples accessible.According to the investigational results of the standard collection, the thorough accuracy rate for all CT instances combined was 86.7%.Gozes et al. (2020) used large number of international datasets included, including data from disease-infected regions of China.They introduce a method that combines clinical knowledge with powerful "2D and 3D deep learning models", changing and reviving existing AI models.As per thoracic CT studies, classification results for patients of "Coronavirus vs. Non-Coronavirus" were 0.996 AUC (95%CI: 0.989-1.00).Shan et al. (2020) has developed the "VB-Net neural network" is used in the DL-based segmentation to split it up COVID-19 infection regions in CT scans.A "Human-In-The-Loop (HITL)" technique is employed to help radiotherapists enhance the instinctive explication of specific case, speeding up the manual demarcation of CT images for training.According to the validation dataset, the developed method produced "Dice similarity coefficients" of 91.6%10.0%between instinctive and physical segmentations and a "mean POI estimation error" of 0.3% for the entire lung.Some of the limitations are as the validation CT datasets, might not be typical of all COVID-19 patients in more topographical zones, were first gathered in a single center.Further validation of the deep learning system's generalization using multi-center datasets is required.Second, the approach was only designed to measure infections; it might not be useful for measuring other types of "pneumonia", like "bacterial pneumonia".Gao et al. (2020) proposed a "Dual-Branch Combination Network (DCN)" for COVID-19 analysis that can attain individual-level classification and lesion segmentation.A unique "lesion attention module" were designed to incorporate the interposed segmentation results and emphasis the classification branch more intensely on the infection locations.The developed DCN outperformed existing models, achieving a classification accuracy of "96.74% on the internal dataset" and "92.87% on the external validation dataset".Amyar et al. (2020) proposed a novel multi-task deep learning model for identifying COVID-19 images and segmenting lesions."Image reconstruction", "Infection segmentation" and "Image classification" are all task of a "Novel Multi Task Learning (MTL)" architecture.This study was helpful to increase "segmentation and classification" attainment by combining information from multiple related tasks.Segmentation method utilizing a multi-task approach with a "dice coefficient of 88%", which is 10% higher than utilizing the condition of the U-net alone.The segmentation findings exceeded similar models lacking a multitask learning technique and when combining a peer of tasks, with a specificity of 99.7% and a sensitivity of 90.2%.Zhou et al. (2021) presented UNet based segmentation network with an attention mechanism.They incorporating an attention mechanism, including a "spatial attention module" and a "channel attention module", to a UNet construction to reweight the feature depiction substantially.To deal with tiny lesion segmentation, a focused Tversky loss is implemented in this study.The obtained Hausdorff Distance is 18.8 and the Dice Score is 83.1%.Some of the limitations in this study are the tiny dataset places restrictions, the system is built to section the solo label, and they intend to use their technique to segment further several categories data chore and contrasted with added relevant approaches.Saeedizadeh et al. (2021) a segmentation system for detecting COVID-19-infected chest areas in CT images.On a voxel level, ground glass areas were detected using an construction equivalent to that of an "Unet model".The presented approach is named as "TV-Unet".Valuation of the forecasted segmentation findings to quantifiable assessment of segmentation effectiveness ("accuracy", "recall", "Dice score", and "mIoU") showed excellent capacity to classify COVID-19 related areas of the lungs, yielding a "mIoU" rate of over 99% and a "Dice score" of over 86%.Zheng et al. (2021) proposed 3D CU-Net is a Convolutional Neural Network that extracts rich features and fuses multiscale global information to automatically identify COVID-19 infected areas from "3D chest CT images"."3D CU-Net" is built on the "3D U-Net architecture".The 3D CU-Net attention method for localized inter data learning in an encoder to progress distinct layers of feature depiction and build a "pyramid fusion module" with "expanded convolutions" at the end of the encoder to fuse multiscale context information from high-level features."Tversky loss" is employed to solve glitches with lesions of varying sizes and distribution.Qiblawey et al. (2021) a "cascaded" approach is presented to slice the lung, locate and measure COVID lesions.Furthermore, the patient's infection percentage in the lung is determined, allowing the sternness of the provided size to be divided into quatern categories grounded on the lung's diseased proportion.The suggested system used the FPN with DenseNet201 encoder to obtain a well-designed efficiency for COVID infection segmentation with a "DSC of 94.13%" and "IoU of 91.85%".Furthermore, the proposed strategy successfully detected COVID-19 with "99.64% sensitivity" and "98.72% specificity".To conclude, over a dataset of 1110 participants, the system was able to distinguish between distinct COVID-19 infection severity levels, with sensitivity values for "mild", "moderate", "severe", and "critical infections", respectively, of 98.3%, 71.2%, 77.8%, and 100%.Table 1 depicts the summary of the various research work for Segmentation on COVID CT scan images.

