Computer vision scoring of endoscopically traced line figures in an inanimate scope tip coordination training model

Area of Science:

• Medical training and simulation
• Computer vision applications in healthcare
• Gastrointestinal endoscopy

Background:

• Fine motor control is crucial for endoscopic submucosal dissection (ESD) and hybrid ESD/Endoscopic Mucosal Resection (EMR) polypectomy.
• An inanimate training model uses paper tracings to assess trainee coordination, similar to needle knife use.
• Traditional scoring relies on manual assessment of tracing completion time and deviations.

Purpose of the Study:

• To evaluate the efficacy of a computer vision (CV) algorithm in scoring tracings from an inanimate endoscopy training model.
• To compare CV algorithm scoring against traditional manual scoring methods.

Main Methods:

• A CV model analyzed 352 tracings by aligning them to a reference template using perspective transformation.
• Color segmentation isolated ink, and deviations were quantified by subtracting the tracing from the template.
• Pixel-based analysis converted deviations to millimeters, with >2 mm classified as significant.

Main Results:

• The CV algorithm detected an average of 1.28 more deviations than manual scoring across various figures.
• Bland-Altman analysis showed good agreement between CV and manual scoring, within the 96% confidence interval.
• Variability in pen stroke thickness was a confounder, improved by adding orienting marks.

Conclusions:

• The CV algorithm effectively and accurately assesses tracing accuracy in the inanimate model, outperforming human scorers.
• Future CV applications could provide real-time feedback for trainees, enhancing endoscopy education.
• AI-driven evaluation holds promise for advanced endoscopy training programs.