Deep Learning Applied to White Light and Narrow Band Imaging Videolaryngoscopy: Toward Real‐Time Laryngeal Cancer Detection

Abstract

Objectives: To assess a new application of artificial intelligence for real-time detection of laryngeal squamous cell carcinoma (LSCC) in both white light (WL) and narrow-band imaging (NBI) videolaryngoscopies based on the You-Only-Look-Once (YOLO) deep learning convolutional neural network (CNN).Study Design: Experimental study with retrospective data. Methods: Recorded videos of LSCC were retrospectively collected from in-office transnasal videoendoscopies and intraoperative rigid endoscopies. LSCC videoframes were extracted for training, validation, and testing of various YOLO models. Different techniques were used to enhance the image analysis: contrast limited adaptive histogram equalization, data augmentation techniques, and test time augmentation (TTA). The best-performing model was used to assess the automatic detection of LSCC in six videolaryngoscopies.Results: Two hundred and nineteen patients were retrospectively enrolled. A total of 624 LSCC videoframes were extracted. The YOLO models were trained after random distribution of images into a training set (82.6%), validation set (8.2%), and testing set (9.2%). Among the various models, the ensemble algorithm (YOLOv5s with YOLOv5m-TTA) achieved the best LSCC detection results, with performance metrics in par with the results reported by other state-of-the-art detection models: 0.66 Precision (positive predicted value), 0.62 Recall (sensitivity), and 0.63 mean Average Precision at 0.5 intersection over union. Tests on the six videolaryngoscopies demonstrated an average computation time per videoframe of 0.026 seconds. Three demonstration videos are provided.Conclusion: This study identified a suitable CNN model for LSCC detection in WL and NBI videolaryngoscopies. Detection performances are highly promising. The limited complexity and quick computational times for LSCC detection make this model ideal for real-time processing.

Detection of Farmland Obstacles Based on an Improved YOLOv5s Algorithm by Using CIoU and Anchor Box Scale Clustering

“…In the literature [16], the improved CBF (Counting Bloom Filter) module based on CBF and a Specter module replacing CSP module were proposed to realize the lightness of YOLOv5s model, and improve the detection speed and accuracy of picking Zanthoxylum fruit. In terms of medicine [17], the LSCC (Laryngeal Squamous Cell Carcinoma) using the integrated algorithm (YOLOv5s + Yolov5M-TTA) had the best results, and its performance indicators were comparable to those reported by other state-of-the-art detection models. In the literature [18], an assist system including embedded system jetson AGX Xavier and binocular depth camera zed 2 was proposed to help visually impaired people walk outdoors.…”

Is the exoscope ready to replace the operative microscope in transoral surgery?

“…imagingbased navigation, neuromonitoring, different light filters and artificial intelligence tools) in a unified view. This may lead to a better identification of essential anatomic landmarks and superficial/deep tumour extension, thus refining surgical access and tumour resection through the growing field of videomics [51][52][53].…”

Artificial intelligence in clinical endoscopy: Insights in the field of videomics

“…With regard to real-time detection, Matava et al ( 19 ) and Azam et al ( 20 ) developed CNN algorithms that were applied in real time during videoendoscopy and which aimed at identifying, on the one hand, normal airway anatomy and, on the other hand, UADT lesions. Using this type of approach, DL may be a useful complementary tool for clinicians in endoscopic examinations, progressively implementing the concept of human–computer collaboration.…”

“…ResNet and Inception achieved a specificity of 0.98 and 0.97 and a sensitivity of 0.89 and 0.86, respectively. Finally, Azam et al ( 20 ) identified a CNN model for real-time laryngeal cancer detection in WL and NBI videoendoscopies. The dataset, consisting of 219 patients, was tested with an algorithm that achieved 0.66 precision (i.e., positive predictive value = true positives over true and false positives), 0.62 recall, and 0.63 mean average precision with an IoU > 0.5.…”