Chapter 9 - Hearing loss classification via stationary wavelet entropy and cat swarm optimization

Abstract

Aim: Pathological images in medicine should be diagnosed in time, and the disease should be accurately and efficiently classified by computer vision. In this chapter, we are devoted to the classification of hearing loss images. Materials: We selected 180 cases and obtained their nuclear magnetic resonance images as the experimental dataset. This group included 60 left-sided hearing loss (LHL) patients, 60 right-sided hearing loss (RHL) patients, and 60 healthy controls (HC). Method: In order to make the process of image extraction more robust, stationary wavelet entropy is selected to extract image features, and the level is selected as two. As a classical classifier, the single-hidden-layer neural network has seven nodes in the input layer. Finally, we choose cat swarm optimization to update the classifier to improve accuracy. Results: Our 10-fold cross-validation results show that the overall accuracy is 90.22±0.95%.The sensitivities are 91.00±1.96%, 89.00±3.87%, and 90.67±2.38%, respectively. Conclusion: The proposed method is superior to the four state-of-the-art approaches (HMI, WE-GA, TS-PSO, WE-CSO) in terms of overall accuracy and sensitivity.