AI-Assisted Bronchoscopy Navigation and Diagnostic System for Severe Pneumonia: A Multicenter Prospective Study

Overview

This study will evaluate an embodied-intelligence-assisted bronchoscopy navigation and diagnostic system for patients with severe pneumonia who require clinically indicated bronchoscopy. The system provides real-time visual cues and voice prompts to help physicians localize target lung segments and describe endobronchial findings; physicians remain fully responsible for all clinical decisions. The trial is designed as a prospective, multicenter, controlled study conducted at about ten hospitals in China, with an anticipated sample size of approximately 100 patients. The primary objective is to determine whether AI assistance improves diagnostic agreement compared with the reference assessment, while secondary objectives include navigation success (e.g., loss-of-path rate), procedure time, and complication rates. The results will provide evidence on the safety and effectiveness of AI-assisted bronchoscopy and support product validation and registration.

Rationale: Severe pneumonia often requires bronchoscopy for diagnosis and therapy, yet accurate segmental localization and consistent interpretation are operator-dependent and time-consuming. Prior work by the sponsor team integrated CT-based planning with real-time bronchoscopic guidance and AI-based lesion characterization. The investigational system overlays guidance and voice prompts during bronchoscopy to standardize navigation and reduce errors. Design: Prospective multicenter controlled study in ICUs/respiratory units. Approximately 100 adults undergoing clinically indicated bronchoscopy for severe pneumonia will be enrolled across \~10 centers in China. Procedures will be performed either with AI assistance or with standard bronchoscopy per site workflow. Outcomes include: (1) diagnostic agreement versus a predefined reference (primary); (2) navigation success/loss-of-path; (3) procedure time; and (4) adverse events (e.g., bleeding, hypoxemia, infection). Enrollment is planned to start in Sep 2025, with anticipated primary completion around Sep 2026. Operations: Data will be collected using a standardized CRF and analyzed per a prespecified SAP. The investigational software is integrated with an endoscopic image processor developed with an industry partner; physicians can follow or ignore AI suggestions at their discretion. The study will generate evidence on effectiveness and safety to inform subsequent registration testing and scaling.