A Preliminary Study on Automatic Characterization and Classification of Vascular Patterns of Contact Endoscopy Images.

The structure and organization of blood vessels in the vocal fold go through changes during the advancement from healthy to benign and further on to malignant stages. Contact Endoscopy (CE) is an optical technique providing real-time information related to the vascular structure of laryngeal mucosa. However, this technique comes with subjectivity in the interpretation of vascular patterns. In this study, a novel automated approach is proposed for vessel pattern charac-terization and classification of larynx CE + Narrow Band Imaging (NBI) images. This method is mainly based on the computation of indicators related to the level of disorder of vessels. 12 features were extracted from the indicators and were fed into two supervised classifiers. Linear Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) showed an accuracy of 95.76% and 93.92% for vascular patterns and 86.04% and 82.23% for larynx histopathologies classification, respectively. These promising results show that the proposed method can potentially solve the subjectivity issues of CE.