An Overview of Medical Imaging Techniques for Knee Osteoarthritis Disease

Osteoarthritis is the most common form of “Arthritis & Joint disease”. Osteoarthritis (OA) is one of the fundamental causes of older and overweight individual’s sickness. It is the main cause of disability in adults. Mostly this disease occurs in people above 45 years of age, in which women suffer more as compared to men. it is basically damaged the Cartilage, because of which bones rub each other causing intense pain and inflammation. this gets thick and makes spurs at the edges. The knee Osteoarthritis is of 4 grades according to X-ray. The first 2 grade and 3rd grade can be recovered with the help of therapy and medications, while the 4th grade is necessary for knee replacement. The emerging Osteoarthritis management approach involves clinical evaluation & diagnostic imaging techniques. Within this research, we explore descriptively and objectively the various medical imaging methods used to diagnose and identify knee osteoarthritis. We study on the automatically detection of recovery rate of human disease and classify Osteoarthritis in the knee from medical images (like Magnetic Resonance image, CT scan, X-ray) from various medical image classification procedures. This paper provides a study that focuses on the various medical imaging methods used to determine osteoarthritis.

This segment offers a brief overview to our study on the immeasurably significant parts of Image processing (IP), medical imaging, knee bone, knee bone disease, and K&L Grades.
The word digital image processing refers to an image being processed through a processor. A picture is a two-dimensional object that represents a measure of certain characteristics, such as brightness or the color of a scene being viewed. In health care, digital image processing (DIP) is becoming increasingly relevant 1 . Some of the essential research areas of image processing such as computer tomography, medical image processing, and morphological imaging, etc. Therefore, digital image processing is an integral part of medical technology. The image processing can usually be divided into digital image processing (DIP) and medical image processing. The article would focus on how medical images are handled. Medical imaging today plays an important role in several other applications in the healthcare field. Medical Imaging is a method used to establish visualization of the body's internal parts, that is, certain organs as well as tissues mainly for clinical study such as diagnosis or treatment of diseases. Medical imaging has undergone significant developments in clinical medicine. Today, there are many good physiological applications of this capacity to obtain a better understanding of the human organ 1 .
Diagnostic imaging is a tactic for creating a visual representation of areas in the interior of the human body in order to identify in addition monitor medical conditions. This has had significant impacts on public health. Medical imaging requires the use of imaging methods and procedures to capture human body images which may assist in medical diagnosis and treatment. This can also be used to monitor any chronic concerns, which can also assist with recovery plans. Medical imaging is the method of generating the inner graphical images Secret structures of the skin and the bones. It is the technique where for clinical diagnosis and medical intervention, we can show the interior of the body. This is the component of biological imaging that includes radiology that utilizes X-ray, MRI, CT scan, etc. 2 .

Knee osteoarthritis
One of the main causes of mainly in women, old aged people and overweight people is Osteoarthritis (OA). OA is a joint condition, which often affects the cartilage. Cartilage is the defensive connective tissue in a joint that covers the ends of bones. Healthy cartilage allows simple gliding of bone in the joint and avoids friction between them 3 . In OA the upper layer of cartilage breaks down & goes away. This allows one another to rub the bones causing pain. It usually affects the knee, hip, spine, and joint in the legs. Age is the most common cause of osteoarthritis of the knee. Two types of OA, primary OA, are seen in older people due to genetic reasons or aging. Secondary OA appears earlier in life due to some injury, diabetes, obesity, athletics, or arthritis patients.
A sample of the normal and affected osteoarthritis knee image is shown in Figure. 1.1 Diagnostic Features • Clinically, knee OA 's main neurotic qualities include narrowing of the joint field, osteophyte structure, and sclerosis 77,78 . Figure 1.1 Exhibits balanced anatomy of the hip and knee affected by OA: joint space contraction (JSN) due to cartilage deterioration and bone spurs (osteophytes). Knee OA symptoms include anatomical deformities, such as loss of articular ligament, menisci, tendons, synovial tissue, and subchondral bone. • The most important signs of OA include discomfort and difficulty in joint mobility, decreased limitations on operation and involvement, morning or after-rest joint stiffness. OA diagnosis is generally based on clinical assessment, symptoms, and basic techniques of radiographic evaluation (X-rays), MRI, CT, etc. While several other methods have been suggested, a reliable method for classifying individual pairs into 5 grades is the Kellgren-Lawrence (KL) scheme.

