Clinical Evaluation of AI-Generated Dental Crowns

Artificial Intelligence (AI) is the science and engineering of machines that act intelligently (1). The Oxford Dictionary defines AI as the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision making, and translation between languages (2). There are many ways AI can be achieved, the most important among them are 1) Machine learning: It is a method where the target is defined and the steps to reach that target is learned by the machine itself by training (gaining experience); 2) Natural language processing, for example, Siri and Google assistant. 3) Computer vision, for example, tesla Autopilot. Many fields have already benefited from AI. In medical field, AI has already been implemented in various medical fields in diagnosis such as diabetic retinopathy, skin cancer and breast lesions (3). In dentistry, most of the application goes to the automatic diagnosis based on CT and radiology images (4).

Digital workflow has become an overwhelming trend in dentistry motivated by the prevalence of intraoral scanners (IOS) and computer aided design and computer aided manufacturing (CAD/CAM). Compared with traditional laboratory methods, which are regarded as time-consuming and technically sensitive, the digital workflow can be greatly convenient and efficient (5). Thus, CAD/CAM facilitates the opportunity for improving the productivity of dental prosthesis.

Current digital workflow consists of four basic elements: 1) tooth preparation and data acquisition (via intraoral scanner, x-ray, CBCT, etc.), 2) data processing and prosthesis design (via CAD), 3) prosthesis fabrication (either laboratory or chairside milling via CAM), and 4) try-in and cementation in the clinic (by the dentist). Despite all the advancements such as the elimination of physical models and labour-saving, many problems still exist in the current workflow. Each dental prosthesis must be customized to meet individual patients' condition and requirement. Designing dental restoration must be conducted and approved by the technician; this is a time-consuming and labour-intensive process even with the assistance of CAD software. In particular, the wrong design in CAD process makes the crowns that can induce major oral problems of: 1) Superocclusion, 2) Infraocclusion, and 3) Overcontour. This said, CAD/CAM does not save a lot of the dentists' and patients' time and cost as advocated. Therefore, there is a need to change the current practice of dental CAD/CAM.

In view of this, with the support of GRF, we have developed a fully automatic algorithm for the design of dental prosthesis by utilizing AI technology. The algorithm was based on two aspects: 1) utilization of the current dental knowledge by learning the materials-human interactions and materials-biomaterials properties to automate the prosthetic design; 2) based on the previous clinically relevant studies, to validate the design from finite element analysis (FEA) results. With the 3D digital dental prosthesis dataset obtained from Prince Philip Dental Hospital, Faculty of Dentistry, The University of Hong Kong, Generative Adversarial Network (GAN) was adopted to train the machine learning model on the design of dental prosthesis. It composed of two deep networks, the generator, and the discriminator. The discriminator could identify the tiny difference between the real and the generated designs, and the generator could create the designs that discriminator cannot tell the difference. Finally, the GAN model converges and produces natural look designs of prosthesis. Afterwards, an AI-enabled FEA algorithm was established in order to achieve the accurate and fast FEA of dental prosthesis. Stress concentration on the prepared tooth and prosthesis, a common cause of the failure, may result from flawed prosthesis design. Based on our published FEA data (6, 7), a validation model was built mainly to detect and correct the errors of design which may cause stress concentration. This FEA machine learning model also served as one of the criteria on evaluating the quality of automatic generated prosthesis.

After the training via GAN and machine learning model, the automatic prosthesis design algorithm needs to be validated by means of mechanical tests in the laboratory and application in clinical practice. Cyclic fatigue is prone to cause failure from stress concentration areas or loading contact points; however, it is hard to be detected by technicians directly (8, 9). In in-vitro validation, specimens were subject to cyclic loading using the Instron universal testing machine (Electro Puls E3000, Instron, Norwood, USA), then failure mode analysis and scanning electron microscopy (SEM) were conducted. Comparable fatigue properties of the automatically designed prosthesis to that of CAD/CAM prosthesis have been confirmed (7, 10).

