Gastroenterology and Artificial Intelligence: 3rd Annual Artificial Intelligence Summit (On-Demand) | September 2021-7- How to Evaluate AI in the GI Literature and Clinical Trials

7- How to Evaluate AI in the GI Literature and Clinical Trials

Pdf Summary

In this presentation on evaluating AI in gastrointestinal (GI) literature and clinical trials, Dr. Cesare Hassan discusses the outcome cycle and the different stages of evaluating AI in the GI field. He emphasizes the importance of assessing both the efficacy (benefit in ideal circumstances) and effectiveness (benefit in the real world) of AI in detecting and predicting various GI conditions.<br /><br />Dr. Hassan highlights the need to evaluate AI's standalone performance through controlled trials and real-life studies. He mentions the challenges of testing AI, such as the need for a representative training dataset and the selection and annotation processes involved. He provides examples of AI systems used in GI, including GI-Genius, Discovery AI, and CAD-EYE, and shares information about their training datasets and the number of cases and centers involved in their development.<br /><br />The presenter also discusses the biases that may exist in the training and testing datasets and emphasizes the importance of assessing the generalizability of AI to community-based endoscopy and the expertise of the endoscopists involved.<br /><br />Dr. Hassan explores the effectiveness of AI in detecting neoplasia and characterizing lesions, discussing the considerations and potential biases involved in randomized controlled trials and sequential setting studies. He also mentions the value of AI in terms of its methodology, outcomes, and exploration of the consequences of false positives.<br /><br />In conclusion, Dr. Hassan emphasizes the importance of understanding the details of the training and testing databases when using an AI system and highlights the need for randomized controlled trials to assess the interaction between human and AI in the field of GI. He also mentions the potential impact of AI on post-polypectomy surveillance and the cost-effectiveness considerations associated with AI implementation in colonoscopy.

Keywords

evaluating AI

gastrointestinal literature

clinical trials

efficacy

effectiveness

detecting GI conditions

standalone performance

controlled trials

training datasets

randomized controlled trials