Radiographic Classification of Knee OA
Knee OA grows progressively as the years progressed, and advances in stages. Knee OA range is usually split into 5 groups. The main (stage 0) matches to the normal healthy knee, and the last (stage 4) matches to the most severe event. Figure 1.2 indicates the contrasting OA frequency levels in the knees. "The International Knee Documentation Committee (IKDC) scheme, the Ahlback scheme and Kellgren & Lawrence system2 are the most generally utilized strategies for grading knee OA". The other commonly used OA assessment method for the non-radiographic knee is "Western Ontario and McMaster Universities osteoarthritis Index (WOMAC3)", which tests pain, rigidity, and functional limitation.
K&L grades can be used for radiographic Osteoarthritis quantification (Kellgren and Lawrence (1957)). As shown in Table 1.1, details the Grades assigned for different Osteoarthritis severities.

Structure of Human Knee
The human psyche has many joints but the knee joint plays a significant role in all joints as it is the largest joint. The aim of the figure is to find a way of making an important point available between thigh & lower leg during development. It includes bones (the femur, tibia, & patella), which are bound by the hyaline ligament articular surfaces (tibiofemoral and patellofemoral). In two significant areas, the femur cooperates with the tibia: average (closer to the body's midline) & parallel (far from the center). Fig.1.4 shows an enticing knee reverberation diagram from a reference anatomical chartbook into a sagittal plane. You will see the ligament as a superb thin coating that protects the bone 6 .

Medical Image Classification
"Image processing tasks consist of three steps: extraction and representation of objects, classification of images, and interpretation of images" 7 . One of the active areas of computer vision is the detection and classification of osteoarthritis of the knee from MIME. One of the most widely discussed issues in computer vision is the classification of pictures. This field aims to detect all the artifacts of a given picture.
Classification of pictures is the most important obstacle to the tasks of image processing. This task was related to the use of methodologies to manipulate the effects of image analysis, the product of pattern recognition and methods of classification, and then authorizing the resulting image classification into health expert information 8 . The primary goal of classifying MIMEs is indeed to achieve high precision but to classify areas of the patient's psyche that have the disease as well 9 . For better clinical treatment an automated diagnostic technique with image data is required in the future. Since the deal with image classification stays an open field and a significant test that nullifies the result of image classification through clinical master information, this paper focuses only on a thorough analysis of the methods used to classify clinical image mining using the updated approach that addresses this problem.
Therefore, this paper is designed as follows. section II presents in detail different types of medical images, Section III presents different types of medical image classification techniques and description of the convolutional neural network (CNN), Section IV presents the conclusion of the review.

Screening Techniques
Tragically, today's typical OA treatment doesn't cure the disease absolutely. It is then of utmost importance to consider the degeneration of the ligament until it becomes irreversible.
Medical imaging comprises a good understanding of imaging media & object, physical imaging, instrumentations & visual display method. There are several modalities of medical imaging which include ionized radiation, Ultrasound, Magnetic Resonance, Nuclear Medicine, also Optical Methods as a media of modality. Each medium of modality has a specific characteristic in addition the response of modifications to the structure of the human body and the tissue of the organs 10 . Medical imaging devices make use of the patient's received signals to generate images. The extent of ligament degeneration in patients is determined by many methods 11 .

Radiography (X-Ray) Imaging
It is a method of electromagnetic radiation by means of a frequency scope of 0.1-10 nm and changed into photons, 12-125 keV, of energy level. The X-ray imaging technology was used almost simultaneously with the need to use research laboratory work as a health instrument for diagnosis. These are used to take pictures of the bones, usually to see whether and where fractures occur. Dentists and orthodontists also use them to look at teeth. We can also see bone tumors on x-rays 11 .
Across the last hundreds of years, Joint Space Width (JWS) X-rays have become the standard OA screening technique. As they are nonobtrusive, they provide considerable advantages over arthroscopy and can be performed again when needed. "The drawback of Radiology is the lack of sensitivity for brief moment monitoring due to how X-ray images appear to measure ligament changes over 2-3 years" 6 . Advantages of X-ray image • The machine does not surround the complete figure so in people with claustrophobia it will not cause anxiety. • The operation takes just a few minutes to complete. Disadvantages of X-ray image • Some dissimilarity mediators could cause undesirable side effects. • X-rays expose the patient to toxic radiation, which may cause cancer, but the total of radiation provided is small 12 .

