Automated Diagnostic System for Laryngeal Hemiplegia Using Endoscopic Image

Abstract

Laryngeal hemiplegia is a naturally occurring disease of a horse recognized as a common reason of the poor performance. This syndrome is diagnosed by a human through an endoscopic observation. The asymmetry of arytenoid cartilages is calculated by left to right angle quotients (LRQ). However, human inspection may misdiagnose the disease. Therefore, an automatic diagnostic system is expected to analysis the asymmetric view of laryngeal hemiplegia. In this paper, we have proposed an automatic way to analysis laryngeal hemiplegia by combining image segmentation and parabola fitting algorithms. Our method includes six steps: (1) hierarchical contour map is achieved by a state-of-the-art hierarchical image segmentation method joining Global Probability of Boundary (gPb), Oriented Watershed Transform (OWT), and Ultrametric Contour Map (UCM); (2) key curves are extracted; (3) boundaries are confirmed by Parabola fitting method; (4) dorsal-most point of rima glottidis is selected; (5) proximal-most point is identified and (6) LRQ is calculated as the final measurement. To evaluate the accuracy of our proposed method, a dataset of horse larynx endoscopic images has been built up and tested. Experimental results have shown that the proposed method has good performance. 