Patellofemoral Arthritis Prediction Using Machine Learning

DOI : 10.17577/IJERTCONV11IS05089

Download Full-Text PDF Cite this Publication

Text Only Version

Patellofemoral Arthritis Prediction Using Machine Learning

Manjula P1, Anu M S2, Ashwin S Mathad 3, Bindu G S 4, Sarswathi S 5

Dept. of CSE, JIT, Dvg.


Patellofemoral arthritis is a degenerative joint disorder that affects both the patella (the kneecap) and femur (the thigh bone). There have been hopeful developments in the diagnosis and management of a number of medical conditions, including arthritis. Patellofemoral arthritis is a prevalent disorder that can cause pain and impairment in patients. One of the diagnostic techniques for detecting patellofemoral arthritis is X-ray imaging. Machine learning techniques have shown great promise for analysing medical images and assisting in the diagnosis and treatment of various ailments. Machine learning algorithms may be trained to analyse X-ray pictures and identify the existence of patellofemoral arthritis based on particular radiological markers. Machine learning has the potential to enhance diagnostic precision, individualise treatment regimens, and ultimately enhance patient outcomes in the processing of X-ray images for patellofemoral arthritis. To properly evaluate the effectiveness and generalization of machine learning algorithms in this situation.

Keywords: Patellofemoral arthritis, X-ray Deep learning, Texture analysis.


Patellofemoral The most prevalent kind of arthritis, which affects millions of people, causes discomfort, movement restrictions, reduces independence, and lowers quality of life. Arthritis is a degenerative condition of the knee joint. Although arthritis has no known treatment, there are a number of environmental, natural, and medicinal risk factors that can be changed and those that cannot that are known to contribute to the onset and progression of the condition. The predetermined variables that characterize arthritis are high-dimensional, diverse, and the few basic logistic retrogression Numerous risk variables and, more importantly, any connections between environmental, other medical, and ecological elements cannot be handled by models. Additionally, they are unable to recognize a healthy subject's propensity to display symptoms of the complaint and its development based on patient-related problems. Despite this, it is important to recognize the strength and importance of a sound research design. "Simple" analysis can yield reliable results in a well-designed research. Due of the serious modelling and computational errors in the arthritis risk vaticination models, the problem must be approached entirely differently. combining sophisticated bracket models with innovative and effective machine literacy approaches such as fuzzy-sense propositions and demarcation criteria (such as collective information gain indicators and Fisher demarcation rates). The issue of excessive dimensionality might be

considerably exacerbated by point selection techniques designed for genuinely enormous data sets in statistical ways applied to the arthritis trouble prophecy problem Machine learning (ML) is the study of how computer algorithmsi.e., machinescan "learn" complicated relationships or patterns from empirical data and then create(fine) models connecting a sizable number of variables to a target variable of interest. Because it can analyze complex circumstances with a significant amount of data and generate the best results, machine learning (ML) is, as was already said, a precious tool in the battle against arthritis. Robotics, medicine, biochemistry, bioinformatics, meteorology, animal husbandry, and the lucrative lores are a few notable domains where machine learning has been used.

Figure 1: An instance of X-ray images of the normal knee and severe Arthritis.


Patellofemoral Arthritis literature assessment, the team searched for and examined a variety of patents, research papers, documents, newspapers, and magazine articles from various scenes. Osteoarthritis severity might be determined using a self-organizing map based.

Anifah et al. [1]. The first stage of this research, which is separated into three parts, is image processing utilising the


Gabor kernel. The learning phase is the second step, while testing is the third. Normalising the picture dimensions to a template of 50 to 200 images is the first step in image processing. The starting learning rate is set to 0.5, and there are a total of 1000 iterations in the learning stage. The weights created during the learning stage are used during the testing step. The conclusion of this study is that KL-Grade may be accurately determined on X-ray pictures of knee osteoarthritis with 94.8% accuracy by medical professionals.

Hegadi et al., [2] A technique to recognise handwritten Marathi digits using a multilayer feed-forward neural network was proposed. Pre-processing is used to remove noise from the scanned picture of the document, and care is made to connect any broken letters. Cubic interpolation is used to split each number from the text and scale it to 7 5 pixels. A method is utilised to better represent each pixel in segmented numerals while scaling. Before being fed to the neural network, this enlarged number is transformed into a 35-value vector. For this experiment, we employed 100 sets of 1000 digits each, of which 50 sets were used to train the network and 50 sets to test it.

K.Sun , Tang, et al., [3] proposed a novel explicit image filter called guided filter. Derived from a local linear model, the guided filter computes the filtering output by considering the content of a guidance image, which can be the input image itself or another different image. The guided filter can be used as an edge-preserving smoothing operator like the popular bilateral filter, but it has better behaviors near edges. The guided filter is also a more generic concept beyond smoothing.

