Non-destructive diagnosis of knee osteoarthritis based on sparse coding of MRI

Abstract

The disability rate of knee osteoarthritis (KOA) is high. A kind of non-destructive diagnosis of KOA based on sparse coding of magnetic resonance imaging (MRI) is presented. The two-dimensional Gabor filter bank is used to extract the high-dimensional features of KOA-MRI images. Secondly, a fitness feedback particle swarm optimisation is proposed to choose three key parameters of the Gabor filter: bandwidth parameters, maximum frequency of the centre and window size. Then the extracted Gabor visual features are described by the sparse coding and sparse coefficient matrix of magnetic resonance images. An improved feature imbalance support vector machine (SVM) is used to classify magnetic resonance images by considering the unbalanced influence of feature contributions. The overall performance of diagnosis has improved.