Classification of endoscopic images based on texture and neural network

Computerized processing of medical images can ease the search of the representative features in the images. The endoscopic images possess rich information expressed by texture. Regions affected by diseases, such as ulcer or coli, may have different texture features. The texture model implemented in this study is Local Binary Pattern (LBP) and a log-likelihood-ratio, called the G-statistic, is used to evaluate the similarity of regions based on LBP. The neural network is used in the classification. SOM and BP are applied and compared. The texture model and classification algorithm are implemented and tested with clinically obtained colonoscopic data. For large amount of colonoscopic images, proper classification results corresponding with unique medical features can be acquired, which suggests that the unsupervised endoscopic image classification is applicable.