(PR032) Artificial Intelligence Models for Detecting Periapical Radiolucent Lesions on Various Imaging Modalities: A Comprehensive Literature Review

Objectives: The study aimed to assess the effectiveness of AI models in detecting periapical radiolucent lesions on different dental imaging modalities. Method and Materials: A comprehensive electronic search was conducted in PubMed/Medline, EMBASE, IEEE Xplore, Google Scholar, Scopus, and Web of Science databases from January 2018 to January 2024 using keywords: Artificial intelligence, deep learning, dental radiograph, image analysis, machine learning, periapical lesion, and periapical radiolucency. This search aimed to identify studies utilizing AI tools for the detection of periapical radiolucency. Initially, 593 articles were included for title and abstract screening. After the application of inclusion and exclusion criteria, a total of 14 articles were selected for this review.

Results: The results indicate that AI models demonstrated high accuracy, sensitivity, and specificity in detecting periapical radiolucent lesions on CBCT, panoramic, and periapical radiographs. Convolutional neural networks (CNNs) emerged as the most frequently utilized algorithm. The training strategy of the network, as well as the lesion category and severity level, significantly affected CNN performance. There are several challenges to overcome, including uncertainty about the generalizability of the developed AI models, concerns regarding data privacy, the necessity for large, high-quality datasets for training AI models, and the integration of AI tools into existing workflows. Attempting to integrate non-imagery data such as clinical signs, symptoms, and history with imagery data in deep learning systems could improve outcomes.

Conclusions: Within the limitations of this study, it is concluded that applying AI for detecting periapical radiolucent lesions can enhance clinical work efficiency and reduce interpretation bias.