In this chapter, a machine learning-based knee Osteoarthritis (OA) detection system from magnetic resonance (MR) images is introduced. This system is capable of detecting the presence of OA, which is one of the most prevalent conditions resulting in disability, particularly in the elderly population. OA is the most common articular disease of the developed world and is a leading cause of chronic disability, mainly as a consequence of knee OA and/or hip OA. Medical images such as MR images are widely used for OA diagnosis. A medical specialist analyzes medical images by measuring the changes, particularly in the compartment of the tibio-femoral cartilage. The proposed method consists mainly of a data processing module and a binary classification module that process the 3-D data from MR images. In this study, a novel knee OA diagnostic approach is presented that can identify the condition using magnetic resonance MR images using the Support Vector Machine (SVM) algorithm. The suggested method is based on using 3-D data from MR scans of an actual cohort and the Independent Component Analysis (ICA) technique. The experimental results showed that the ICA-SVM machine learning model achieved 86% of testing accuracy with both 72% of specificity and 100% of sensitivity, once trained with a small MR image dataset. Furthermore, a benchmark evaluation was performed. The results suggest that using a larger and more diverse dataset could ensure the robustness of the proposed method. In future works, the complementary use of ICA components from MR images and a convolutional neural network (CNN) will be studied to try to achieve better predictive rates in supervised learning using a larger dataset. Overall, the proposed method has the potential to revolutionize the diagnosis of knee OA and improve the quality of life of millions of people suffering from this debilitating condition.
Author(s) Details:
Marco Oyarzo Huichaqueo,
School of Engineering, Rovira i Virgili University, 43007, Tarragona, Spain.
To Read the Complete Chapter See Here