Computed Tomography (CT scan)
CT is a method in medical imaging that has had a significant effect on health judgment & evaluations. This is a common multimodal modality Medical fusion picture 13-26 . A Scans is just the same as an X-ray producing crosssectional images within a given area 's body. A CT production from one's knee, for example, will allow specialists to examine illness or to examine wounds on one's knee. The body is circled by a CT scanner, and pictures are forwarded to a Screen. These images are used by the Computer point by point to the point of view. This allows specialists to see the muscles, ligaments, tendons, lungs, & bones that make up one's leg, in specialism and training. Likewise, a CT scan is also called a CAT scan. The output is assessed in a laboratory or in a specified casualty test facility. The key benefits of the CT scan are the comparatively tall processing periods then the high resolution of the images 6 .

Advantages of Computed Tomography
• CT scans are relatively short-they last just about 10-20 minutes.
• Results compared to some other types of scans are extremely fast.
• CT scans are non-invasive, as they are painless.

Disadvantages of Computed Tomography
• As with many scans, your body is exposed to a certain amount of radiation. The more the patient's body is scanned the more they are exposed to radiation. These are also designed to limit the sensitivity to radiation.
• An allergic reaction to the used dye is likely. 12 .

Magnetic Resonance imaging (MRI)
MRI is a mostly modern standard ligament screening technique as it doesn't use radiation therapy, is non-intrusive & reproducible, and offers high contrast and clarity with decent image quality. X-rays transmit images in a proprietary format that can be processed and easily retrieved, and offer a set of requirements for the appropriate supply of images. 6 . The MRI approach uses a strong magnetic field to coordinate the nuclear magnetization of water molecules with hydrogen atoms. The upside of MRI is that it is fit as a fiddle for pregnant ladies and youngsters since it doesn't permit introduction to radiation. What's more, delicate tissue structures are precisely reproduced in organs, for example,   Clinical imaging likewise gives a premise of typical life structures and physiology for the recognizable proof of anomalies. Despite the fact that imaging of separated organs and tissues can be performed for clinical purposes, such methodology is generally viewed as a major aspect of pathology as opposed to clinical imaging. Image classification is a pixel or group of pixels with gray value-based labeling. Classification is one of the most commonly used forms of knowledge extraction. In Classification, multiple features are usually used to collect pixels, i.e., multiple images of a given object are needed. The procedure that will be used for the image classification is as follows: • Image Acquisition: It is the retrieving of an image from a source. • Image Enhancement: It aims to improve the image qualitatively with respect to a specific task. • Feature Extraction: This is the data cleaning process in which the features related to the classification are extracted from the cleaned images.
• Classification: Classification of images is the marking of a pixel or group of pixels based on their gray color.

Texture classification
Texture offers essential features and properties for computer vision & IP difficulties such as investigation of images, classification, & analysis of segmentations. Texture classification requires a result as to the type of texture belongs to the trial image using prior class information and classification algorithms 28 .
It can be useful to any digital image modality & assists in obtaining Image spectral properties. Besides, the description of textural characteristics may be useful for clinical analysis by the radiologist and Partitioning the selected features room by class/category of tissue. A wide range of techniques has been proposed for describing the texture of the image. Approaches of texture analysis can be broken down into four

k-Nearest Neighbor Pattern Classification (KNN)
The KNN was going to be a nonparametric set of rules. First, the set of rules stores the vectors of the Training set function, and then you can find a novel case classification set of k closest training cases in the Function space, then assign the case to the class with more cases in the set 30 . "In Suguna & Thanushkodi's work 31 an upgraded kNN utilizing a genetic set of rules has  Marked narrowing of joint space,Large osteophytes,Severe sclerosis and definite of bone ends been utilized to diminish the high aim complexity with low reliance on the training set and no weight variance b/w each class". "Latifoglu et al., 32 have proposed a clinical technique for the conclusion of atherosclerosis sickness dependent on fundamental component analysis, KNN weight pre-processing, & Artificial Immune recognition System (AIRS)". Nonetheless, this algorithm has some limitations: it can only store highly computationally complex local statistics, and it takes longer to compute new queries. It can deal with binary & continuous attributes yet it doesn't expressly discrete those 33 . By the by, scarcely any ongoing investigations look to determine the restriction of customary kNN, e.g., in the 34,35,31 study, can yield better results.

Support Vector Machine (SVM)
SVM is a theory of statistical learning which analyzes data & recognizes patterns. It's a good thing. Supervised way of studying. "The key training course in arrears SVM is to find the ideal linear hyper-lane, thus reducing the error predictable in classifying unknown samples of the analysis" 28 . SVM 's benefits: Both continuous and binary attributes can be handled; rating speed and accuracy are fine. Yet there are few inconveniences, such as SVM will take longer to train data set and does not manage discrete attributes 36 . "Kharrat et. al 37 recommend a classification technique for mind MRI utilizing an SVM genetic calculation and fit for characterizing cerebrum tissue into an ordinary, generous, or dangerous tumor". In any case, when managing multi-measurements and consistent usefulness SVM will in general improve. It takes a large sample size to reach optimum predictive accuracy.

