BRIEF INTRODUCTION TO DENTAL SCANNER TECHNOLOGY
Dental zirconia implants, —along with all the necessary appliances for their manufacture—, have been with us for several decades. During all this time, milling machines, sintering furnaces and processing software have evolved very quickly. But probably, the element that has undergone the greatest technological advance has been the dental scanner. The development of facial recognition and intelligent vision technologies, together with the important advances in hardware, has meant a complete revolution for this type of device in the last five years.
Unfortunately, many times the dental technician does not have sufficient knowledge and basis to judge the different scanner models present on the market. A well-known proverb says that we shouldn’t judge a book by its cover. Nor a scanner because its external appearance is more or less futuristic, or because of the design of its user interface.
This article tries to be a simple guide that clears the ideas from an exclusively technological perspective, leaving aside the dental practice. Its character is divulging, so the concepts have been simplified and the points that could be more difficult for an average dental technician to assimilate have been eliminated. An in-depth analysis of the operation of a modern scanner would involve entering the field of artificial vision, micro-electronics, advanced algorithms and advanced computer programming, among others. All of it, is beyond the scope of this article. We only hope that it can be a help tool for professionals interested in study deeper into the subject, or in objectively choosing a good scanner from the offer on the market.
1) Principle of operation
The principle of operation of any scanner is very simple. We all know the laser levels that are used in masonry and are capable of projecting a straight line (usually red color), on a surface. Now suppose that one of these levels projects the line onto a wall and we place any object in front of that same wall. When the line falls on the plane of the wall, it generates a perfect straight line. But when it is projected on the object, it acquires a warped shape that faithfully coincides with the silhouette of the object. Also, the brightness of each point will depend on its depth in space. Intuitively, it is seen that this warped line creates a generatrix of the outer surface of the object. If we now move the level from top to bottom, throughout the displacement infinite generatrix will be created with different shapes depending on the position of each surface point of the object.
With what has been said, we have almost finished the process. We only need to place two cameras in front of the scanned object to record the laser line when it moves. If the object, the cameras and the apparatus that projects the line are in known positions respect to a reference point, it will be very easy using trigonometry, to determine the position in space occupied by each of the points. Or, to put it another way: based on the images captured by the camera, a computer with the appropriate software can place on the space —depending on the silhouette of the object—, the points of all the lines generated during the movement of the laser line. That point cloud faithfully matches the surface of the object.
This operating principle has multitude of applications: the cameras and the laser projector can be installed in an airplane that sweeps a wide surface from the sky, generating a 3D map; or, by rotating the projector and cameras on the roof of an self-driving car to draw a 3D map with distances to any object at an angle of 360º; or on the head of a missile sweeping the surface over which it flies to detect obstacles or targets. Of course, these elements can also be placed on a frame that rotates dental models. The possibilities of this technology are limitless.
In any case, we always talk about three fundamental components that are the key of this technology: the projector, the cameras and the hardware/ software in charge of sort the information supplied by cameras. Without a doubt, software and hardware has the greatest hit on the final result. Let’s go on to analyze each of these components separately.
It is one of the key elements in a good scanning system. We can say without doubt that it has a big influence on the final result. In the previous point, —and for reasons of clarity in the exposition—, some simplifications have been made. In reality, scanner projectors do not use light similar to masonry laser levels. They also do not use a single scrolling line. Today the “red laser line” is used mostly in barcode reading. The principle of operation is identical to the scanner, but in barcode case we are working in 2D and. Moreover, with a pattern of lines with different sizes and only two colors: black and white. All of this is a huge simplification over modern 3D scanners.
Instead of the “red laser line”, today’s dental scanners use blue light projectors that project geometric patterns onto the object to be treated. The patterns are adjusted to the shape of the object and the cameras read them identifying points in space very quickly. If a single moving line of light is used, the time required for a scan would skyrocket and the accuracy would leave much to be desired.
There are many image projection systems on the market that can be implemented in scanners, but it is evident that the better the resolution of the geometric patterns projected on the object, the greater precision of the images captured by the cameras. Consequently, the precision of scanning is also higher.
