Application of principal component analysis and a fuzzy C-means clustering algorithm to wear debris morphology classification

Abstract

Abstract

The application of principal component analysis (PCA) and fuzzy C-means clustering algorithm to the classification of ultrahigh molecular weight polyethylene (UHMWPE) wear debris from artificial joints has been described in this article. Wear particles were extracted and isolated from peri-prosthetic tissues collected during revision surgery, which was revised for loosening. The implant life of the hip prosthesis was 12 years. The particles were examined by scanning electron microscopy. Digitized particle images were analysed on a computer by specially developed software ‘Image-Pro Plus’. The following 19 numerical descriptors were used to characterize the particles: particle area, length, width, perimeter, boundary fractal dimension, and shape parameters such as form factor, roundness, convexity, aspect ratio, and others. PCA algorithm was applied to reduce the amount of parameters to simplify the following calculation. Furthermore, main factors and important parameters such as mean diameter, equivalent circle diameter, and perimeter were found out by PCA. However, C-means clustering algorithm was applied to classify the UHMWPE wear debris into 4–7 clusters. The Xie—Beni index was introduced to determine the optimal number of clusters and illuminate the clustering validity. The result of the calculation indicates that five clusters is the optimal clustering number. The feature of the debris in each cluster is also described in this article.