JMIR Publications

ABSTRACT

Background:

The integration of artificial intelligence (AI) into intraoperative surgical imaging represents an emerging frontier in digital health. Despite advances in preoperative computed tomography (CT)–based surgical planning, real-time translation of imaging data into actionable intraoperative guidance remains limited by CT-to-body divergence—a fundamental information gap between preoperative digital models and the dynamic surgical field. This divergence, driven by lung deflation under anesthesia and positional changes, represents a critical digital-to-physical registration challenge that current preoperative imaging workflows fail to address in real time. This study evaluated the performance and safety of the LungVision system, a portable AI-driven digital platform that integrates preoperative CT data with real-time fluoroscopic image fusion, for intraoperative tumor localization during thoracoscopic lung resection.

Objective:

This study aimed to evaluate the clinical feasibility, localization accuracy, and safety of the LungVision system—an AI-augmented fluoroscopic navigation platform—for real-time intraoperative localization of small pulmonary nodules during thoracoscopic surgery.

Methods:

A prospective single-center study enrolled fourteen patients with pulmonary nodules requiring localization prior to thoracoscopic resection between March and September 2024. The platform comprises a passive radiopaque positioning board, an AI-powered computing unit for real-time image processing, and a tablet-based interface for procedural planning and augmented visualization. All patients received dual localization with either preoperative CT-guided dye injection or Archimedes virtual bronchoscopic navigation, followed by intraoperative localization with the LungVision system and video-assisted thoracoscopic surgery. Demographic data, lesion characteristics, procedural performance, and procedure-related complications were collected.

Results:

The mean patient age was 57.2 years, and 92.9% were non-smokers. Most nodules were peripherally located (85.7%), with a mean diameter of 9.3 ± 5.3 mm and a mean CT attenuation of −320.1 Hounsfield units. LungVision successfully localized all target lesions intraoperatively, with a mean navigation time of 38.6 minutes. Complete resection was achieved in all cases, and 71.4% of nodules were pathologically malignant. No intraoperative or localization-related complications were observed. The system was integrated into the existing operating room without additional infrastructure modifications.

Conclusions:

The LungVision system demonstrated high accuracy and safety for intraoperative localization of small, hypodense pulmonary nodules. By minimizing CT-to-body divergence and integrating seamlessly into existing bronchoscopic and surgical workflows, this AI-driven platform represents a scalable and infrastructure-light alternative to conventional localization strategies, warranting further evaluation for broader clinical implementation.