We develop artificial intelligence (AI) systems based on machine learning – specifically, deep neural networks. Not because it is the trend, but because this technology helps us to efficiently accomplish a certain class of production tasks we used to face.
At the same time, it should be noted that in the field of three-dimensional surface processing based on machine learning, there are still no off-the-shelf solutions that could be combined to solve an applied problem. We are at the forefront of this research and development.
Below are just some of the most vivid examples of what we can do using AI Machine Learning:
We use neural networks to segment a jaw scan for purposes of definition, recognition and identification of teeth. That is, we need not only to locate the surface of the gums and the teeth, but also to correctly identify each tooth by assigning it an ID as per standards.
Solving this problem by geometric and heuristic techniques only is very difficult. At the same time, our research shows that a machine-trained AI alone is not equal to the task of segmentation, which is done better by combining an AI with geometric and heuristic techniques.
Therefore, we perform the segmentation task very efficiently by using machine-trained AI in combination with geometric algorithms.
Automatic crown generation
Another important task is the automatic generation of a dental crown. The challenge here is that anatomical, technological and aesthetic requirements must be met all at the same time. An artificial tooth should not only be perfect by itself, but should also fit perfectly into its environment – it should mesh in with its counterpart, blend with its dental neighbors, slot into its spot with a clearance of ten microns, leaving no irregularity or backlash, be manufacturable, and fittable into the jaw of a real-life patient. And last but not least, it must be easy on the eye when revealed in a smile.
In practical dentistry, if a crown is made manually by a prosthetic technician, there will be as many “perfect” crown designs for each specific case as there are technicians involved. It is no exaggeration to say that the dental crown design is an art. And this is actually a problem!
If there are more than one “perfect” solution, then geometric algorithms and heuristic intelligent systems are not the way to go. And this is where machine learning-based AI comes to the rescue.
We were the first to start using generative-competitive neural networks (GAN) for real-life dental production, for actual treatment of people. In solving this problem early on, we very successfully collaborated with the University of California, Berkeley. As a result of further research, we constructed a specific neural network architecture and trained it on millions of restorations made by Glidewell’s experienced dental technicians. GAN is now successfully using it to mass produce dental crowns.
Other AI utilization
Having accumulated significant experience in the application of machine learning in the dental field, we use this powerful tool in many tasks involving high uncertainty and variability – for example, to quickly and efficiently distinguish between bad and good jaw scans and dental restorations, improve the accuracy of surface reconstruction during CT scanning, build the best interface between the crown and the tooth preparation, etc.