Abstract
Knee osteoarthritis (OA) is a degenerative joint disease that occurs due to wear down of cartilage. Early diagnosis has a pivotal role in providing effective treatment and in attenuating further effects. This chapter aims to grade the severity of knee OA into three classes, namely absence of OA, mild OA, and severe OA, from radiographic images. Pre-processing steps include CLAHE and anisotropic diffusion for contrast enhancement and noise reduction, respectively. Niblack thresholding algorithm is used to segment the cartilage region. GLCM features like contrast, correlation, energy, homogeneity, and cartilage features such as area, medial, and lateral thickness are extracted from the segmented region. These features are fed to random forest classifier to assess the severity of OA. Performance of random forest classifier is compared with ANFIS and Naïve Bayes classifier. The classifiers are trained with 120 images and tested with 45 images. Experimental results show that random forest classifier achieves a higher accuracy of 88.8% compared to ANFIS and Naïve Bayes classifier.
TopMachine intelligence models have received impressive results in various healthcare problems. This is attributed to the availability of data and advancements in algorithms. Several studies have been carried out on the diagnosis of knee osteoarthritis employing the computer-aided methods. Each study applied various segmentation techniques, feature extraction methods and classifiers to diagnose the knee OA. This section describes various researcher’s works and studies of related research problems.
Brahim et al. (2019) applied circular Fourier filtering to retain necessary information related to tibial trabecular bone structure. Independent Component Analysis (ICA) was adopted for feature extraction and the first ten discriminant components were used for classification using Naive Bayes and Random Forest classifier. This method classified radiographic images with an accuracy of 82.98%, a sensitivity of 87.15%, and a specificity of 80.65%.
Thomson et al. (2015) developed an automated grading method by identifying the outlines of bones to standardise the measurement of OA features of the knee. The features derived from both bone shape and image texture in the tibia were given to Random Forest classifiers. The weighted sum of the outputs of two Random Forest classifiers improved the performance. Alternatively, the experimental results proved that Random Forest classifiers trained on simple pixel ratio features are as effective as the texture and shape features.
Anifah et al. (2013) employed Gabor kernel, template matching, row sum graph and gray level center of mass method for segmentation. A classic Self Organizing Map algorithm was trained with Gray Level Co-occurrence Matrix (GLCM) features. The experimental results proved excellent classification accuracy for grade 0, 1 and 4, whereas the grade 2 and 3 were failures.
Wahyuningrum et al. (2016) applied Structural 2-Dimensional Principal Component Analysis (S2DPCA) for feature extraction and Support Vector Machine (SVM) for classification. The maximum average classification accuracy was compared with Gaussian kernel and Polynomial Kernel. The experimental results proved that the hybrid of S2DPCA and SVM could differentiate KL grade 0 from the other grades with accuracy up to 94.33%.
Key Terms in this Chapter
Osteophytes: Abnormal bony outgrowth or projection that form along the joint margins.
Sclerosis: Hardening or stiffening of tissues caused by diseases such as osteoarthritis, diabetes, scleroderma.
Computer-Aided Diagnosis: Systems that assist medical professionals in analysis and interpretation of medical data.
Radiology Informatics: A field of imaging informatics that is concerned to apply information science to radiology.
Arthritis: A disease causing painful inflammation and stiffening of bones.