Feedback training with a novel artificial intelligence (AI)-based automatic bronchial segment identification system helps novices perform faster, more systematic, more complete bronchoscopies, according to study findings published in Chest.
Researchers assessed whether an AI-based automatic bronchial segment identification system offering feedback improves the end-of-training performance of novice bronchoscopists.
The researchers conducted a randomized controlled trial in a simulated setting at the Copenhagen Academy for Medical Education and Simulation (CAMES), Rigshospitalet, Denmark, from mid-April to mid-May 2023. The study included 20 participants recruited from the local university’s Facebook group for medical students. Those with prior clinical or simulated endoscopic procedure experience were excluded. The study authors stated that the training approach would not be relevant if more than 20 trainees were needed to show an effect.
All participants watched a 3-minute video introduction to simulation-based training explaining basic bronchoscope handling and the 2-part study design, training, and testing. Participants were then stratified by sex and randomly assigned 1:1 into the feedback group (n=10), in which participants watched a 4-minute instructional video on how to use the AI software, and the control group (n=10), in which participants watched a 2-minute instructional video with on traditional training using the Four Landmarks Approach, which splits the bronchial tree into 4 landmarks and couples these with the angle of the scope.
Training guided by this novel AI makes novices perform more complete, more systematic, and faster bronchoscopies.
Participants in the feedback group chose which of 3 metrics (diagnostic completeness, structured progress, procedure time) to receive feedback from, and after each bronchoscopy on a mannequin, participants received reports with procedure time and a checklist of visualized segments. The feedback group practiced with the AI on and off.
Participants in the control group received an instruction book and a poster highlighting the training method and the bronchial anatomy of the 18 bronchial segments. This group received procedure time feedback only.
All participants were allowed to train as long as they wanted up to a max of 3.5 hours. Participants then immediately performed a full bronchoscopy without guidance or aids, in the same circumstances for both groups.
The feedback group trained for a median of 49 minutes longer than the control group (P =.029). As a predictor variable in multiple linear regression, time spent training did not affect diagnostic completeness or structured progress, but did affect procedure time.
Researchers found the feedback group significantly outperformed the control group for the tested metrics for diagnostic completeness (median difference [MD], 3.5 segments; P <.001), structured progress (MD, 13.5 correct progressions; P <.001), and procedure time (MD, -214 seconds; P =.002).
The feedback group scored significantly higher for the total intrinsic motivation inventory (a 6-statement questionnaire to rate each statement from 1 [not at all true] to 7 [very true]) and specifically on the inventory statements (“I did well compared with other students”; “I would recommend this training system”; “I’d like to continue using this system”).
Study limitations include lack of testing for retention of skill and the inclusion of only novice bronchoscopists.
The researchers concluded, “Training guided by this novel AI makes novices perform more complete, more systematic, and faster bronchoscopies.” They added, “The [feedback group] not only visualized more of the bronchial tree, but they also did so in a much more structured order.” The researchers said future studies should examine use of this AI in a clinical setting and its effects on more advanced learners.
Disclosure: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.