BACKGROUND: Quality control can reduce variations in operators' performance conducting magnetically controlled capsule gastroscopy (MCCG). However, an optimal method for quality control in routine MCCG procedures is still lacking. This study aimed to develop an automatic quality control system (AQCS) and assess its effectiveness in a clinical trial. METHODS: We developed the AQCS using convolutional neural network (CNN) models to monitor inspection completeness, evaluate gastric cleanliness, and identify suspicious lesions. Then, patients were prospectively randomized to undergo routine MCCG with or without AQCS assistance. The primary outcome was the blind spot rate in the AQCS and control groups. RESULTS: The CNN model demonstrated specificity of 98.27-99.30% and sensitivity of 76.33-96.35% in gastric site identification. Between August 27, 2021, and July 28, 2022, a total of 200 patients were randomized, with 98 and 96 patients analyzed in the AQCS and control groups, respectively. Compared to the control group, the AQCS group achieved lower blind spot rates (median: 0.00% vs. 16.67%, P < 0.01), higher lesion detection rates (75.51% vs. 60.42%, P = 0.02), and comparable gastric examination time (28.63 min vs. 27.58 min, P = 0.48). Additionally, AQCS showed high consistency with expert evaluations in cleanliness assessment (Kappa = 0.95, P < 0.01). No serious adverse events occurred in either group. CONCLUSIONS: AQCS significantly reduced the blind spot rate during MCCG procedures. It could be a powerful assistant tool to mitigate operator skill variability and enhance the overall quality of routine MCCG examinations. TRIAL REGISTRATION: ClinicalTrials.gov Identifier NCT04954677.
Wang et al. (Sat,) studied this question.