DATASeT DeSCRIPTION
349 CT images positive for COVID-19 from 216 individuals, as well as 397 CT scans negative for COVID-19, are included in the COVID-CT collection.The dataset is openly accessible to the public to encourage study and development on COVID-19 Computed tomography diagnostics.COVID-CT is a dataset that can be found at https://github.com/UCSD-AI4H/COVID-CT. Table 2 depicts the summary of the dataset used in this research work.
For the segmentation tasks, all of the images were converted to Portable Network Graphics (PNG) format and downsized to 256*256 pixels.

CONTOUR LUNG MASK SeGeMeNTATION MODeL (CLMSM)
Computer-guided analyses of "parenchymal density", airway and "emphysema evaluation", "lung nodule detection", and "lobe-based studies" all require lung segmentation.Because airflow has a low lessening rate on CT scans and reside in the majority of the lung, the "lung tissue" appears dark while the surrounding tissue appears bright.Due to variances in "pulmonary inflation" and an elastic chest wall, there may be significant discrepancies in volumes and margins when trying to automate the segmentation of lungs.Moreover, the lung margins may fail if there is lung disease present.Further difficulties with lung region separation include the following.Some of the other challenges for separation of lung region are as: Segmenting the lung regions is a key step in the process of detecting lesion region in COVID images.Some post-refinement techniques on the first lung mask are required to solve this problem.The lung area was separated with a "Contour Lung Mask Segmentation Model".
The First step in Contour Lung Mask Segmentation Model.From the grey scale image need to create a binarized mask.If the pixel's value is below the threshold, it will be given the maximum value, which is white, in inverse binary thresholding.It will be allocated 0 (black) if it is greater than the threshold.Three arguments are required for this function: Input image as Grey Scale COVID image, threshold value as mean of the image, and maximum value as 255.The Algorithm 1 pseudocode for binary inverse thresholding is presented below.

INPUT-GREY SCALE IMAGE G (X, Y) OUTPUT-BINARY MASK IMAGE START READ Grey Scale Image as source image (X, Y) SET as Threshold value T = MEAN of Grey
A multi-stage method is used by the Canny edge detector, an edge detection operator, to find a variety of edges in images.Canny edge detection is a method for collecting important structural information from a variety of visual objects while drastically lowering the quantity of data that has to be processed.The Canny edge detection technique is divided into five steps: Recursively locate "NON-EDGE PIXELS" using M ³ T l and with an "EDGE PIXEL" as a neighbour.
Appended as edge pixels to them.END Apply Morphological operation to the canny edge detected image.The largest value of all the pixels in the input pixel's neighbourhood is the value of the output pixel.If any of the pixels in a binary image are assigned to the value 1, the output pixel is also set to 1.The Algorithm 3 pseudocode for dilated image is presented below,