Convolutional Neural Networks (CNNs)
A CNN is an actual form of ANN, with very few interlay connections, aimed at maintaining spatial data relations. In CNN, the input is settled in a net assembly and then fed through layers which preserve some layers, with respectively layer acting on a specific area of the previous layer ( Figure. 3.2) 38 . However, we encounter three different types of layers in these networks instead of plain perceptron layers (a) Convolutional layers, (b) Pooling layers, (c) Completely connected layers (often called FC-layers). We define these layers   Figure 3.2. The architecture of the convolutional neural network (CNN).

Convolutional layers
This layer is made up of several feature maps, which are gained by kernel convolution with the input signal. "For a two-dimensional (2-D) image of a solitary way, every convolution kernel is a weight matrix that can be a 3 × 3 or 5 × 5 matrix" [39]. An example of the 2-D convolution is shown in Figure. 3.3.
The convolution strategy gives an approach to the process variable size input sources utilizing the convolution kernels, and extraordinary inputs features are separated by convolution activity in the convolution layer. "The main layer draws tests of features of the lower level, for example, sides, corners, and endpoints. At that point, the higher layer takes out increasingly complex and more elevated level capacities by changing the features at the lower level" [39]. In the primary, the Convolution layer has minimal contact characteristics & weight sharing. Feature maps are fed through nonlinear activation functions from a convolutional layer. This permits almost any nonlinear function to be approximated for the whole neural network. "The activation functions are commonly exceptionally basic direct amended units, or ReLUs, characterized as ReLU(z) = max (0, z), or variations, for example, defective ReLUs or parametric ReLU. For more detail on these & other empowering features. Taking care of the feature maps with an activation function delivers new tensors, all the more regularly known as feature maps" 39 .

Polling layer
In general, a pooling layer is then pooled into each feature map made by taking care of the information through at least one convolutionary layer. "Pooling tasks, accepting little network locales as data sources, yield single numbers for every area. Commonly, the number is resolved to utilize the capacity (max-pooling) or mean (average pooling)" 38 . The pooling layers give the CNN an invariance of an interpretation since a slight move in the input picture prompts minor changes in the activation maps.
"One way to get the sampling effect of pooling down is to use convections with the extended phase duration. Evacuating the pooling layers rearranges the system's configuration without performance" 39 . The most widely used layer of max-pooling is given away in Figure. 3.4.

Fully connected layer
The image's feature map is separated after a progression of convolution and pooling layers, and all the neurons in the element map are changed over into a fully connected network. To finish, SoftMax is capable of classifying output by layer. The completely linked layer aims to combine local knowledge with class distinguishing in both the convolution layer and the pooling layer 39 in order to increase the CNN's overall effectiveness.

DISCuSSION
Research on CNNs has progressed rapidly over the years but computing resources and memory limitations frequently constrain the realistic implementation of these models. What has also led to extensive research on ConvNets is the accuracy of difficult classification tasks that involve an understanding of abstract concepts in pictures. Another reason why CNN is so popular is due to its architecture-the best thing is that there is no need for feature extraction. The computer learns to extract characteristics and the central concept of CNN is that it uses image and philtre convolution to establish invariant features that are passed on to the next layer. The next layer results are compared with unique philters to create more invariant and abstract features, and hence the process is repeated until the final feature / output (say, X face) becomes invariant to occlusions.
ConvNets are more effective than machine learning algorithms, and are therefore more accurate in numerical terms. ConvNets are also highly successful in areas involving vast unstructured data, such as image detection, speech recognition, & natural language processing. "We have developed deep learning & ML techniques for analysing knee bone images". Deep learning & ML methods are set out in Table 4.1 below

CONCluSION
The principal motivation behind this review is to advance the state-of-the-art computeraided diagnosis of knee OA severity through the advancement of predictive methods based on deep learning. Classification of health images is an important part of the study; it combines the diagnostic question with the goals of the scientific review. This paper introduced an intensive examination of knee osteoarthritis in clinical imaging, concentrating on automatic detection and methods for human body sickness analysis, in which the imaging modalities utilized for each Dataset and pros & cons. Improvements in image classification techniques for future research will improve the accuracy value and can be used for later computer-assisted diagnosis, and more reliable methods can be developed. 16 (2007).