Mishra, M., Srivastava, et al., [4] A theory of artificial neural networks was put out.This essay discusses artificial neural networks (ANNs), its numerous properties, and their business applications. This essay also explains what neural networks are and why they are so crucial to the field of artificial intelligence today. due to the development of intelligent systems, some of which were influenced by biological neural networks.

Chen et al., [5] proposed Deep convolutional neural networks were suggested by him for the automated identification and categorization of the severity of knee osteoarthritis. The automated detection and evaluation of the severity of patellofemoral arthritis from X-ray images using deep learning is presented in this research. When it came to identifying and categorising the severity of knee osteoarthritis, the suggested strategy had good accuracy.

SS Hu et al., [6] proposed utilising biomechanical characteristics and machine learning techniques to predict the course of osteoarthritis of the knee. This study suggests utilising gait analysis-derived biomechanical parameters to

predict the development of knee osteoarthritis. High accuracy was attained in forecasting the course of knee osteoarthritis using the suggested method.

H.Wang et al.,[7] proposed a Quantifying the severity of knee osteoarthritis using deep learning-based analysis of knee radiographs was the idea put up by them.This study describes a deep learning-based method for evaluating the patellofemoral arthritis using radiographs of the knee. The suggested method has excellent accuracy when assessing the degree of osteoarthritis in the knee.

N. Ghavami et al. [8] proposed an Using clinical and radiographic data, presented an automated categorizaton of Patellofemoral Arthritis severity. The automated classification of the severity of patellofemoral arthritis using clinical and radiographic data is presented in this research as a machine learning-based method. The suggested method classified the severity of knee osteoarthritis with good accuracy.

X. Wang et al., [9] proposed a Wearable sensors were suggested by to be used in a joint motion analysis for the early diagnosis of knee osteoarthritis. In this study, a wearable sensor-based method for the early identification of osteoarthritis of the knee utilising joint motion analysis is proposed. The suggested method has excellent early osteoarthritis knee detection accuracy.



    A set of medical images aimed at identifying Patellofemoral arthritis is the knee orthopedic detection dataset.

    There are five classes in the dataset: Dubious, Solid, Negligible, Moderate, and Extreme. Each picture is named with the seriousness of the infection, taking into consideration the improvement of a grouping model. The dataset contains a total of 8260 images, of which 1656 are from the test set, 5778 are from the preparation set, and 826 are from the approval set. The photos are stored in JPEG files and have a resolution of

    224 by 224 pixels.The dataset is intended for use with supervised machine learning, in which the model is trained on labeled data to identify patterns and make predictions about data that has not been seen.

    The images are pre-processed to prepare the data for machine learning, the photos are pre-processed using the Image Data Generator class in Keras. By applying erratic modifications to the images, such as shear range, zoom reach, and flat turning, the Image Data Generator class is used to extend the preparation information. In order to verify that the pixel values lie within the range of 0-1, the Image Data Generator also


    rescales the images by a factor of 1./255. In general, the dataset for detecting patellofemoral arthritis in the knee is a helpful tool for academics and developers.The large and diverse dataset allows for the development of trustworthy and precise classification models.

    Creating a Dataset

    The X-ray images of the knee that are going to be classified are divided into a training dataset and a testing dataset, with the test size ranging from 30 to 20 percent.


    Resizing an Image: By setting the image height and image width variables to 224, the code resizes all of the images in the dataset to 224×224 pixels. In order to train a CNN, it is essential to resize the images to a standard size so that they all have the same dimensions.

    Information increase: Shearing, zooming, and flat flipping are only a couple of the picture expansion techniques that the train_data object, which was made utilizing the Image Data Generator class, applies to the preparation information. Data augmentation increases the size and variety of the training set, which can improve model performance and prevent overfitting.

    Encoding one-time: The target labels become one-hot encoded vectors when the class mode parameter of the flow from directory method is set to "categorical."

    Each class in one-hot encoding receives its own vector with a value of 1 at the class index and 0s elsewhere. Classification tasks frequently employ this approach.

    Normalization: By guaranteeing that the pixel values are on a comparable scale, standardization is important to upgrade the intermingling of the improvement calculation during preparing. In any case, apparently the code you gave does exclude standardization.

    Data Splitting: The code divides the dataset into three sets: the training, validation, and testing sets. The validation set is used to modify the model's hyperparameters and avoid overfitting while the training set is used to train the model. The testing set's goal is to evaluate how well the model performs when used with fresh, untested data.


Figure 2: The activity diagram.

Figure 3: The use case diagram


Figure 4: Sequence Diagram


Convolutional Neural Networks (CNN's) can be used in patellofemoral arthritis research for various tasks such as medical image analysis, diagnosis, and prediction of disease progression. Here are a few examples of how CNN's have been used in patellofemoral arthritis.

Figure 1: Architecture of CNN Algorithm

Patellofemoral Arthritis diagnosis: CNN's have been used to classify patellofemoral arthritis from radio graphic images, where the model learns to identify specific features or patterns that are associated with arthritis. This can aid in early detection and diagnosis of the disease.

