Firefly algorithm for training the radial basis function network in ultrasonic supraspinatus image classification

Chih-Feng Chao, Ming-Huwi Horng

COMPUTER MODELLING & NEW TECHNOLOGIES 2014 18(3) 77-83

Department of Computer Science and Information Engineering, National PingTung Institute of Commerce, Pingtung, Taiwan

The physicians observed the echo-texture and the shape changes of supraspinatus to decide the severity of rotator cuff disease in the clinical standard ultrasound examination. It is not reliable because the accuracy of visual observation depends on the experience of physicians. This article proposes a new algorithm called Firefly RBF network to training the radial basis function neural network by applying the firefly algorithm for classifying the different supraspinatus disease groups that are normal, tendon inflammation, calcific tendonitis and tendon tears of the ultrasound supraspinatus images based on the texture analysis technology. The texture features are generated from four methods those are the grey-level co-occurrence matrix, the texture spectrum, the fractal dimension and the texture feature coding method to analyse the tissue characteristic of supraspinatus. The F-score measurement are used to select powerful features those are generated from the four texture analysis methods for comparison in the training stage, meanwhile, the proposed Firefly RBF network is used to discriminate test images into one of the four disease groups in the classification stage. Experimental results showed that the percentage of correct classification was more than 93.7% that is superior to other methods in the classification of ultrasonic supraspinatus images.