Accuracy of convolutional neural network in the diagnosis of alveolar bone loss due to periodontal disease: A systematic review and meta-analysis

Abstract

Periodontitis is an inflammation of the supporting structures of teeth, involving progressive alveolar bone loss. A computer-assisted detection system based on a deep convolutional neural network (CNN) algorithm was developed. This study aimed to assess the existing literature to determine the accuracy of the CNNs for diagnosing and measuring periodontal bone loss (PBL). We used a modified approach to Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension diagnostic test accuracy by searching the following databases: Scopus, PubMed, Cochrane, and Web of Science, in addition to gray literature. Medical Subject Headings terms and Keywords in the title and abstract fields, as well as subject headings for both periodontal disease/bone loss and CNN/artificial intelligence, were used to search the existing literature for publications relevant to the evaluation of the accuracy of CNN for the detection and measurement of alveolar bone loss over the past three decades (January 1990–May 2021). Quality analysis was performed using the quality assessment and diagnostic accuracy tool-2. Four thousand six hundred and ninety potentially relevant titles and abstracts were found after an initial electronic and manual search and removal of duplicates. Applying the inclusion and exclusion criteria yielded 75 publications, which were further analyzed for relevance and applicability. Most of the included studies were observational. Following the critical analysis, eight publications were used to assess CNN's precision of the CNN for PBL measurements. The CNN system successfully determined PBL. Therefore, it can facilitate the diagnosis and treatment planning for dentists in the future.