Machine Learning in Laryngoscopy Analysis: A Proof of Concept Observational Study for the Identification of Post-Extubation Ulcerations and Granulomas

Abstract

Objective: Computer-aided analysis of laryngoscopy images has potential to add objectivity to subjective evaluations. Automated classification of biomedical images is extremely challenging due to the precision required and the limited amount of annotated data available for training. Convolutional neural networks (CNNs) have the potential to improve image analysis and have demonstrated good performance in many settings. This study applied machine-learning technologies to laryngoscopy to determine the accuracy of computer recognition of known laryngeal lesions found in patients post-extubation. Methods: This is a proof of concept study that used a convenience sample of transnasal, flexible, distal-chip laryngoscopy images from patients post-extubation in the intensive care unit. After manually annotating images at the pixel-level, we applied a CNN-based method for analysis of granulomas and ulcerations to test potential machine-learning approaches for laryngoscopy analysis. Results: A total of 127 images from 25 patients were manually annotated for presence and shape of these lesions—100 for training, 27 for evaluating the system. There were 193 ulcerations (148 in the training set; 45 in the evaluation set) and 272 granulomas (208 in the training set; 64 in the evaluation set) identified. Time to annotate each image was approximately 3 minutes. Machine-based analysis demonstrated per-pixel sensitivity of 82.0% and 62.8% for granulomas and ulcerations respectively; specificity was 99.0% and 99.6%. Conclusion: This work demonstrates the feasibility of machine learning via CNN-based methods to add objectivity to laryngoscopy analysis, suggesting that CNN may aid in laryngoscopy analysis for other conditions in the future.

Diagnostic Accuracies of Laryngeal Diseases Using a Convolutional Neural Network‐Based Image Classification System

“…To address this problem, computer-assisted diagnosis (CAD) with deep learning technique has been implemented in laryngoscopy. [2][3][4][5][6][7][8] It has been demonstrated that deep neural networks (DNNs) can be applied to laryngoscopic images to discriminate between two diseases (i.e. binary classification).…”

“…binary classification). 2,3,6 Other studies reported outstanding results in discriminating laryngeal neoplasms (benign, precancerous lesions, and cancer) compared to clinicians' visual assessment. 7,8 However, many studies on the laryngeal image classification focused only on specific diseases and are limited in their applicability to real clinical practice for diagnosing various laryngeal disease.…”

Where do we stand in AI for endoscopic image analysis? Deciphering gaps and future directions

“…Here, a semantic segmentation approach was taken, which yielded in dice similarity coefficient of 0.78 ± 0.24 and 0.75 ± 0.26 on retrospective and prospective test sets, respectively. Similarly, for the laryngoscopy 74 , various lesions were annotated in 127 images from 25 patients to train a UNet architecture showing per-pixel sensitivity of 82% and for granulomas and 62.8% for ulceration. Segmentation of recurrent laryngeal nerve, responsible for human speech, during surgery (thyroidectomy) was achieved using the widely known mask R-CNN (instance segmentation) approach 75 .…”

“…AI in other endoscopic procedures. Some other types of endoscopic images-based deep learning applications include (a) detection of nasopharyngeal malignancies 73 , and segmentation of granulomas and ulcerations on images acquired by laryngoscopy 74 , (b) an end-to-end deep learning algorithm to segment and measure laryngeal nerves during thyroidectomy (a surgical procedure) 75 , and (c) deep-learning-based anatomical interpretation of video bronchoscopy images 76 . A recent review and metaanalysis paper on laryngeal endoscopy 77 suggested the AI models presented high overall accuracy between 0.806 and 0.997.…”

Detection of laryngeal carcinoma during endoscopy using artificial intelligence

“…Only in recent years has there been a constant rise in publications of articles investigating the application of AI in laryngoscopic images. Although we did not performed a systematic review of literature, a thorough literature search revealed over 15 publications in the last 6 years 16–32 . A rough summary of these articles showed that all but one research group investigated still images (i.e., photos) of laryngeal lesions in a postprocessing method.…”