MUSCULOSKELETAL DISORDER DETECTION OF WRIST USING VOTING-BASED CLASSIFICATION SYSTEM

According to World Health Organization report, almost one out of the five people including children are suffering from musculoskeletal disorders. It is the second biggest cause of disability. Musculoskeletal disorder affects the major area of musculoskeletal system i.e. shoulder, forearm, wrist, and

Project Title

MUSCULOSKELETAL DISORDER DETECTION OF WRIST USING VOTING-BASED CLASSIFICATION SYSTEM

Project Area of Specialization

Artificial Intelligence

Project Summary

According to World Health Organization report, almost one out of the five people including children are suffering from musculoskeletal disorders. It is the second biggest cause of disability. Musculoskeletal disorder affects the major area of musculoskeletal system i.e. shoulder, forearm, wrist, and legs which causes severe pain, joint noises, and disability. To detect the abnormality, radiologists analyze different XRAY taken from the different angles of patient’s body. The automatic detection of abnormality in musculoskeletal system is a challenging issue. In past, many researchers detected the abnormality in musculoskeletal system from radiographs images by using several deep learning techniques such as capsule network, 169-layer convolutional neural network and group normalized convolutional neural network. According to our research, none of the researchers worked on the proposal of a voting base system that includes multiple CNN. Furthermore, there is no web application available to detect the abnormality of musculoskeletal system of wrist or any other joint. To overcome these gaps, we will develop an architecture that consist of three CNN model to detect abnormality in radiographs of wrist, then the decision will be made on the bases of votes of each model. We will also develop a web application to predict the abnormalities by taking different angles of wrist radiographs by users.

Project Objectives

There are some limitations in the traditional radiology system that is,

1. The Limited number of radiologists available across the country.

2. Disability ratio increasing continuously due to late detection of disorders.

3. The quality can be compromised due to insufficient training and skills [9].

4. The radiologist workload increases which may affect the quality.

To overcome these issues, we will develop an application using deep learning and image processing that automatically detects the musculoskeletal disorders of the wrist. Our application will take the radiographs images of different angles of the patient input and classify the image as either normal or abnormal. Our application will assist the radiologist in disorder detection of the wrist and reduce the workload.

Project Implementation Method

. The application will use the proposed voting-based model that will consist of three convolutional neural networks. To perform accurate detection of high-level features in the initial layers of all three networks, initial layers of all three networks will be frozen whereas middle and last layers will be trained using wrist images in the MURA dataset to learn low-level features from it. As these networks are trained on natural images, it might be possible that transfer learning may be used after re-training those networks on thousands of X-Ray images as their high-level features will be more relevant compared to the natural images. The output of all three networks will be used for voting to get predictions with good accuracy.

Benefits of the Project

In Pakistan, musculoskeletal abnormality of the wrist is detected by radiologists. Radiologists examine the different angles of radiographic images of patients to detect the abnormality. The whole procedure carried out manually which requires a lot of time and increases the radiologist’s workload.  There is not any automated tool or system which can reduce the workload. So, there is a need for an automated system that can assist the radiologist in determining the abnormality in the musculoskeletal system that is the goal of our project.

To achieve our goal, we divided our project into five phases. In the first phase, we will perform a literature review and understand the available dataset. In the second phase, we will study and implement the CNN models using MATLAB. In the third phase, we will develop the web-based application and integrate the CNN model with the application using PHP language. The next phase will be the deployment of the application PHP server and then we will document our findings

Technical Details of Final Deliverable

To study and examine the pre-trained networks such as VGG16, VGG19, Alexnet, squeezenet, and shufflenet.

2.  Enhancing the quality of X-Ray images to improve the training as well as removing the unwanted text tag.

3. To apply dataset balancing techniques such as augmentation etc. to improve the accuracy.

4. Application of transfer learning for the better extraction of high-level features.

5. Voting based model that is based on the voting of the three convolutional neural networks that will be selected based on an examination of all the available options.

 So, the performance of our purposed model will be better or at least equal to the highly performed existing model.   

Software: Basically, we will develop a system in MATLAB using three trained CNN models that will be taken from considered options of VGG-19, VGG-16, squeezenet, shufflenet and Alex net. For the development and deployment of web-based application, we will use php language and php server.

Final Deliverable of the Project

Software System

Core Industry

Medical

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com