END IF END IF END FOR END
Contour detection can be used to find boundaries for any object in an image.A contour is a closed curve that connects all continuous points of varying color or intensity, and it represents the contours of objects in an image.Shape analysis and object detection and recognition improve the contour detection.The shapes of the objects in the image are represented by contours, which are abstract groupings of points and segments.As a result, manipulate the contours such as count the number of contours present and classify the forms of object, crop objects from an image segmentation.The Contour detection technique is divided into five steps: 1. Apply to the dilated image.2. Apply Binary thresholding to the image.This changes the images to black and white, emphasizing the important areas to enable the contour-detection algorithm do its job more quickly.Thresholding renders the target border of the image completely white, with a consistent intensity across all pixels.The algorithm can now determine the edges of the object from these white pixels.3. Find Contours and get the larger contours first, run them via the sorted function.Thresholding is one of the most basic and often used image segmentation algorithms for separating light objects from dark backgrounds based on image features including color, intensity, and texture (Qiblawey et al.,2021;Abubakar, 2013).Edge segmentation finds the edge information in an image and transforms the original image into edge images using edge detection operators.Pixels that connect the boundaries of two sections constitute an edge.The region-based approach, also known as "Similarity-Based Segmentation" is used to directly determine region (Yogamangalam. & Karthikeyan, 2013;Manikannan & SenthilMurugan, 2015).It divides an image into similar sub-areas grounded on texture, color, intensity, and other characteristics.Fig. 3 represents the various segmentation approaches.
Although there have been many segmentation algorithms developed, each has its own set of applications and restrictions.The discusses the features of various approaches in (Hou et al., 2006)."Histogram-based thresholding", which implies that homogenous items in the image exhibit itself as groups, is one of the most extensively used approaches for image segmentation.The essential to the histogram-based technique is choosing a set of criteria that can distinguish between foreground and background images.Over the times, several histogram-based thresholding approaches have been presented.These techniques can be divided into two types.The first category includes thresholding algorithms that optimize a certain objective function to obtain the best thresholds (Li & Lee, 1993;Pal et al., 1996;Małyszko et al., 2010;Otsu et al., 1979;Albuquerque et al., 2004;Yin, 2007;Thanh et al., 2019).Entropy-based approaches are the most common among these thresholding strategies, and several algorithms have been presented in this approach.
A histogram is a graph that counts or visualizes the frequency of data (i.e., the number of occurrences) across bins of discrete intervals.Histograms are useful in many areas of data and image processing.The histogram is made by dividing the data range into equal-sized bins which are called classes, in other words called as regions.The number of points from the data set that fall into each bin is then counted for each bin.In vertical axis counts for each bin is calculated and variations are calculated in horizontal axis.The horizontal axis in image histograms is made up of pixels values are given.In this research, proposed Novel Adaptive Histogram Binning based Lesion Segmentation (NAHBLS), a novel segmentation system that employs a binning histogram structure to adaptively find threshold values.
In this work, images are collection of pixel elements which is ranging from 0 to 255.For ranging of the pixels, histograms help to visualize the distribution of pixel in images.The histogram's x-axis reflects the count of bins, while the y-axis shows the frequency of every bin.The volume of bins is a number that can be changed depending on how you choose to represent your data's distribution.The goal is to choose a bin width that produces the best accurate representation of your data.
Using Equal Frequency Binning which helps to produce bins have an equal frequency, divided the range of the pixel values into intervals that contain equal number of pixels values.To calculate the best number of bins to utilized in an equal frequency binning apply Sturge's Rule.The following formula is used by Sturge's Rule to calculate the optimal number of bins to deploy in an equal frequency binning value:

PROPOSeD APPROACH FOR FINDING INFeCTION QUANTIFICATION USING LUNG AND LeSION SeGMeNTATION
Based on the binning thresholds, segmented the pixel values into different regions.Regions are R 1 , R 2 , R 3 , R 4 in the image.For these regions separated into different colors.Saturation measures the amount of grey in a particular color using the Hue Saturation Values (HSV), where Hue is the pigment aspect of the model, and Value determines the intensity of the colour.Apply the HSV to the Segmented Images and this will help to separate the Lung Region in one color, Lesion Area in other color and Background in other color.

Infection Rate (IR)
The Lesion region (sum of green pixels) over the Lung area (sum of red pixels) for CT scan image were used to compute the Infection Rate (IR).Fig. 5 shows the suggested method for computing the infection rate percentage for CT images:

IR Number of Green Pixel in Lesion Area
Number of Red Pixel = over Lung Area (2)

eVALUATION MeTRICS FOR SeGMeNTATION
The performance measures used to estimate chest CT scan images of the lung and lesion segmentation include "Dice Similarity Coefficient", "Intersection over Union", "Accuracy", "Precision", "Sensitivity", "Specificity"," F1-score" and "Structural Similarity Index Method".The performance measures that are widely used in medical image evaluation methods are shown below.

Sorensen-Dice Similarity
Consider A to be the segmented regions for which insist to measure the quality, and B to be the ground truth.The following formula is used to calculate the Sorensen-Dice similarity (Hore et al., 2010): (3) The Sorensen-Dice similarity statistic has a value of 0 to 1.The improved the "segmentation result", the higher the Sorensen-Dice value."TP", "TN", "FP", "FN" represent the "True Positive", "True Negative", "False Positive", and "False Negative", respectively.

Intersection Over Union (IoU)
The overlay amongst the prediction and the ground truth is measured by Intersection Over Union (IoU)."Intersection-Over-Union" is a frequent image segmentation evaluation metric and the formula is shown below:

Precision
Precision is considered as the proportion of accurately expected optimistic cases, as shown in the equation below:

Sensitivity (Recall)
The recall given in the following equation is the ratio: SENSITIVITY = + TP TP FN (6)

Specificity
The percentage of successfully "predicted negative class" samples to all "negative class samples" is known as specificity:

F1-Score
The "F1-score" is the proportional mean of "precision and recall", as shown in the equation below.

