30 - Assesing the reliability of artificial intelligence-based automatic digital model analysis compared to manual system

Abstract

In orthodontic treatment planning, model analysis plays a crucial role in evaluating the patient's dental and skeletal structure. Measurements and analyses are essential for understanding the severity of malocclusion and developing individualized treatment plans to achieve functional and aesthetic alignment. The accuracy of these measurements directly impacts treatment outcomes and patient satisfaction. Technological advancements have significantly influenced orthodontic model analysis. Within these digital systems, measurements can be obtained through manual segmentation of STL files uploaded into a software. With the advent of artificial intelligence (AI), software now enables automatic segmentation, offering precise measurements for model analysis parameters. AI-based automatic digital model analysis represents a paradigm shift in orthodontics, utilizing advanced algorithms to analyze digital models with greater speed, efficiency, and consistency than traditional methods. These systems promise transformative improvements but also raise questions regarding their diagnostic accuracy and clinical utility. Evaluating the differences between digital and manual methods is essential to ensure the reliability and safety of patient care. This study evaluates the accuracy and reliability of AI-supported digital model analysis compared to auto-segmented digital (AD), manual digital (MD), and manual plaster (MP) methods. Parameters such as tooth segmentation, measurement accuracy, and analysis efficiency are assessed. There remains a significant gap in the literature regarding studies that compare digital systems in their own methods of analysis options. Most of the digital software systems include manual segmentation and automatic segmentation options. The majority of the studies tend to compare digital systems to conventional plaster models, focusing on the differences and similarities between the two techniques. However, there is a lack of comprehensive analysis investigating how various digital methods perform relative to one another, without incorporating or relying on comparisons to traditional plaster model techniques. This gap highlights the need for studies that focus solely on the intra-domain evaluation of digital systems, which could provide deeper insights into their precision and reliability under different diagnostic and operational conditions. Twenty-four individuals, comprising 8 males and 16 females (19.5± 6,05 years of age), with all permanent teeth and mild to moderate crowding were selected. Linear measurements were made with a digital caliper on the plaster models of the selected patients. The digital models were manually and automatically segmented and measured using a Software. Mesiodistal tooth dimensions, arch widths were measured, and Bolton analysis was performed. Repeated measurements and statistical analysis show that all methods give consistent results. Low standard deviations and narrow confidence intervals indicate that each system is reliable, but that some methods may be more precise. In comparing the methods, a significant difference was found when examining the size of teeth other than premolars (p<0.05). There was no statistically significant difference between AD and MD systems in parameters such as arch length deficiency, and Bolton discrepancy. (p>0.05). The AD and MD methods produced consistent measurement results. The MP method showed a higher probability of error due to conventional processing steps. Enhancing digital methods with deep learning can further improve AI-based AD accuracy, bringing it closer to the MD method.