Artificial intelligence and machine learning in orthodontics

Abstract

Artificial intelligence (AI) has had an enormous impact on medical and life sciences such as bioinformatics and genomics.1 With the recent acceleration of digitalization and advances in medical device technology, the amount of clinical data is also expanding, and there is a huge demand for support in the analysis of these data. AI and machine learning could enable processing of these expanded data and reduce our efforts in this process.2 These data can be collected efficiently with further support from advanced data transfer and data storage systems. Consequently, accumulated big clinical data are expected to provide the new findings with the help of the AI using current data science technology.3 Orthodontics, by nature, has applied information technology in practice. Since the 1920s, orthodontists have utilized standardized cephalography and converted shape information into a small set of quantifiable indices.4, 5 Various dental/skeletal parameters are also extracted from dental plaster models. These quantifiable data are further collected and processed, enabling orthodontists to describe malocclusion using pre-defined objectives and standards.6 Converting such varied biometric information into quantifiable parameters, integrating these data and supporting treatment decisions are analogous to the current IT analytics. In other words, orthodontists have already adopted digital thinking in their diagnostic and treatment systems since the analogue era. With the advent of CBCT,7 the acquisition and usage of 3D digital information has been widely adopted. Recently, 3D cameras and intraoral scanners have enabled the easy digital capture of the face and dental arch.8 Although the hardware for digitization of analytical subjects has progressed, data processing remains an obsolete, which is difficult for a clinician to manage the exceedingly huge data. For example, landmark identification is necessary for a 3D morphological analysis. AI is expected to automate this process, which would consequently reduce the effort required for manual identification and activate the use of digital data in orthodontic clinics and research. In modern medicine, treatment evidence is supported by traditional mission-driven clinical studies. However, this style of analysis is associated with several concerns when applied in the validation of treatment efficiency.9 In orthodontic research, there is enormous diversity in samples and treatment modalities, and it is possible to underestimate treatment efficiency. In the near future, big data will be further utilized with the advanced support of data transfer and data storage systems, and clinical operation and research will become more data-driven.10 The mining of big data using AI is expected to yield novel findings and lead to a paradigm shift in diagnosis and treatment planning in future. In this AI and machine learning special issue, we have included 21 articles. Many of these studies have gone beyond machine learning and most used deep learning approaches. Among them, three review articles were included. Monill-González et aland Gili et al review the current status of the application of AI in orthodontic clinics and provide perspectives on this technology.11, 12 AI Turkestani et al reviews the data science approaches for clinical decision support systems in orthodontics and also introduce a web-based data management platform.13 Cephalometric analyses are performed by plotting reference points manually; AI can automate this process. Several commercial services have already been launched; however, such models lack sufficient scientific and theoretical support. In this special issue, three research groups of Bulatova et al Tanikawa et al and Kim et al evaluated the accuracy of automatic landmark identification on cephalograms.14-17 All groups reported that the accuracy of AI detection is acceptable but that there is still room for improvement. Actually, the accuracy is affected by the quality of the cephalometry, and Tanikawa et al also evaluated the number of cephalograms needed by the AI to re-learn for different quality images when an AI system is introduced in a clinic.16 Two groups of Kim et al and Kök et al evaluated the automated diagnosis of skeletal maturation using cervical vertebrae images on cephalometry and that AI-based decision supporting is available for evaluating the optimal timing of treatment for growing orthodontic patients.18, 19 Both groups also found that that additional patient information increased the accuracy of the prediction. Computer vision technology has been already widely used in facial recognition. In this issue, two groups evaluated facial morphology. Yurdakurban et al developed a method of automatic detection of the facial midline using AI.20 Rousseau et al trained AI to detect morphological differences in Osteogenesis imperfecta (OI) types. This method could be applied to the detection of the facial morphology of other syndromes.21 Segmentation of CT images is essential for the quantification and printing of three-dimensional (3D) images for the diagnosis of patients and treatment planning. Delineating the craniofacial structures is challenging due to their complex morphology, and manual annotations have limited reproducibility and are very time-consuming. AI and machine learning help us by providing automatic image segmentation, which reduces the burden of manual handling with improved reliability. In this issue, Lo Giudice et al developed a method that enables the automatic segmentation of the mandible.22 Wang et al evaluated reconstructed 3D images of the maxillary bones of cleft lip and palate (CL/P) patients and described their defective phenotypes.23 Sin et al applied this AI process in the delineation of the airway morphology.24 Niu et al aslo evaluated the nasal cavity and pharyngeal airway in the Rapid palatal expansion treatment.25 The shape of the palate and the dental arch varies among the patients with malocclusion and is influenced by various genetic and environmental factors. Although essential for the success of AI-based diagnosis, the objective extraction of these morphological features is not easy. Nauwelaers et al developed a novel AI-based method for the analysis of palatal 3D shape26 and Croquet et al developed methods of automatic landmarking for analysis of the palatal shape.27 Machine/deep learning is a powerful tool for assessment of important clinical parameters and the extraction of clinical knowledge from complex patient information and data that affects the treatment outcomes. Lim et al developed comprehensive predictive models to determine prognostic factors and actually evaluated the need for orthognathic surgery (OGS) in mixed dentition patients using CL/P by AI systems.28 Auconi et al applied case-based reasoning in predicting the progression of Class III malocclusion.29 Di Carlo et al applied a network analysis to verify the possible association between the orthodontic features and upper airway characteristics and found the most closely connected features.30 The decision on whether or not to perform extraction is one of the most debated issues in clinical orthodontics. This decision is not made simply based on morphological dental and/or skeletal problems, rather it is influenced by various factors. Rongo et al shows that a 3D stereophotogrammetric analysis allows for a more comprehensive assessment of the facial soft tissue effects by extraction and non-extraction treatment.31 Etemad et al applied AI in this extraction issue and demonstrated that a denoise machine learning technique could improve the prediction for a diverse patient population.32 We sincerely appreciate the opportunity to serve as the guest editors for this issue and wish to thank the editor-in-chief, Dr Ambra Michelotti and Wiley.

Ähnliche Arbeiten