First-in-human pilot study of a robotic-assisted system for image-guided trans-thoracic lung biopsy: Clinical feasibility and exploratory post-hoc AI trajectory analysis.
Authors
Affiliations (6)
Affiliations (6)
- Department of Vascular & Interventional Radiology, Singapore General Hospital, Singapore. Electronic address: [email protected].
- Department of Vascular & Interventional Radiology, Singapore General Hospital, Singapore.
- NDR Medical Technologies, Singapore.
- Centre of Quantitative Medicine, Duke-NUS Medical School, Singapore.
- Department of Radiology, Nagoya City University Graduate School of Medical Sciences, Japan.
- Division of Interventional Radiology, Department of Radiology, Stanford University School of Medicine, USA.
Abstract
This prospective, single-center clinical trial evaluated the technical feasibility and clinical success of the Automatic Needle Targeting (ANT-C) patient-mounted robotic needle-guidance platform for CT-guided trans-thoracic lung biopsy under local anesthesia, incorporating an exploratory post-hoc geometric analysis of parallel artificial intelligence path planning via the NDAnalyzer software. Out of thirty-one recruited patients, one was excluded due to on-table lesion resolution, leaving a final cohort of thirty evaluable subjects with a median nodule size of 24 mm. While interventional radiologists executed target acquisition using independently planned trajectories, the automated software generated alternative paths in a parallel, blinded fashion to ensure zero clinical influence on the live procedure. The platform demonstrated a high technical success rate of 93.3% and a 100% clinical success rate, successfully harvesting adequate tissue cores for full histopathological and molecular profiling in all cases. Operational efficiency improved significantly as operators gained familiarity, with median procedural durations decreasing from 61.4 min in the first ten cases to 30.5 min in the final ten cases. Concurrently, median total radiation exposure significantly decreased from 427.3 mGy·cm to 303.7 mGy·cm between the early and late deciles. No major complications occurred, and minor, self-limiting Grade A adverse events were limited to asymptomatic perilesional hemorrhage and trace pneumothoraces. Furthermore, exploratory modeling revealed a median angular deviation of only 3.5° between the operator's selected path and the closest automated alternative. The robotic system is clinically viable and technically feasible for awake patients, though the AI's automated path-ranking cost-function requires further refinement prior to live workflow integration.