Multicenter Validation of Video-based Deep Learning to Evaluate Defecation Patterns on 3-dimensional High-definition...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning shows promise in diagnosing gastrointestinal motility disorders.
• Three-dimensional high-definition anal manometry (3D-HDAM) offers detailed insights.
• Validating AI for 3D-HDAM interpretation is crucial.

Purpose of the Study:

• To validate a deep learning algorithm for spatiotemporal analysis of 3D-HDAM.
• To assess the algorithm's diagnostic accuracy in a multicenter setting.
• To compare AI performance against expert interpretation.

Main Methods:

• 1214 anorectal manometry studies from 3 health care systems (2018-2022).
• Deep learning algorithm performance compared to London consensus protocol (expert interpretation).
• Area Under the Curve (AUC) calculated via bootstrap sampling; Wilcoxon tests for confidence score correlation; Gaussian Mixture Modeling for feature clustering.

Main Results:

• The deep hybrid learning algorithm achieved high AUCs (0.99, 0.90, 0.79) across sites.
• Algorithm confidence scores correlated well with expert interpretation of ambiguity.
• Two novel dyssynergia patterns were identified, potentially representing distinct phenotypes.

Conclusions:

• Three-dimensional high-definition anal manometry (3D-HDAM) with video-based deep learning is effective for evaluating anorectal dyssynergia.
• This technology is clinically relevant and aids in diagnosis.
• Future applications may extend to other motility disorders and treatment evaluations.