When we talk about extremely precise projection systems, we should talk about the DLP® technology developed by the American firm Texas Instruments. There is a resounding consensus within imaging specialists to point out it as, —now days, autumn 2020—, the queen technology for very high-resolution projection systems. Without extending too much, we will say that DLP® technology has a matrix of tiny mirrors with an approximate size of about 5.4 um. Some of these matrices can contain up to 8 million points. Each of these mirrors is coated with nanosilver, a material that is practically perfect for light reflection. In addition, the micromirrors can be oriented independently and synchronously, rotating at angles of up to +/- 17º on their base.
When a light beam falls on the micro-mirror, it reflects it towards the projection surface. The projected light source can be of any type: visible frequencies, infrared, RGB color range, ultraviolet, etc… It must be remark that moving millions of micromirrors independently at a reasonable speed and at variable angles is not an easy task. However, the final result of the projection regarding quality and sharpness is unmatched.
Normally, blue light is used as a light source in dental scanner projectors because it has the advantage that it projects finer geometries, (greater precision) and is more immune to object glare, transparencies and imperfections or shadows. Blue light is more coherent in space (ability of a light beam to remain confined into a small space when it moves through the space); and in time, (Ability of a light beam to maintain its frequency range constant regardless of external conditions). Another advantage is it is not concentrated; therefore, the blue light is safer for the eyes.
3) The cameras
Camera is another decisive element for a quality end result. In addition, it marks significant differences between the different models of dental scanners present on the market. When analyzing a camera, we can basically talk about two key elements: the optics, which would be the set of lenses present in the objective; and the sensor, which would be the element in charge of capturing the light.
Starting with the optics and if we take into account that in a dental scanner the lighting conditions are optimal due to the action of the projector and the position and distance of the object to be scanned is known (constant focal distance), with the proper reserve we could say that, even though it is a decisive element, the optical assembly has a relative practical impact on the final result of scanning dental models.
Intuition tells us that it will not always be this way. In the case of an airplane that flies over an extensive territory to obtain 3D maps, the variable lighting conditions, the different focal distance depending on the speed and altitude of the airplane, the vibrations produced during the flight, the infinity of landscape colors, makes us suspect that in this case the optics will be decisive for a good scan. The same could be said about a self-driving car which tries to detect a moving pedestrian on the stripes of a zebra crossing at a distance of 20 meters on a rainy night.
The second of the constructive elements in a scanner camera is the sensor. It is a plate covered with photosensitive material where the light falls after passing through the optics. The photosensitive material is divided into tiny dots called “Pixels”. Each pixel is independently capable of measuring the intensity and frequency (color) of the light that falls on it. The quality of this element is essential to achieve an optimal scan. To estimate it, we must attend to the following points:
- Number of Pixels– Obviously, the more points a sensor has, the greater precision, because the information collected is also greater.
- Surface of each pixel– The bigger the pixel, the greater the amount of light that falls on it, and therefore, the information will be much better nuanced.
- Information dump technology-. The information read by the pixel must be subsequently transferred to the computer so that it can be processed. To carry out this task there are two procedures:
- Rolling shutter– The set of pixels is divided into lines or blocks and the information is transferred by blocks. The computer does not have the value of all pixels (complete image) until the last block is transferred.
- Global shutter– The dump of the value of the pixels is carried out in a single step, the transfer of information is instantaneous made.
- Number of images per second. We know that film and TV images are a succession of “photographs” that change very quickly. From 24 images per second, the human brain is unable to appreciate jumps, visualizing the image as a continuous movement. The vast majority of dental scanners work with cameras at 30 fps (Images per second). Only a few manufacturers dare with 60 fps. Synchronously combined a DLP® projector with a 60fps capture camera leads to a simply perfect image quality.
- Image threading technology: In the case of systems that work with two cameras, it will be necessary to combine the 2D information from the two images to get a single 3D image. During “stitching” of images, the resolution doubles due to the sum of information delivered by each of the cameras. Thus, a set of two cameras of, let’s say 1.3 Mpixels, will deliver final images close to 2.6 Mp. For technical reasons complicated to explain, the exact arithmetic sum will never be reached.
