Comparison of graphical user interfaces for computer-aided detection of Barrett’s neoplasia

Aims Despite the surge of artificial intelligence applications in endoscopy, the interaction between the endoscopist and AI system remains an underexplored aspect. This endoscopist-AI interaction ultimately may have significant impact on the performance of the AI system in daily clinical practice. The aim of this study was to compare two graphical user interfaces for a computer aided detection (CADe) system for Barrett’s neoplasia.

Methods This study involved a comparative analysis between two distinct graphical user interface (GUI) designs for a computer-aided detection (CADe) system: the traditional bounding box GUI and an alternative heatmap GUI. For this study, we utilized a well-established and rigorously evaluated CADe system. A group of 37 endoscopists from 6 countries assessed 70 Barrett’s esophagus videos. All videos were analyzed by the CADe system and comprised, at some point, a CADe detection, regardless of the actual presence of neoplasia. The study had two phases. Initially, videos were shown with either a bounding box or heatmap. After a two-week wash-out period, the same videos were reordered and displayed with the alternate GUI. Endoscopists marked potential neoplastic lesions and biopsy sites and provided their personal GUI preference.

Results The study found no significant difference in classification performance between the bounding box and heatmap visualizations (sensitivity 83% vs. 83%, p=0.29; specificity 86% vs 86%, p=0.09). Localization accuracy also did not differ significantly between the two methods, both achieving a median score of 97%. In total, 23 endoscopists favored the heatmap, while 14 preferred the bounding box.

Conclusions Although endoscopists expressed a preference for the heatmap GUI, this was not associated with a statistical difference in performance outcomes.

Publication History

Article published online:
15 April 2024