PROGNOSTICATION OF LOSS AND CRITICAL COMPROMISE OF THE DENTAL IMPLANTS FUNCTIONAL STATE BY USING MACHINE LEARNING METHODS: PROSPECTS FOR CLINICAL IMPLEMENTATION

Abstract

Introduction. Accumulation and targeted processing of data regarding trends in development and actual experimental effectiveness of proposed artificial intelligence models focused on quantifying the probability of dental implant loss or critical compromise of their functional status will contribute to the objectification of their prospective applicability and significance during initial clinical prognosis while selecting prosthetic rehabilitation methods using different types of intraosseous fixtures under variable distribution scenarios. Objective of the research. To assess the predictive significance and feasibility of the direct clinical implementation of machine learning systems for forecasting the probability of dental implant loss, as well as quantifying the risk of conditions leading to critical compromise of their functional capacity, rendering them as incapable of fulfilling the role of intraosseous support for prosthetic structures. Materials and methods. Study design was based on the principles of systematic analysis and mapping of scientific literature sources covering the application of machine learning and artificial intelligence methods for predicting the performance, survival, and risk of failure of dental implants. Provided research adhered to guidelines ensuring transparency and reproducibility of analytical reviews in medical science, with a focus on clinically-oriented predictive models based on artificial intelligence technologies. Results and discussion. Analysis of scientific literature indicated that machine learning and artificial intelligence systems generally demonstrate high potential effectiveness in predicting dental implant performance, survival, and risk of failure, with reported accuracy levels ranging from 70% to 96,13%, depending on the type of data used and model architecture. The most promising results were observed with multimodal and ensemble approaches that integrate clinical, anamnestic, and radiographic data. However, most existing models are characterized by sample heterogeneity, limited external validation, and a predominant focus on detecting already manifested pathological changes rather than performing true early risk prediction. Conclusions. Majority of current models are oriented toward the detection of already manifested pathological changes, such as periimplant bone loss or signs of peri-implantitis, which essentially restrict their functionality to classification of clinically unfavorable states rather than early prediction. Lack of systematic external and multicenter validation, insufficient calibration of models for clinically significant endpoints, and deficit of unified and standardized datasets reduce the level of evidence and limit the widespread clinical implementation of machine learning and artificial intelligence systems in this domain.