As a curiosity we will say that, actually to perform a 3D scan, only a camera placed at a fixed angle respect to the projector position is enough, as long as the object to be scanned rotates on one of its polar axes. The reasons why two cameras are placed instead one are the increased of precision; flexibility in the distances and position between object, cameras and projector; The flexibility in the movement of the object; and the highest speed in capturing points.
Continuing our tour with the different elements that make up a modern dental scanner, we come to the hardware. We understand by hardware the set of components, chips and electronic circuitry that represent the physical and tangible part of the system. Without doubt, the real heart of the hardware is the microprocessor. This small element is the brain that executes all the calculations and logical operations to put in order and later display on the screen in an understandable way, the enormous amounts of information that the cameras capture.
Depending on each dental scanner, we can talk about systems that use one, two and in more exceptional cases, up to three different microprocessors working in parallel and exchanging information and processes in real time. In this way, we can make a quick classification of the microprocessors associated with a dental scanner based on their type:
- Central processing microprocessor (CPU): It is located on the PC on which the main scanning software runs. It is easy to identify in dental scanners because it is the PC where the monitor, keyboard, mouse, etc. are connected.
In microelectronics, intelligent systems are often classified as “Masters” and “Slaves”. According to this classification, the Masters deliver information and instructions to the slaves and they execute it. When a slave finishes its task, the result is returned to the Master. The Master is on charge of the process and their respective Slaves must collaborate with him following his instructions. In our case, the PC’s microprocessor would act as the master of the system.
- Microprocessors with FPGA technology– Normally, the PC-class desktop computers that we use, have a generalist microprocessor. They have been designed to perform a lot of different tasks: from a simple mathematical calculation, to viewing an internet video, or searching information in a database. This generality forces to create a complex and heavy internal architecture, in addition to an extensive set of assembler instructions, that matches to the needs of the multitude of programming languages currently used by computer technicians. The result is microprocessors that are easy to program and fits into a multitude of different tasks, but also very lazy in certain tasks.
In front of the generalist microprocessors there is another type: the dedicated ones. These are designed with the right blocks and instruction set to perform a few very specific tasks. Its processing speed and efficiency is more much higher than a generalist microprocessor. In the past, dedicated microprocessors had to be manufactured for a very specific application. However, FPGA (Field Programmable Gate Array) technology allows the engineer himself to design his microprocessor with custom blocks and develop the corresponding set of specific instructions.
To be honest, FPGA technology is any sane engineer’s nightmare. It involves a massive amount of design-only work. As if that were not enough, the source program must be written in an assembler developed specifically for this task. A real torture compared to modern high-level programming languages. As a counterpart to such high effort, the speed and precision in data processing has no comparison with the obtained by working with the generalist microprocessor of the PC computer.
In the case of dental scanners, very few manufacturers dare with this technology. Normally these few manufacturers use it to create a dedicated microprocessor built-in the scanner and which is responsible for very specific tasks; such as managing the information delivered by the cameras; sew 2D images to get 3D reproduction; storing the information in a battery of DDRIII RAM memories associated with the FPGA; bidirectional communication with the central microprocessor; coordinate projector shots and motor spin, etc. Using this technology, the central PC (Master) is freed from all these tasks that are performed by its FPGA (Slave). The increase in processing speed is spectacular.
- Graphics microprocessor (GPU) Years ago, computer graphics cards were more or less “Dummies” elements, which were limited to draw on screen the images provided by the PC’s microprocessor. Due to the sophistication of video games, the important multimedia component of the Internet, and the increasing demand for monitors (4K and now 8K), modern graphic cards and their specific microprocessors (GPU’s), have evolved to become in real computers, with a high computing power and a specialization unimaginable just a few years ago.
