Analysis of artificial intelligence and machine learning models’ effectiveness in regards to bone loss verification as a criterion for assessing the state of periodontal tissues based on orthopantomography data

Abstract

Introduction. Artificial intelligence models provide over 70% accuracy in classifying periodontitis cases using datasets of various nature, however, the most significant proportion of such was represented by orthopantomograms.Objective of the research. To analyze data on the effectiveness of using various artificial intelligence and machine learning models for the bone loss verification within the projection of natural teeth as a criterion for assessing the state of periodontal tissues based on orthopantomography data and to establish the accuracy indicators of such models in the structure of comprehensive diagnostics of periodontal patients.Materials and methods. Processing of publications selected for the primary cohort of scientific works was carried out by analyzing their text and providing manual data extraction in accordance with the following research categories: criteria that were used to assess the effectiveness of an artificial intelligence models focused on the detection, quantification and/or classification of the bone loss level in the projection of natural teeth based on orthopantomography data; indicators of the tested models performance effectiveness in accordance with the criteria used in different studies; the technologies (algorithms), which formed the basis for the development of the proposed target artificial intelligence models.Results and discussions. The prevailing part of the studies demonstrated the experience of using convolutional neural networks as the main approach in the structure of artificial intelligence models focused on verifying the level of bone loss in the projection of natural teeth based on the orthopantomography data. According to previously conducted studies cumulative diagnostic accuracy of such networks used for the above-mentioned purpose was equaled to 0,85, while cumulative sensitivity was equaled to 0,84, and cumulative specificity was equaled to 0,85. The processing of orthopantomograms using machine learning models was characterized by high efficiency in verifying cumulative bone loss in the projection of remaining dentition.Conclusions. During the literature review conducted to assess the effectiveness of clinically-oriented artificial intelligence and machine learning models for the verification of bone loss in the projection of remaining dentition, it was found that the accuracy of these models according to previously published studies has increased significantly after 2020, critically approaching the average indicator of 90%, meanwhile some studies demonstrated data controversial to generally established tendencies. Literature data of the 2020–2024 period indicates a pronounced positive growth in the sensitivity indicator of the above-mentioned models, while positive changes in the specificity indicator are less pronounced in terms of dynamics. Convolutional neural networks represent the approach most frequently described in the literature as usable for the development of artificial intelligence models focused on detecting and classifying levels of bone loss in the projection of natural teeth based on the orthopantomography data.