Artificial Intelligence in laryngeal endoscopy: Systematic Review and Meta-Analysis

Abstract

AbstractBackgroundEarly and proper diagnosis of laryngeal lesions is necessary to begin treatment of the patient as soon as possible with the possibility of preserve organ functions. Imaging examinations are oft aided by artificial intelligence (AI) to improve quality and facilitate appropriate diagnosis. The aim of the study is to investigate of the diagnostic utility of AI in laryngeal endoscopy.MethodsFive electronic databases (PubMed, Embase, Cochrane, Scopus, Web of Science) were searched for studies published before October 15, 2021 implementing artificial intelligence (AI) enhanced models assessing images of laryngeal lesions taken during laryngeal endoscopy. Outcomes were analyzed in terms of accuracy, sensitivity and specificity.ResultsAll 13 included studies presented overall low risk of bias. The overall accuracy of AI models was very high (from 0.806 to 0.997) and the number of images used to build and evaluate the models ranged from 120 to 24,667. The accuracy was significantly higher in studies using larger database. The pooled sensitivity and specificity for identification of healthy laryngeal tissue (8 studies) was 0.91 (95% CI: 0.83-0.98) and 0.97 (95% CI: 0.96-0.99), respectively. The same values for differentiation between benign and malignant lesions (7 studies) were 0.91 (95% CI: 0.86-0.96) and 0.95 (95% CI: 0.90-0.99), respectively. The analysis was extended to a comparison of sensitivity and specificity of AI models assessing Narrow Band Imaging (3 studies) and white light endoscopy images (4 studies). The results were similar for both methods, no subgroup effect was revealed (p = 0.406 for sensitivity and p = 0.817 for specificity).ConclusionsIn assessing images of laryngeal lesions, AI demonstrates extraordinarily high accuracy, sensitivity, and specificity. AI enhanced diagnostic tools should be introduced into everyday clinical work. The performance of AI diagnoses increases efficacy with the size of the image database when using similar standards for evaluating images. The multicentre cooperation should concentrate on creation of huge database of laryngeal lesions images and implement their sharing, which allows building AI modes with the best performance, based on vast amount of images for learning and testing.