Basically, we have two ways of making the PC’s microprocessor (Master) interact with the graphic microprocessor (GPU):
- The Master-Slave relationship between graphics card and PC is done using the general protocols and instructions of the graphics GPU. This case is the most common of all because it simplifies programming a lot. However, the speed, performs and flexibility achieved is reduced, because the two microprocessors (CPU / GPU) interact following the protocol and instructions of the graphics card manufacturer, which is usually very generic to adapt into multitude different situations.
- The Master-Slave relationship between graphics card and computer is done using protocols and instructions developed for a specific task. In this other case, a series of algorithms and specific instructions are designed to be processed by the GPU. In some way, the programmer uses the computing power of the graphics card (GPU) on his own advantage, using instructions and protocols development by him, to customize the tasks they deem appropriate. Custom software runs on the GPU. For this arrangement, it will be necessary to select a specific family of GPU’s. The software will not work properly if it is used in a different family.
Last item to be analyzed in a dental scanner is software. If we look at the part of the software that, for practical purposes, most affects resolution, speed and user experience, we must certainly talk about algorithms. For less initiated readers we will say that an algorithm is an ordered sequence of steps or instructions aimed to solving a specific problem. As an example, we could propose an algorithm to add 2 + 2. Obviously, this algorithm has no practical interest, but it is very simple and illustrative. To solve the proposed problem there are different algorithms:
- Take a piece of chalk; make two marks on the wall; then make two more marks; Count the total number of marks on the wall.
- Turn on a desktop calculator; Press key “2”; Press the “+” key; press key “2” again; Press the “=” symbol;
- Address to the person next to you; smile; ask him: “Can you tell me how many are 2 + 2“?
- Take the first number and rise to the Third power; multiply it by 2; take the square root; that is the result.
As can be seen, there are many algorithms to solve the same problem. However, we must underline two important things:
- Some algorithms are simpler than others. It’s about combining speed, accuracy, resolving power, and simplicity. Depending on the nature of each algorithm, results more or less satisfactory, will be achieved.
- Good algorithms generally belong to the know-how of each company or programmer. There are free algorithms that are accessible to any programmer. However, the most powerful and sophisticated are rarely shared. This means that the quality of the software results will depend a lot on the quality of the algorithms used by the programming team.
When we talk about intelligent vision in general and more specifically about dental scanners, we should consider two basic families of algorithms:
- Algorithm “Cluster or Hough transform” – Developed in the 1960’s by mathematician Paul Hough, it aims to find geometric shapes that fit a mathematical equation (lines, circles, ellipses, etc.) within a point cloud. A classic example can be the orbits of planets or comets. As Kepler stated in his second law, the stars and comets orbit in elliptical paths. If we have a sequence of photographs of the sky and we superimpose them, the positions of a certain comet are located in an ellipse. Hough’s algorithm is programmed in this case to identify ellipses in clouds of celestial points and, consequently, trajectories of celestial bodies.
Volcanologists try to identify, in aerial photographs of the earth, the lines or curves that connect the calderas of different volcanoes and the points where earthquakes occur. It is intended to find the lines that identify the tectonic faults to be able to predict earthquakes.
The above two examples are very simple. If we complicate the algorithm, we could consider identifying much more complex curves, such as: within a facial recognition system, finding in a cloud of points (face silhouette), the line that joins the two pupils of the eyes; the equation of the curve that draws the eye socket; or the position of the sagittal plane. We could also try to identify the characteristic curve drawn by the turret of a war tank in a battlefield… It is intuited that the level of refinement achieved now days by Hough’s algorithms is spectacular and besides there are many variants in terms of quality and precision.
- Algorithm “FPFH (Fast Point Feature Histograms)” also known as Feature Extracting. Contrary to Hough’s algorithm that identifies curves with mathematical equations defined in a point cloud, the FPFH algorithm is in charge of comparing a given point with all those around it. A possible example could be the basting of different separate images, to create a panoramic photograph: we would start by taking a singular point in both images, —usually an edge, a corner, or a certain reference—, to later analyze the characteristics of the points that surround them. When the surrounding points are the same in the two images, we can sew them to create a consolidate one.
In the case of facial recognition, if we have a database with significant points of people’s faces previously identified by Hough algorithms, we can compare them with a specific image using FPFH algorithms to recognize an individual person. The same can be said of the significant points of the cloud of points that form the image of a war tank on a battlefield.
