Automated vision-based assistance tools in bronchoscopy: stenosis severity estimation

TL;DR

This work targets automated, CT-free estimation of subglottic stenosis severity from bronchoscopy video. It introduces a two-step pipeline: first, select an optimal frame by tracking the darkest lumen region using illumination decline; second, reconstruct a 3D airway from that frame with LightDepth to measure stenosis and compute $PSA = (1 - \frac{A_{stenosis}}{A_k}) \times 100$ and $PSD = (1 - \frac{diameter_{stenosis}}{diameter_k}) \times 100$. The authors release the Subglottic Stenosis Dataset and demonstrate consistency with CT-ground-truth and expert estimations, achieving robust repeatability and faster assessment (about 7 seconds per keyframe on a RTX3090). The approach reduces subjectivity in SGS evaluation, lowers exploration time, and minimizes patient radiation exposure by avoiding CT scans, while providing a publicly available benchmark to spur further research. Overall, the method offers a practical, repeatable, and accessible tool for automated SGS severity assessment from bronchoscopy alone, with potential integration into clinical workflows.

Abstract

Purpose: Subglottic stenosis refers to the narrowing of the subglottis, the airway between the vocal cords and the trachea. Its severity is typically evaluated by estimating the percentage of obstructed airway. This estimation can be obtained from CT data or through visual inspection by experts exploring the region. However, visual inspections are inherently subjective, leading to less consistent and robust diagnoses. No public methods or datasets are currently available for automated evaluation of this condition from bronchoscopy video. Methods: We propose a pipeline for automated subglottic stenosis severity estimation during the bronchoscopy exploration, without requiring the physician to traverse the stenosed region. Our approach exploits the physical effect of illumination decline in endoscopy to segment and track the lumen and obtain a 3D model of the airway. This 3D model is obtained from a single frame and is used to measure the airway narrowing. Results: Our pipeline is the first to enable automated and robust subglottic stenosis severity measurement using bronchoscopy images. The results show consistency with ground-truth estimations from CT scans and expert estimations, and reliable repeatability across multiple estimations on the same patient. Our evaluation is performed on our new Subglottic Stenosis Dataset of real bronchoscopy procedures data. Conclusion: We demonstrate how to automate evaluation of subglottic stenosis severity using only bronchoscopy. Our approach can assist with and shorten diagnosis and monitoring procedures, with automated and repeatable estimations and less exploration time, and save radiation exposure to patients as no CT is required. Additionally, we release the first public benchmark for subglottic stenosis severity assessment.

Figures (4)

• Figure 1: Overview of the proposed automatic stenosis estimation approach. Our two-step pipeline takes as input the bronchoscopy video, starting in front of the vocal cords. In each frame, the lumen is segmented and tracked until the camera reaches the subglottis and starts to see the stenosis. At that selected keyframe, we compute a 3D reconstruction of the explored airway, in which the stenosis is measured.
• Figure 2: Overview of our proposed approach. The first step detects the darkest area through grayscale intensity segmentation and tracks it along the sequence using IoU tracker. Once the camera reaches the adequate location to measure the stenosis, the darkest segment changes shape significantly. The tracking is lost and the keyframe is reached. That keyframe is then passed to our second step, which produces a 3D reconstruction of the airway using LightDepth model. The stenosed and non-stenosed airway areas are measured in that model.
• Figure 3: Examples of subglottic area appearance for patients A, B, C, D, G and H from our proposed dataset. Patients A, B, C and D present varying degrees of stenosis. Patients G and H are healthy, and the trachea with its characteristic cartilaginous rings can be seen in the back.
• Figure 4: Obstruction measurement as done clinically (Myer-Cotton classification) compared to our approach. The green area represents the stenosed area, for which the measurement is the same clinically and in our case. The red area corresponds to the healthy reference airway: clinically measured in the trachea below the stenosis, in our approach measured between the vocal cords and the stenosis. The obstruction (in blue) is the difference between the healthy and stenosed areas.