Accuracy
Accuracy is defined as the fraction of "correct predictions" among the "total number of predictions", as shown in equation below:

Structural Similarity Index Method
The "structural similarity index" is a method for determining exactly similar two images seem (Hore et al., 2010, Nivetha, et al., 2022a,b, 2021).In this method, image degradation is considered as the change of perception in structural information.Pixels that are considerably interconnected or temporally constrained are referred to as having structural information.Thus, it functions with various essential perception-related data including brightness masking, contrasting masking, and other related information.The technique used to make an image's deformation fewer obvious at its edges is known as luminance masking.Contrast masking, on the other hand, describes the method of lessening the visibility of texture distortions in an image: The mean intensity can be estimated as: The Standard deviations can be estimated as: The Co-Variance can be estimated as: where µ A and µ B denotes the mean values of original and distorted images.And σ A and σ B denotes the standard deviation of original and distorted images, and σ AB is the covariance of both images and C 1 and C 2 are constants.

CONCLUSION
A Novel Automated Histogram Binning Based Lesion Segmentation (NAHBLS) approach is proposed in this work.Grounded on the Histogram Binning approaches for identifying the lesion areas of the lungs in COVID19 patients.This study demonstrated a strategy for segmenting COVID-19 infections and measuring the results.The dataset used in this study is GITHUB CT dataset ("https://github.com/UCSD-AI4H/COVID-CT").Furthermore, the proposed approach included Contour Lung Mask Segmentation Model (CLMSM), Novel Automated Histogram Binning Based Lesion Segmentation and Infection Quantification.The proposed approach worked well, and all of the procedures were necessary for producing outstanding outcomes.The proposed method got a Dice of 96.58% with 95.52%accuracy for Lesion segmentation.This shows how effective the proposed strategy was at detecting lesions infections and finding infection rate.As a result, it is expected that the proposed strategy will play an important role in clinical practice.It is useful in practice since it can alert specialists to both illnesses and strategies to combat them.In Future, COVID-19 lung abnormalities can be classified into different kinds of infection, which can provide useful information depending on the severity and stage of the condition.New architectures could be tested and constructed in future for detecting severity of the disease.

Figure 1 .
Figure 1.The most prevalent CT patterns found in COVID-19 pneumonia: (a) The GGO pattern is shown with white arrows, and (b) the consolidation pattern is shown with red arrows.(Shi et al., 2020).(c) Pleura Effusion is shown with white arrow 4. Draw the contours to the image.5. Get mask of the image.Algorithm 4 pseudo-code for largest contour approximation is presented below., Fig. 2 Shows that Lung Segmentation from CT COVID Images A) Original Grey Scale Image B) Thresholded Image C) Canny Edge Image D) Segmented Lung.Algorithm 4: Largest Contour Approximation INPUT: DILATED EDGES DE (X, Y) OUTPUT: SEGMENTED MASK IMAGE SEG_MASK (X, Y) START READ Dilated Edges Image as source image, DE (X, Y) COMPUTE Find Contour: FOR DE (X, Y) Image segmentation approaches include Pixel-Based, Edge-Based and Region-based segmentation.

Figure 2 .
Figure 2. Process of lung segmentation from CT COVID images: a) original grey scale image b) thresholded image c) canny edge image d) segmented lung

Figure 5 .
Figure 5. Proposed method for calculating the CT image "infection rate" percentage

Table 2 . Overview of the dataset employed in this study
Smooth the image with a Gaussian filter to remove the noise.2. Find the image's intensity gradients.3. To eliminate erroneous edge detection responses, use gradient magnitude thresholding.4. To find probable edges, use a twofold threshold.5. Hysteresis-based edge tracking: Complete the edge detection process by suppressing all weak and extraneous edges.
helps to rounding up to the nearest integer.Optimal number of bins can be calculated for the image pixel values using Sturge Rule.Next Step, to determine the threshold values in the optimal binning, Use Equal Width Binning Approach which helps to determine the threshold values for each optimum binning.From the threshold values, form the different regions in the image.Regions are having the background, lungs regions, Affected Region and Unaffected Region.Algorithm 5 depicts the procedure for Novel Adaptive Histogram Binning based Lesion Segmentation (NAHBLS).Fig.4represents the complete view of Proposed Methodology for COVID-19 Severity Segmentation.