A cavity is a serious dental infection that needs to be treated quickly. An operation called dental preparation removes the infection and prepares the tooth to receive a crown that will protect it. However, because of the uniqueness of every human tooth, dental preparations often adjust poorly to the crown, which can lead to detachment of the crown or development of new cavities. To solve this problem, methods using artificial intelligence (AI) are now being created to generate more accurate dental preparations at lower cost, thereby improving dentistry and making it more accessible.

A Statistics Canada report says that some 96% of adults have already had at least one cavity, and will continue to have them throughout their lives. While some small cavities can be treated with a simple filling, others are deeper and require a dental crown to cover and protect the tooth. Before being fitted with a crown, the tooth must be “prepared,” by removing any decay. Because every tooth is unique, the preparation must be personalized. If a cavity is too deep or located between two teeth, dentists are faced with a difficult preparation choice, which they have to make quickly during a single visit. New AI-based methods now make it possible to simplify this choice by generating a customized dental preparation from a scan of the original tooth. This enables the dentist to identify the parts of the tooth that need to be removed adapt the preparation precisely to the specific tooth.

In practice, the parts of an original tooth that need to be removed to allow a crown to be fitted differ considerably from one tooth to another. Like a key and its lock, a dental crown must fit snugly on to the tooth preparation because a properly fitted crown will be more durable and less likely to fall out. However, the lack of quality standards and personalized procedures for dental preparations can lead to misalignment between the dentist’s preparation and the crown made by the dental technician. If the original preparation does not meet the retention requirements defined by the technician, the crown cannot be fitted and the dentist will have to trim the tooth further in a second procedure to adjust its shape to fit the crown. A wax mould, called a dental cap, will also have to be made by the  technician to help identify where more enamel needs to be removed. This trial-and-error process is considered inefficient by dental specialists, as it requires a wax cap to be made and increases the number of sessions the patient needs.

My research project, co-supervised by Professors Farida Cheriet and François Guibault at Polytechnique Montréal, aims to solve this issue by establishing an automated approach for validating and optimizing a dental preparation based on a 3D model of the original tooth. This validation during the operation will be done by comparing the dentist’s preparation with that generated by our algorithm. The system will automatically generate a 3D image of the preparation, which will be superimposed on the 3D model of the original tooth, enabling the dentist to quickly identify any areas where excess tooth needs to be trimmed. This superimposition will also be accompanied by a rendering of the inside of the original tooth, obtained by X-ray scanning or tomodensitometry, which will prevent the suggested preparation from removing too much tooth and impinging on the nerve.

To train the algorithm, we use a database containing hundreds of digital scans of the inside of the mouths of patients who have already undergone dental preparation to design a shape deformation model. Shape deformation is a sub-field of neural network learning that aims to automatically create a new shape—such as a tooth—from a series of existing shapes. Generating new shapes is similar to the workings of our own imagination. For example, based on its knowledge of geometric shapes, the human brain is able to imagine a sphere turning into a cube, even though it has never seen this specific series of images in the past. Similarly, AI can generate a new shape that it has never seen before, based on an amalgam of similar shapes. This generation of new information is also enhanced by access to a wider variety of data: the more different cases the algorithm sees, the better it will perform on complicated dental shapes. Thanks to a collaboration with Kerenor Dental Studio, a laboratory based in Montréal, our research team has access to a large number of dental preparations from a variety of real cases, enabling the algorithm to be trained more effectively.

In theory, this new AI-based method plays a very similar role to the wax dental cap currently used by specialists. However, our method is purely computer-based and does not involve any hardware design, which reduces production and labour costs. In terms of education, this tool will also give dental students the opportunity to practise their procedures on a wider variety of dental preparations. What’s more, since the results of the assessment will be available immediately with this new computational tool, adjustments to the dental preparation can be made in a single session, to patients’ great delight.