Knee Osteoarthritis (OA) is a joint disease that is commonly observed in people around the world. Osteoarthritis commonly affects patients who are obese and those above the age of 60. A valid knee image was generated by Computed Tomography (CT). In this work, efficient segmentation of CT images using Elephant Herding Optimization (EHO) optimization is implemented. The initial stage employs, the CT image normalization and the normalized image is incited to image enhancement through histogram correlation. Consequently, the enhanced image is segmented by utilizing Niblack and Bernsen algorithm. The (EHO) optimized outcome is evaluated in two steps. The initial step includes image enhancement with the measure of Mean square error (MSE), Peak signal to noise ratio (PSNR) and Structural similarity index (SSIM). The following step includes the segmentation which includes the measure of Accuracy, Sensitivity and Specificity. The comparative analysis of EHO provides 95% of accuracy, 94% of specificity and 93% of sensitivity than that of Active contour and Otsu threshold.
Osteoarthritis is a type of disease that mainly destruct the joint in knee, hip and hand and this disorder may happen when the cartilage cushion down the knee which starts to wear down [
In Osteoarthritis there are several imaging techniques applied to restore the images from the disadvantage of unclear images and accuracy in image. The techniques are as follows: deep learning algorithm using Conventional Neural Networks (CNN) and for segmentation Active Contour Algorithm may be existed [
This section of this paper explains the related papers that deal with the Detection of Osteoarthritis Based on Thresholding methods. In 2019, Shiva and Gornale et al., have proposed a segmentation technique for evaluating knee osteoarthritis to overcome the disease using different segmentation methods. Though they have used the Prewit and Sobel edge methods, the accuracy enhancement and the classification rate are the main limitations. In 2019, Sicheng Wang et al., have done research on segmentation aware of denoising without true segmentation to generate image denoisers with better quality with no ground truth segmentation. enhancing U-SAID's generalizability in three ways: denoising unseen types of images, denoising as pre-processing for segmenting unseen noisy images, and denoising for unseen high-level tasks. To improve the accuracy of the image and to enable automatic evaluation, in 2018, Archit Raj et al., proposed an automatic knee cartilage segmentation using CNN in a 3D structure. Despite the fact that they proposed a technique to improve image accuracy, they are still in the limitation stage. In 2017, Shivan and Gornale et al., proposed a determination of OA using a histogram of oriented gradients and multiclass SVM. This method uses a HOG technique and can be processed using multiclass SVM for evaluating the KL grading system. Its results can be evaluated by medical expertizers. Still has the disadvantage of segmentation accuracy and could improve the classification rate with improved techniques for segmentation and preprocessing.
In 2019, Prajna desai et al., had proposed a paper called “Knee cartilage segmentation and thickness measurement from 2D ultrasound. This paper proposes local-phase based image processing to enhance the modality of the image, and different segmentation techniques are proposed called RW, Watershed, and graph-cut-based methods. This existing method also tried to improve the accuracy. Though it doesn't fulfill the objective of improving accuracy. In 2020, Ridhma et al., proposed a paper on the Review of automated segmentation approaches for knee images, owning the different accuracy and difficulty of the various data sets of the knee image. This paper gives various segmentation techniques in detail for making a proposal. They did a lot of research to get better results for knee image segmentation. To separate the original histogram of a 16 bit biomedical image into two Gaussian that cover the pixel region and bright pixel regions was proposed by Joyce Sia Sin Yin et al. in 2020 in “Prominent Region of Interest Contrast Enhancement for Knee MRI: Data from the Osteoarthritis Initiative”, though it has a trade-off of longer time in image processing. It is observed from the review of the literature that the existing papers talked about Detection of Osteoarthritis based on Thresholding methods. Hence, there is a need to develop an efficient Thresholding methods model for Detection of Osteoarthritis.
The color and contrast of an image can be analyzed and processed in digital images using a tool called image histograms. In our proposed work, the knee osteoarthritis image can be enhanced using a method called Correlation Histogram dependent on the gray scale level of CT images, referred to as the CHA_CT algorithm. They are ideally suited for analyzing the effect of various processing and coding algorithms. The image can be fed up with a number of pixels. Each pixel represents a tissue in a CT image, and a gray-scale value between 0 and 255 is assigned to it.