Therefore, several clinically relevant parameters, such as anatomical form, marginal adaptation, wear behaviour of the restoration and the antagonist, and integrity of the restoration and the abutment tooth, are aimed to be evaluated clinically using World Dental Federation (FDI) criteria (11). Compared with the modified USPHS criteria, FDI criteria may give more sensitive results in relatively short-term clinical trials, as it has more scoping options (12). The criteria can be categorized into aesthetic parameters (4 items), functional parameters (6 items) and biological parameters (6 items). In this study, 10 items are selected as they are relevant to the design procedure of the prosthesis. Data collection will be accomplished using oral examination and grading and IOS.

In evaluation of the amount of wear, digital impressions captured by IOS can be superimposed and analysed directly in the software. The replica is no longer needed, the data capture procedure is simplified, so the error can be reduced. IOS has been utilized in several clinical assessments and its accuracy has been confirmed (15).

This proposal aims at validating prosthesis design by the fully automatic algorithm both clinical side and also to further optimize the algorithm. Restorations designed by the algorithm will be comprehensively evaluated according to the FDI criteria while the CAD/CAM prostheses designed by technicians using ordinary computer-aided design software serve as the control group.

Reference:

1. Norvig P. Artificial intelligence: Early ambitions. New Scientist. 2012;216(2889):ii-iii.
2. OxfordUniversityPress. English Oxford Dictionaries. Oxford University Press; 2019.
3. He J, Baxter SL, Xu J, Xu J, Zhou X, Zhang K. The practical implementation of artificial intelligence technologies in medicine. Nature medicine. 2019;25(1):30.
4. Hwang J-J, Jung Y-H, Cho B-H, Heo M-S. An overview of deep learning in the field of dentistry. 2019.
5. Li RW, Chow TW, Matinlinna JP. Ceramic dental biomaterials and CAD/CAM technology: state of the art. J Prosthodont Res. 2014;58(4):208-16.
6. Maghami E, Homaei E, Farhangdoost K, Pow EHN, Matinlinna JP, Tsoi JK-H. Effect of preparation design for all-ceramic restoration on maxillary premolar: a 3D finite element study. Journal of prosthodontic research. 2018;62(4):436-42.
7. Homaei E, Jin X-Z, Pow EHN, Matinlinna JP, Tsoi JK-H, Farhangdoost K. Numerical fatigue analysis of premolars restored by CAD/CAM ceramic crowns. Dental Materials. 2018;34(7):e149-e57.
8. Keulemans F, Palav P, Aboushelib MM, van Dalen A, Kleverlaan CJ, Feilzer AJ. Fracture strength and fatigue resistance of dental resin-based composites. Dent Mater. 2009;25(11):1433-41.
9. Zhang Z, Guazzato M, Sornsuwan T, Scherrer SS, Rungsiyakull C, Li W, et al. Thermally induced fracture for core-veneered dental ceramic structures. Acta Biomater. 2013;9(9):8394-402.
10. Homaei E, Pow EHN, Matinlinna JP, Akbari M, Tsoi JK-H. Fatigue resistance of monolithic CAD/CAM ceramic crowns on human premolars. Ceramics International. 2016;42(14):15709-17.
11. Hickel R, Peschke A, Tyas M, Mjor I, Bayne S, Peters M, et al. FDI World Dental Federation: clinical criteria for the evaluation of direct and indirect restorations-update and clinical examples. Clin Oral Investig. 2010;14(4):349-66.
12. Cakir NN, Demirbuga S. The effect of five different universal adhesives on the clinical success of class I restorations: 24-month clinical follow-up. Clin Oral Investig. 2019;23(6):2767-76.
13. Sinescu C, Negrutiu M, Topala F, Ionita C, Negru R, Fabriky M, et al. Ceramic and Polymeric Dental Onlays Evaluated by Photo elasticity, Optical Coherence Tomography and Micro Computed Tomography. Proc Spie. 2011;8172.
14. Fujita R, Komada W, Nozaki K, Miura H. Measurement of the remaining dentin thickness using optical coherence tomography for crown preparation. Dent Mater J. 2014;33(3):355-62.
15. Aladag A, Oguz D, Comlekoglu ME, Akan E. In vivo wear determination of novel CAD/CAM ceramic crowns by using 3D alignment. J Adv Prosthodont. 2019;11(2):120-7.