Clinical feasibility study of generative depth estimation to diagnose airway stenosis using 3D slicer

Airway Stenosis (AS) refers to the obstruction observed in the expiration phase of breathing. Bronchoscopy is a minimally invasive procedure for diagnosing and treating AS. Physicians assess the severity of the obstruction using the stenosis index (SI). However, the prevailing method for estimating SI is subjective and has been found to need more precision.

This study aimed to develop a generative depth estimation integrated into a 3D Slicer to assist clinicians in determining the stenosis index from CT and bronchoscopic images. We compared the efficacy of this video-based SI estimation to the CT-based method, using subjective human-based bronchoscopic assessments as a reference standard. We integrated the earlier conceptualized generative depth estimation technique into a 3D Slicer module, making it publicly accessible. An interim clinical analysis was conducted, juxtaposing the SI quantification performance of an expert bronchoscopist in the operating room (human-based) with the video-based and CT-based estimates provided by our new tool on 30 patient datasets assessed by three other expert bronchoscopists.

The correlation coefficient between the human-based SI estimation and CT-based assessment was 0.58 ( $p<0.001$ ). Comparatively, the correlation between subjective SI estimation and video-based SI estimation was 0.78 ( $p<0.001$ ), and that between CT-based and video-based assessments was 0.34 ( $p<0.001$ ).

Our study introduces a public tool that employs depth estimation to evaluate AS observed in bronchoscopy. When comparing our method to human subjective assessments and CT-based evaluations, the results underscore the potential of our video-based AS quantification tool in aiding physicians during bronchoscopy.