Cartilage segmentation: CNN's have been used to segment cartilage from MRI images, which can help in assessing cartilage thickness and volume, as well as detecting cartilage lesions and defects.

Knee joint alignment: CNN's have been used to measure knee joint alignment from radiographic images, which can provide information about the progression of patellofemoral arthritis and help in the development of personalized treatment plans.

Prediction of disease progression: CNN's have been used to predict the progression of patellofemoral arthritis from radio graphic images or patient data. This can help in identifying patients who are at high risk of disease progression and can benefit from early intervention or treatment.


Image showing the Patellofemoral Arthritis login

and registration page allows people to create an account or log in to an existing account

Figure 5: Creating an Account


Registration of Users:

The user have to establish a user registration interface that allows users to establish accounts by entering personal data such their name, email address, and password..Image showing the Patellofemoral Arthritis registration page that allows people to create an account. The registration page may have fields for username, email address, username and password.


Figure 6: Registration Page

The patellofemoral arthritis login page allows people with an existing account to log in and access patellofemoral arthritis information and resources.

Figure 7: Login Page

A dashboard for downloading patellofemoral arthritis reports would allow people to securely upload x-rays and save reports. Filling the form with the register name , uploading of x-ray image ,gender , age etc, with completing filling the details for report generation.

Figure 8: Dashboard Page Dashboard with uploaded Patellofemoral Arthritis

X-rays and option to view Patellofemoral Arthritis reports.

Figure 9: Report Page.


This page creates a report that provides an in-depth picture of a person's arthritis condition based on X-ray images that healthcare providers can use to create or modify treatment plans.

Figure 10: Report Generated.

This page allows users to view and download generated reports

Figure 11: View of Reports.


This study examined a number of novel techniques for automatically calculating the severity of knee using CNN's. Finding the knee joint location is the initial stage in the procedure. As an alternative to template matching, we suggest training a linear SVM on horizontal picture gradients, which is both more accurate and quicker than template matching. Our original method for categorizing the severity of knee utilized features taken from CNN's that had already been trained. It was discovered that the characteristics from the pooling and

convolutional layers were more precise than the fully connected layers. Accuracy of multi-class categorization was further enhanced by fine-tuning the networks by swapping out the top fully connected layer. Their techniques have previously been evaluated using binary and multi-class classification metrics. We suggest that treating KL grades as a continuous variable and measuring accuracy using mean squared error are preferable approaches. By employing regression loss during model training, mistakes are penalized proportionately to their severity, leading to more accurate predictions. The wonderful thing about this method is that the forecasts might vary by grade, which fits with the idea of a continuous illness development.


The goal of future research is to increase the accuracy of knee joint detection utilizing a CNN or region-based CNN rather than the suggested linear model on Sobel gradients, as well as to improve the evaluation of knee severity. It is obvious that there are significant differences between the distribution of pictures in ImageNet and knee radiographs. On the knee OA images, it would be possible to train a model from scratch, which would probably be better suited to the domain given a large number of training examples. Semi-supervised methods like ladder networks may be more successful than the domain adaption method employed here if there aren't many instances with labels. Currently, there are separate steps for the detection of knee joints, feature extraction, and classification and regression. Future research will also look at an integrated end-to-end deep learning system.



[1]"Automated Knee Osteoarthritis Diagnosis from Plain Radiographs: A Deep Learning-Based Approach" by M. J. B. Haar et al. (2019)

[2]"Automated Detection of Knee Osteoarthritis Severity Using Convolutional Neural Networks" by A. Anthimopoulos et al. (2020)

[3] "A Systematic Review of Machine Learning Techniques for Knee Osteoarthritis Research" by K. Adhikari et al. (2019)

[4] "Deep Learning for Knee Osteoarthritis Diagnosis Using X- ray Images" by Y. Chen et al. (2019)

[5] "Predicting Progression of Knee Osteoarthritis Using MRI- Based Deep Learning" by L. Zhang et al. (2019)

[6] "Radiographic Knee Osteoarthritis Detection Using Deep Convolutional Neural Networks: A Feasibility Study" by S. P. Yap et al. (2018)


[7]"Detecting Osteoarthritis from X-Ray Images using Convolutional Neural Networks" by P. Rajpurkar et al. (2018)

[8]"Automated Classification of Radiographic Knee Osteoarthritis Severity using Deep Convolutional Neural Networks" by A. Anthimopoulos et al. (2018)

[9]"Machine Learning for Early Detection of Knee Osteoarthritis: A Systematic Review" by A. Bechar et al. (2021)

[10]"Predicting Knee Osteoarthritis Incidence Using a Combination of Machine Learning Methods and Clinical Features" by S. Schwen et al. (2020)

[11] "Automatic Detection and Grading of Knee Osteoarthritis on Magnetic Resonance Images" by Z. Zhang et al. (2020)