Comparison of AI-based radiographic interpretation versus endodontic specialists for identifying periapical lesions: An in vitro study

TL;DR

An AI system using CNN was compared to endodontic specialists in identifying periapical lesions from radiographs and showed similar accuracy but faster processing.

Contribution

The study demonstrates that AI can match expert performance in detecting periapical lesions while significantly reducing time.

Findings

AI achieved 89.2% sensitivity, 91.5% specificity, and 90.2% accuracy in lesion detection.

AI processed images over 20 times faster than endodontic specialists.

No significant differences were found in diagnostic measures between AI and experts.

Abstract

The detection of periapical lesion is a diagnostic problem that sometimes entails the use of experts who can correctly interpret the
radiographs to be accurate. Therefore, it is of interest to compare an AI system that uses CNN with endodontic experts in the classification
of periapical lesions on 500 digital periapical radiographs. The AI was sensitive, specific and accurate with 89.2, 91.5, and 90.2
respectively, which is close to the performance of specialists and is able to process images more than 20 times faster. There were no
differences in diagnostic measures of AI and experts. Thus, we show that AI-aided radiographic analysis may be used as a safe, time-saving
supplement to the endodontic practice to detect periapical lesions.