To present a proof-of-concept model for the automatic identification and characterization of morphologic patterns associated with common benign anorectal conditions in endoanal ultrasound (EAUS), highlighting its significance in improving diagnostic accuracy.
Key Findings:
The CNN achieved 100% sensitivity, specificity, and accuracy for anal fissures, indicating its potential as a reliable diagnostic tool.
For external lacerations, sensitivity was 82.5%, specificity 93.5%, and accuracy 88.2%, suggesting areas for further improvement.
For internal lacerations, sensitivity was 91.7%, specificity 85.9%, and accuracy 88.2%, demonstrating the model's effectiveness in diverse lesion types.
Interpretation:
The study demonstrates the potential of AI to enhance diagnostic accuracy in EAUS, specifically addressing the limitations of current practices such as the steep learning curve and variability in results.
Limitations:
Relatively steep learning curve for EAUS, which may hinder widespread adoption.
Limited accessibility to EAUS, often restricted to specialized centers, potentially biasing the sample.
Intra- and intervariability affecting reproducibility of results, which could impact clinical trust in AI applications.
Conclusion:
This study represents a pioneering effort in applying AI for automatic detection of benign anal lesions in EAUS, laying the groundwork for future advancements in proctology and suggesting further research to validate these findings in broader clinical settings.
