A novel automated classification technique for diagnosing liver disorders using wavelet and texture features on liver ultrasound images

Abstract

A novel automated classification technique for diagnosing liver disorders is contributed in this paper by utilizing the merits of wavelet and texture features of ultrasound images. In this automated classification technique, initially the diseased part of the ultrasound image is isolated based on the application of improved active contour-based segmentation scheme. Improved active contour-based segmentation is mainly for preventing the issue of worse convergence, which is prevalent in the concave boundary regions of ultrasonic images. After segmentation, shift variant bi-orthogonal wavelet transform is applied for decomposing the region of focus into diagonal, vertical and horizontal component images. This shift variant bi-orthogonal wavelet transform is used in this approach for reducing the degree of prediction errors that are most possible in the classical discrete wavelet transform schemas. Finally, an improved random forest classifier (IRFC) is used for classifying the features that are extracted from the wavelet filtered images using gray level run length matrix (GLRLM). The performance of this scheme is evaluated based on sensitivity, specificity and accuracy metrics and shows the comparison of each classifier performance. The results of the proposed scheme infer an overall classification accuracy rate of 97.8% and confirm better results using GLRLM.