AI Isn't Science Fiction: It's Digital Design's Reality

2021 has brought several promising technologies that will lead the way by altering and enhancing how dental laboratories and the dental industry as a whole will use technology in this digital world. Many of those promising technologies involve new CAD/CAM software capabilities, and one in particular should incite both excitement and concern among our profession: Artificial intelligence.

Several software manufacturers have introduced augmented reality capabilities that involve a level of artificial intelligence (AI), including exocad, Ivoclar Vivadent's IvoSmile, and others. Earlier this year, however, one manufacturer made a very exciting (and somewhat controversial) move when it unveiled a new platform that takes AI to the next level. 3Shape Automate (3Shape) is a revolutionary new technology that utilizes AI, machine learning, and a subset of machine learning called deep learning that utilizes neural networks in order to design crowns (Figure 1).

What do all of those words actually mean? A common misconception is that an AI and deep learning platform is just some enhanced auto-workflow or auto-crown algorithm. The intuition is to imagine that a programmer punched in some parameters that can be set by an end user and the software then attempts to make a crown based on those pre-set rules very quickly—much quicker than any human can. Although this is, indeed, a very valid method of creating a productive and streamlined workflow using an algorithm, it is not "true" artificial intelligence. What many do not realize is that there is no standard CAD software system on the other end of that portal at all.

What you will find at the other end of that portal is a data center up in the cloud, and within mere moments, your designs are ready to be reviewed for downloading. This is not just a computer that can design and click through CAD software quickly; it is a completely different animal altogether.

Deep learning and neural networks are not a new concept, but their application in dentistry is absolutely revolutionary. It is also very helpful that the dataset used for training these neural networks happens to be an ideal match.

In very basic terms, neural networks are a method of machine learning whereby a computer learns to perform a task based on training from existing examples. These samples are fed to the machine and have been identified and labeled in advance. The more data sets that get fed to the machine, the better the output result will be.

Now, many readers may be pausing here to say to themselves, "Dental anatomy and morphology are very complex and subjective; how does a machine learn to generate teeth well enough to create clinically acceptable crowns?"

On the surface, yes, teeth do come in many shapes and sizes. Furthermore, understanding anatomy and morphology well enough to create natural-looking restorations takes years of studying and experience. Even when we are armed with knowledge and experience, human judgement is inherently biased and our work always includes a level of subjectivity. Deep learning, conversely, is purely objective. It relies only on data, which results in very consistent outputs. So how does a machine interpret such complex data as teeth?

This technology is able to exist today thanks to a very simple primary shape that we all learned about in elementary school: the triangle. When we try to get our point across in layman's terms when dealing with computers, often you hear the phrase, "It's just a bunch of 1s and 0s." Although that is true, with deep learning for dentistry, the lowest common denominator is actually triangles (Figure 2).

The de facto standard file format in our industry today is STL. This format was created by 3D Systems¹ and consists of a mesh made up of many thousands of tiny triangles. (Fun fact: STL has several "backronyms," such as "standard triangle language" and "standard tessellation language"). Zooming into these meshes, you'll see thousands of small triangles grouped together (Figure 3).

Artificial intelligence is notoriously bad at solving problems involving subjectivity, rationality, and reasoning. If we strip away our own bias and look at an STL file not for the teeth, but as a machine would as a collection of triangles in a particular pattern, then we start to see and understand why artificial intelligence is so well suited for the task of designing crowns.

One single STL file means nothing to a machine, but a sample of hundreds of thousands or even millions of STL meshes is plenty of good data for the machine to digest. With this data set, the machine is able to look at these collections of triangular patterns and draw conclusions. It can determine what a collection of triangles in this particular spot (eg, a lower first premolar) should look like relative to this other set of patterned triangles (eg, a lower second premolar and lower canine). It then outputs a new collection of triangles in a mesh that will follow that specific pattern that it recognizes. The result is a crown that looks very good and fits very well (Figure 4).

This is AI in its truest sense. Drop a file in there, the machine interprets what tooth it is based on customer metadata and outputs what it has been taught, and it subsequently produces an STL file for machining. It is very fascinating and revolutionary technology.

Of course, technology can be weird and scary, and there will always be backlash whenever something new comes along. As much fanfare as this technology was met with, there was an equal amount of skepticism and fear. In an industry faced with a race to the bottom, increased competition from overseas, and a general looming uncertainty about the economy in general, it is understandable that many technicians would be opposed to a technology that could potentially replace them for a much cheaper price. At the moment, this technology is only available to dental laboratories and only for single molars in the posterior region. However, many technicians have voiced concerns over it simply being a matter of time before the flood gates ultimately open. That still remains to be seen and only time will tell, but there is no doubt that this technology is indeed disruptive in nature.

Beyond the novelty of "new and exciting" in the grand scheme of things, deep learning represents a very important milestone in dental technology. Streaming, and on-demand in general, has become the standard in so many segments of our increasingly online-and-connected world—whether with media consumption such as YouTube, Netflix, and Spotify; in business with Zoom or Dropbox; or in enterprise applications such as Amazon Web Services or Microsoft Azure. Everything is up in the cloud, streamed, and on-demand. Deep learning is another example of that on-demand, cloud-based type of service.

The ultimate question, of course, is: Does it work? 3Shape's service boasts a 92% acceptance rate with a sample size of more than 65,000 approved designs. In independent blind tests conducted, AI-designed crowns were consistently chosen over crowns designed by humans.

Human skill and knowledge likely will continue to have a place in the process, but this could certainly change the landscape of designing and digital dentistry in a big way. It is disruptive in nature and revolutionary in its implementation. Only time will tell exactly how significantly AI and deep learning will change the dental industry, and if it is for better or worse.

Minh Tran
Founder and Creative Director,
DentalTechTips

1. Park WJ, Park JB. History and application of artificial neural networks in dentistry. Eur J Dent. 2018;12(4):594-601. doi:10.4103/ejd.ejd_325_18

2. Chen, H., Zhang, K., Lyu, P. et al. A deep learning approach to automatic teeth detection and numbering based on object detection in dental periapical films. Sci Rep 9, 3840 (2019). https://doi.org/10.1038/s41598-019-40414-y

3. Stanley M, Paz AG, Miguel I, Coachman C. Fully digital workflow, integrating dental scan, smile design and CAD-CAM: case report. BMC Oral Health. 2018 Aug 7;18(1):134. doi: 10.1186/s12903-018-0597-0. PMID: 30086753; PMCID: PMC6081948.

4. Lerner H, Mouhyi J, Admakin O, Mangano F. Artificial intelligence in fixed implant prosthodontics: a retrospective study of 106 implant-supported monolithic zirconia crowns inserted in the posterior jaws of 90 patients. BMC Oral Health. 2020;20(1):80. Published 2020 Mar 19. doi:10.1186/s12903-020-1062-4

5. Vasamsetty, P.; Pss, T.; Kukkala, D.; Singamshetty, M.; Gajula, S. 3D printing in dentistry-Exploring the new horizons. Mater. Today 2020, 26, 838-841.

6. Pillai S, Upadhyay A, Khayambashi P, Farooq I, Sabri H, Tarar M, Lee KT, Harb I, Zhou S, Wang Y, Tran SD. Dental 3D-Printing: Transferring Art from the Laboratories to the Clinics. Polymers (Basel). 2021 Jan 4;13(1):157. doi: 10.3390/polym13010157. PMID: 33406617; PMCID: PMC7795531.