Human-machine interaction in optical polyp diagnosis: decision-making after CADx polyp diagnosis

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Artificial intelligence (AI)-based computer-aided diagnosis (CADx) systems are increasingly used for colorectal polyp classification during colonoscopy.
• The effectiveness of AI CADx systems is influenced by how endoscopists interpret AI-generated predictions.

Purpose of the Study:

• To evaluate endoscopist decision-making processes when reviewing AI CADx outputs for colorectal polyps.
• To compare endoscopist interpretations against histopathology as the ground truth.

Main Methods:

• Secondary analysis of a prospective study (NCT06822816) involving patients with polyps ≤10 mm.
• Included cases with documented AI CADx output and endoscopist diagnosis.
• Assessed accuracy of accepted AI diagnoses and reclassification patterns after AI output rejection.

Main Results:

• The study analyzed 754 polyps from 367 patients.
• Accepted AI diagnoses were correct for 89.1% of neoplastic and 68.7% of hyperplastic polyp predictions.
• Endoscopists more frequently accepted incorrect AI classifications for neoplastic polyps (64.9%) than hyperplastic (26.5%).
• Rejection of correct AI neoplastic predictions led to misclassification as sessile serrated lesions (SSLs) in 23.4% of cases; 29.5% of rejected correct hyperplastic predictions were also misclassified as SSLs.

Conclusions:

• Endoscopists demonstrate a diagnostic bias towards neoplasia, frequently misclassifying hyperplastic polyps as SSLs.
• This bias may enhance safety but could lead to unnecessary surveillance colonoscopies and increased healthcare costs.