Comparison of the Diagnostic Performance of Artificial Intelligence-Based Models and Endoscopists for the Diagnosis of Colorectal Cancer Using Colonoscopy: a Systematic Review

Abstract

Background:

Colorectal cancer is known as one of the most common cancers with a high mortality rate. Early diagnosis of the disease helps to reduce mortality. Regarding the rapid development of artificial intelligence (AI) models for the diagnosis of different cancers, this study compares the diagnostic performance of AI models with the ability of experts or non-expert endoscopists to diagnose cancerous polyps.

Materials and Methods:

 The present study was conducted as a systematic review in PubMed, Scopus, and Web of Science databases, as well as those studies that were designed according to the research question and referenced methods to compare the performance of endoscopists and AI models. Data extraction was done by two researchers, and the criteria for comparison were accuracy, sensitivity, specificity, and positive and negative predictive values.

Results:

Out of the total 838 articles obtained from the database search, 112 duplicates and 683 irrelevant records were excluded. Besides, 35 records were removed after content analysis, and finally, nine articles remained for data extraction. Based on the results, AI-based models can improve the diagnostic performance of less experienced experts. However, by considering quantitative performance indicators such as accuracy, sensitivity, and specificity, the performance of experienced endoscopists was significantly higher than AI models and the less experienced experts.

Conclusion:

AI-based models can be suitable for improving the diagnostic performance of endoscopists; however, the focus of using these models should be on helping less experienced ones.