Fig. 5: Results of in vivo experiments.
a CT images of the live porcine lung. b Bronchoscopic images obtained by the expert without the AI co-pilot and the attending doctor with the AI co-pilot, corresponding to the two paths in Supplementary Fig. 24. c Actuation displacement curves and heatmaps of the mean values and fluctuation ranges. d Actuation force curves and heatmaps of the mean values and fluctuation ranges. e Image error during bronchoscopy. The results show that the attending doctor with the AI co-pilot consistently maintained smaller image errors than the expert without the AI co-pilot. Line plots with error bands show time-varying mean image errors ±95% CI within a time window of 20 frames during each procedure. f Statistical results for the image errors. The numbers of recorded frames are n = 18653 for Expert and n = 17354 for AI. Bar plots depict the mean image error and the 95% confidence interval for all frames during the bronchoscopy procedures in the live porcine lung, i.e., the mean ± 95% CI image error for Path 1 and Path 2. The results show the safer steering achieved in AI-assisted bronchoscopy. g Statistical results for the human intervention ratios, determined by the number of time stamps at which the action of the doctor’s hand changed from the last time stamp relative to the total number of time stamps recorded throughout the whole bronchoscopy procedure. The results demonstrate that AI-assisted bronchoscopy can effectively minimise the physical strain and cognitive burden on doctors. Human intervention ratios of n = 8 independent experiments are presented as mean values ± 95% CI, and specific data points are overlaid on the bar plot.
