Automatic Detecting Cornea Fungi Based on Texture Analysis

Fungal keratitis has a high rate of blindness and can cause big harm to health. In recent years the incidence of it is rising. So the accurate diagnosis of fungal keratitis is important, but traditional diagnostic methods have some deficiencies. This paper proposes a new cornea fungi automatic detection method based on texture analysis and support vector machine (SVM). The detection of hyphae is a typical evidence of the infection of fungal keratitis, so in this method we detect cornea fungi by hyphae. First, we enhance the confocal microscopic images of cornea to separate hyphae and nerve from the background visually. After that, we extract the texture feature based on local binary patterns (LBP). Then we train the SVM classifier using the extracted feature and labels. Finally testing the accuracy of this method, and dyeing hyphae and nerve with different colors to distinguish the images of fungal keratitis from healthy cornea. The accuracy of this method is 93.5294%. Eventually, we make it able to separate images of fungal keratitis from healthy cornea, thus this method can assist doctors to make further diagnosis whether patients are infected with fungal keratitis.