Testing of a novel approach for an automated classification of compact endoscopic vascular patterns in laryngeal lesions

Introduction:

The combination of contact endoscopy and narrow band imaging (compact endoscopy) is suitable for the examination of laryngeal vascular patterns and provides information on the dignity of the lesions. However, the evaluation of the shape of the vessel is partly subjective and dependent on experience. On the other hand, vascular changes can be characterized mathematically as an increase in the disorder of the gradient vector and the curvature of the blood vessels.

Methods:

We have tested a novel approach to the automated classification of compact endoscopic vessel patterns using image and signal processing techniques. Videoendoscopic data from 22 patients were evaluated. First, the automated classification of the studied samples was tested in three groups: ordered patterns, disordered patterns, and patterns with a high degree of disorder. Furthermore, the allocation into four classification scenarios was tested on the basis of histological diagnoses.

Results:

A total of 907 compact endoscopic images were evaluated. Of these, 40% were for training and 60% for testing. Sensitivity was 94%, specificity 97% and accuracy 94% for the automated classification of vascular patterns in one of the three groups. The classification according to histological diagnoses was achieved with a sensitivity of 85%, a specificity of 94% and an accuracy of 84%.

Conclusions:

It was shown that the automated classification of compact endoscopic vascular patterns is feasible. The algorithm can assist physicians in clinical decisions and can be used in diagnostics and tumor follow-up of laryngeal dysplasia and carcinoma.

Publication History

Publication Date:
23 April 2019 (online)