Efficacy of a virtual bronchoscopic navigation system improved by deep learning for biopsy of peripheral lung lesions: a single-center randomized controlled trial

Background: Existing virtual bronchoscopic navigation (VBN) biopsy systems are ineffective in reconstructing small airway trees of 2-3 mm, which leads to an inability to accurately guide the biopsy of peripheral pulmonary lesions (PPL). This study intended to compare the diagnostic rate of PPL by the modified SARS-pro (small airway reconstruction system-pro) system and the original VBN system. Methods: This single-center randomized controlled trial enrolled subjects aged ≥18 years who had one or more PPLs between August 2023 and December 2024 at a hospital. Subjects were randomly assigned to the SARS-pro system and the VBN system (1:1). The outcomes were the diagnostic positive rate of PPL and adverse events in subjects. The diagnosis rate outcomes were evaluated in the per-protocol set (PPS). Results: Ninety-five eligible subjects were recruited, 95 subjects were included in the full analysis set (FAS), and 92 subjects were included in PPS. The SARS-pro system exhibited a higher positive diagnostic rate for PPL than the VBN system on FAS [43 (91.49%) vs. 30 (62.50%), P = 0.002] and PPS [43 (93.48%) vs. 30 (65.22%), P = 0.002]. Furthermore, the positive diagnostic rate of PPL was significantly higher for the SARS-pro system than for the VBN system in some characteristic subgroups, such as females, no smoking history, lesion size of 1-2 cm, and no bronchial signs. Conclusion: The improved SARS-pro system presented a higher diagnostic rate for PPL than the VBN system, which may suggest the application of deep learning technology in improving the diagnostic rate of lung biopsy.