The aim of this study is to evaluate the performance of real-time auxiliary system based on artificial intelligence algorithm in lesion detection and quality control in magnetically controlled capsule endoscopy.
Magnetically controlled capsule endoscopy (MCE) has been used in clinical practice for gastric examination, with high sensitivity and specificity of 90.4% and 94.7%, respectively. Therefore, a real-time auxiliary system based on convolutional neural network deep learning framework was developed to assist clinicians to improve the quality in MCE examinations. Patients referred for magnetically controlled capsule endoscopy (MCE) in the participating center were prospectively enrolled. After passage through the esophagus, physician will finish the gastric examination under magnetic steering with the real-time auxiliary system. Professional operators guarantee the integrity of the examination and the diagnostic results of professional endoscopist was used as the gold standard. The system diagnosis results was recorded at the same time. The sensitivity, delay time, specificity of lesions and anatomical landmarks will be analyzed.
Study Type
OBSERVATIONAL
Enrollment
50
The patients swallowed the MCE with a small amount of water in the left lateral decubitus position. Once the capsule reached the stomach after investigating the esophagus, it was lifted away from the posterior wall, rotated and advanced to the fundus and cardiac regions, and then to the gastric body, angulus, antrum and pylorus. The magnetic steering time for passing through the pylorus was not allowed more than 15 min. The real-time auxiliary system implemented real-time processing for the output image of MCE system. Professional operators guarantee the integrity of the examination and the diagnostic results of professional endoscopist was used as the gold standard. The system diagnosis results was recorded at the same time. The sensitivity, delay time, specificity of lesions and anatomical landmarks will be analyzed.
Changhai Hospital
Shanghai, China
RECRUITINGSensitivity
The sensitivity of lesions detected by system
Time frame: up to 2 weeks
Delay time
Average preprocessing and displaying times before and after the execution of system
Time frame: up to 2 weeks
Specificity
The specificity of detected by system
Time frame: up to 2 weeks
Accuracy of anatomic landmarks identification
Accuracy of anatomic landmarks ( cardia, fundus, body, lesser curvature, greater curvature, angle, antrum and pylorus) identified by real-time identification system
Time frame: up to 2 weeks
Completeness of real-time observation with assistance
Whether the clinician observed all anatomic landmarks of stomach (cardia, fundus, body, angulus, antrum and pylorus).
Time frame: up to 2 weeks
Accuracy of heat map
The rate of the highlighted area indicated in the lesion
Time frame: up to 2 weeks
Lesion detection yield
The ratio of lesions detected by system to all lesions
Time frame: up to 2 weeks
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.