Classification of laryngeal disorders based on shape and vascular patterns of vocal folds

Vocal folds disorders such as vocal nodules, vocal polyps and laryngitis may cause hoarseness, breathing and swallowing difficulties due to vocal fold malfunction. Diagnosis of vocal folds disorders is based on examining structural defect of the vocal folds and vocal fold vibrations by medical imaging devices such as videolaryngostroboscopy and endoscopic high-speed cameras. However clinical diagnosis of vocal folds disorders is prone to error because of their visual similarity. In such cases, surgical procedure may help the physicians make a more accurate diagnosis. Recent medical researchers reported that vocal fold anomalies affect the anatomic structure of blood vessels. Researchers noted that since the blood vessels patterns vary with the types of vocal fold disorders, they can be used as a diagnostic tool. However there is no reported study in literature that evaluates the vessel structures on vocal folds by image processing and pattern recognition methods. In this paper, we design and implement a novel system that can aid diagnosis of laryngeal disorders exploiting alignment of blood vessels and structural defects of vocal folds on color vocal fold images acquired by direct-micro laryngoscopy and videolaryngostroboscopy. We first detected the vocal folds on video-laryngostroboscopy images by using Histogram of Oriented Gradients (HOG) Descriptors. Then glottal area is segmented by region growing algorithm. Evaluation of vocal fold edge features is done by analyzing the vocal fold edge contour to measure the location, shape and size of the vocal fold pathologies. To extract the vessel features we first perform initial vessel segmentation by a region growing process. Then vessel centerlines are obtained by executing false vessel elimination and vessel linking steps. At the last step we categorized vocal folds by using vascular features and shape features of vocal folds. In this study two different classification tasks namely, cyst and nodule classification and healthy, nodule, polyp, sulcus vocalis and laryngitis classification are discussed. Vocal fold vessel features are evaluated for cyst and nodule classification. We used Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Random Forest (RF) classifiers and evaluated their results. We achieved an average success rate of 75% by using SVM. We exploit vascular structure and shape features of vocal folds for healthy, nodule, polyp, sulcus vocalis and laryngitis classification. We propose binary decision tree architecture for classification and achieved an average success rate of %79.