In order to reduce the noise of the image along with less blurring, Median Filter can be used in our research. Because of the black and white spots that are overlaid in it, it must be called pepper and salt noise [
We extracted a Region of Interest (ROI) of 140 × 140 mm for each knee to pre-process the Osteoarthritis Initiative (OAI) images utilizing an ad-hoc template and Bone Finder software31 that allows precise completely anatomical. To standardize the communication frame between patients and data acquisition centers, this Technique was made [ ROI extraction Creating histogram analysis Correlation of histogram Analyze image contrast Enhanced image
To convert knee image into binary image in ROI extraction to enhance the knee,
The standard deviation of the region of interest can be calculated. The enhancement parameter is calculated by,
l is the mean estimation of the Region of interest
The outcome is then analyzed by qualitative and quantitative analysis to determine the fundamental features of the equalized histogram image, using a total of 10 patients that are used to contrast the image for research.
a, b are the coordinates of the threshold value point p (a, b)
f (a, b) are points the gray level image pixels.
Threshold image g (a, b) can be defined:
K implies constant adjacent object boundary and
Here N represents number of pixels in the gray scale image.
The contrast measure of bernsen algorithm includes the following steps;
Step 1: It defines the maximum and minimum value of segmented 5 × 5 gray scale image.
Step 2: It calculate the local contract with predefined value d
Step 3: if d >
Step 4: if d <
The detection of osteoarthritis in the knee using EH Optimized Thresholding Based Segmentation is a unique method. Thresholding is used to create binary images from a grayscale image in this manner. EHO is a metaheuristic optimization method inspired by nature and based on the herding behavior of pictures. It's used to solve the problem of multilevel image thresholding. The EHO optimization mechanism is as follows.
The elephant clan is selected during initialization, i.e., the gray scale section of pixels was chosen. Solution creation means that initiation is a crucial step in the process of optimization, leading you to identify the optimal solution easily.
The pixel with absolute value is considered as the best pixel and considered as matriarch. After generating the solution, the fitness function is evaluated and then chooses the best solution. Optimization algorithm for the most part relies upon its fitness function to acquire the best solution.
The movement of elephant in the clan the following state is approached through matron, then desired new location is given as,
The suitable elephant,
The suitable position of the elephant is obtained by the gathering of information's from all Elephants
The centre of the clan is given by
The elephant clutches in the time the male elephants reach adolescence and they travel and survive solitarily [
The evolution of EHO is obtained by the separation and the updating operator. The obtained optimum threshold value from EHO is fed to the processing and segmentation of gray scale image using Benson and Niblack algorithm to obtain the resulted image [
The proposed method was implemented by Matlab 2017. The outcome of the proposed method shows the result of pre-processing and segmentation of CT image for knee osteoarthritis and the performance was evaluated with different measurement parameters as accuracy, sensitivity and specificity. To evaluate the efficiency, the output of proposed med technique is compared with different existing techniques.
