Use of Artificial Intelligence (AI) for the intraoperative evaluation of vocal fold leukoplakias

Introduction Assessing vocal fold leukoplakia can be challenging despite modern endoscopic methods. The characterization of the morphology of adjacent vocal fold vessels is of great importance but depends heavily on the clinical experience of the observer. Intraoperative contact endoscopy with Narrow Band Imaging (NBI-CE) enables optimized visualization of vascular changes while the data generated can well be used for an automated evaluation using Artificial Intelligence (AI) methods.

Methods In the present study, the adjacent vessels of 40 vocal cord leukoplakias were recorded intraoperatively using NBI-CE. The generated data was evaluated using machine learning methods with the classification scenarios Support Vector Machine with Polynomial Kernel (SVM) and k-Nearest Neighbor (kNN). After the histology was obtained, the sensitivity, specificity and accuracy of both classifiers were calculated in the classification between benign and malignant findings.

Results In total, 1998 contact endoscopy images were evaluated in 16 benign and 24 malignant leukoplakias. The vascular changes could be mathematically characterized by the algorithms as an increase in the disorder of the gradient vector and the level of curvature. The sensitivity, specificity and accuracy of the automated classification were 100 % , 77.2 %  and 90.6 %  for the SVM and 100 % , 79.8 %  and 91.7 %  for the kNN.

Conclusion The use of methods of AI and machine learning allows an automated evaluation of the vascular changes in vocal cord leukoplakia. The algorithms used can support doctors in the clinical characterization of leukoplakia as potentially benign or malignant.

Poster-PDF A-1534.pdf

Publikationsverlauf

Artikel online veröffentlicht:
13. Mai 2021