In our proposed work, the sample CT bone images were considered from CT-ORG database. The sample dataset has 140 images of which 85 images were healthy (training set) and rests were used for testing purpose [
The experimental result of Knee OA CT image using a correlation histogram analysis and the optimized thresholding is analysed through different steps;
From the experimental analysis, it is obvious that the osteoarthritis can be analysed and processed by this technique. As we mentioned in
The experimental results were evaluated with MSE, PSNR and SSIM [
Patient number | Local phase based bone enhancement | Bone shadow region enhancement | Proposed correlation histogram | ||||||
---|---|---|---|---|---|---|---|---|---|
PSNR | SSIM | MSE | PSNR | SSIM | MSE | PSNR | SSIM | MSE | |
1 | 56.41 | 0.966 | 0.48 | 67.98 | 0.99 | 0.10 | 69.72 | 0.999 | 0.01 |
2 | 55.42 | 0.976 | 0.46 | 66.49 | 0.991 | 0.20 | 68.97 | 0.999 | 0.023 |
3 | 55.98 | 0.945 | 0.42 | 65.53 | 0.958 | 0.25 | 67.88 | 0.999 | 0.15 |
4 | 53.32 | 0.954 | 0.441 | 65.25 | 0.967 | 0.18 | 66.65 | 0.999 | 0.126 |
5 | 53.99 | 0.975 | 0.53 | 64.98 | 0.988 | 0.114 | 66.43 | 0.998 | 0.17 |
6 | 52.65 | 0.966 | 0.56 | 64.61 | 0.97 | 0.13 | 66.41 | 0.998 | 0.19 |
7 | 51.34 | 0.949 | 0.57 | 63.78 | 0.977 | 0.25 | 65.89 | 0.998 | 0.26 |
8 | 51.10 | 0.952 | 0.59 | 62.01 | 0.964 | 0.26 | 64.21 | 0.997 | 0.28 |
9 | 50.66 | 0.962 | 0.61 | 62.50 | 0.960 | 0.12 | 63.52 | 0.996 | 0.33 |
10 | 50.12 | 0.931 | 0.62 | 61.09 | 0.944 | 0.21 | 62.11 | 0.995 | 0.31 |
Patient number | Active contour | Ostu threshold | Proposed threshold tech |
||||||
---|---|---|---|---|---|---|---|---|---|
Accuracy | Sensitivity | Specificity | Accuracy | Sensitivity | Specificity | Accuracy | Sensitivity | Specificity | |
1 | 70.5 | 72.3 | 78.4 | 77.2 | 76.0 | 80.1 | 98.3 | 98.1 | 89.2 |
2 | 87.5 | 77.3 | 76.6 | 78.9 | 71.5 | 85.4 | 96.3 | 95.7 | 88.2 |
3 | 82.1 | 66.4 | 77.3 | 75.6 | 66.5 | 76.5 | 98.6 | 89.4 | 98.1 |
4 | 69.8 | 81.5 | 69.1 | 65.5 | 76.8 | 77.1 | 89.2 | 88.3 | 96.2 |
5 | 66.5 | 79.9 | 88.2 | 71.3 | 67.6 | 74.2 | 97.5 | 87.4 | 96.4 |
6 | 70.5 | 83.2 | 83.4 | 72.4 | 66.8 | 77.4 | 99.1 | 98.4 | 95.4 |
7 | 73.4 | 88.2 | 87.4 | 75.4 | 78.9 | 76 | 91.7 | 95.6 | 97.9 |
8 | 87.1 | 69.4 | 77.9 | 79.5 | 79.1 | 74.5 | 93.6 | 93.5 | 95.7 |
9 | 65.8 | 71.4 | 76.8 | 86.3 | 74.2 | 77.7 | 95.4 | 94.3 | 91.4 |
10 | 75.9 | 73.5 | 73.0 | 88.4 | 76.3 | 74.5 | 92.2 | 90.1 | 99.1 |
The comparative analysis of existing techniques with our proposed may show the outcome of accuracy of Knee OA segmentation technique is shown in
The percentage of actual positive values that are accurately detected is measured by sensitivity, while the percentage of negative values that are correctly identified is measured by specificity. The proportion of people who got a positive response on this test who truly have the illness is called sensitivity. The sensitivity had increased in our proposed technique when comparing with existing methods like active contour and Ostu thresholding technique is represented in
The sensitivity had increased in our proposed technique when comparing with existing methods like active contour and Ostu thresholding technique, which is shown in
This paper proposes an EHO algorithm based on thresholding and segmentation to improve the efficiency of the CT image. The efficiency can be evaluated by accuracy, sensitivity and specificity. The image characteristics become more significant when the image is enhanced. The CT image enhancement relies on the image normalization followed by the enhancement of the image using histogram correlation. The further segmentation of the image takes place using the thresholding techniques named by EHO assimilated Niblack and Bernsen optimization. The time-efficient utility is needed to be enhanced in this approach for future enhancement.