If we now go down to the level of dental scanners, starting from a dental model we will have to identify the contact points between teeth and molars of the upper and lower jaw; and the meshing contours between valleys and cusps. To do this, we will compare the point clouds that surround each significant point belonging to the two teeth arches.
Depending on the quality of the algorithms used, the result will not be the same when we compare two different dental scanners. As a guide, the indications of a software that has excellent quality intelligent vision algorithms are:
- Speed with which the fitting procedure is performed. The faster the better the quality of the algorithms.
- The precision of the fitting: Understood as the accuracy with which both jaws matches each other.
- The high rate of success in the matching. Poor algorithms lead to few successful matchings. In many cases, they end up making manual match.
Apart from those mentioned, another indication of the excellent quality of algorithms of intelligent vision software which has the ability to read textures: specifically curves drawn with pencil that are later identified by the scanner and treated with the dental modeling program. If we take it to the field of facial recognition, the operation would be equivalent to identifying a mark, mole, scar or tattoo within the cloud of points associated with a face. Or certain distinctive in the silhouette of a battle tank if we talk about military applications.
Identifying these textures on the point cloud and later associating them with curves with mathematical equations requires giving the algorithms a further twist. We would speak of state-of-the-art technology available only for few programmers. This is the reason that very few dental scanner manufacturers currently have it.
Throughout this brief introduction we have tried to show in a simple and simplified way the most significant points that determine the quality of a dental scanner. Basically, and as a summary, we can group them into three categories:
Aspects that improve scanning speed
The main characteristic that affects speed is a Master/ Slave type architecture in which the main computer (PC) sends tasks to two slaves: an FPGA built-in the same scanner and the GPU in the graphics card. With this arrangement we can launch processes in parallel by significantly speed increasing. Also, the use of state-of-the-art algorithms is an excellent help to improve speed.
A communication through custom language and algorithms between the GPU and the computer also helps, as well as the Globe Shutter technology to dump the information read by the cameras. Finally, the use of a USB type 3.0 port involves multiplying the speed of information exchange between scanner and computer by eight.
Issues related to scan resolution
Regarding the resolution, we must point out the importance of having a DLP® technology projector combined with cameras of at least 1.3 Mpixels of resolution and a sequence of at least 60 Fps. If we add pixel size of at least 5.6 um to this, the resolution is greatly optimized. The possibility of direct communication between the central computer and the GPU through specific algorithms, is very beneficial to improve the resolution of the scanner too.
Aspects that condition the user experience
A software with advanced algorithms to fit elements quickly and intelligently (dental models, scanbodies, etc.). Utilities that allow to read textures such as lines drawn in pencil or the possibility of being able to scan at the same time and more comfortably two or more superimposed abutments, are examples of functionalities that indicate the good quality of a scanner. The user’s experience increases.
There is no doubt that one of the fields of technology that has undergone the greatest changes in recent years is that of smart vision. The strong irruption of facial recognition, the impact of the latest generation video games, 4K and 8K resolutions, multimedia, military applications and the identification of products in industrial processes; has provoke this branch of engineering to be immersed in a deep revolution in just last five years.
However, the vast majority of dental scanners on the market today were conceptually designed more than 10 years ago. An eternity when you consider the breakneck speed with which intelligent vision advances. Despite the great effort of scanner manufacturers to try to keep their designs alive by periodically incorporating small improvements that can be perceived by the user, actually many of them do not have a state-of-the-art concept, which allows real improvements: a Volkswagen Beetle can be “upgraded” by adapting a modern sound system, an air conditioning system, or electric windows. However, its chassis, aerodynamics, and rear engine / traction arrangement, pose heavy handicap when compared to more current concept products.
The development of a new hardware/ software that can incorporate the latest generation of advances in intelligent vision is a great effort for any company. It will probably be difficult to assume until they have paid off the investment made a few years ago to develop his current product. Not to mention the obvious risk involved in working with high technology: changes occur very quickly and sometimes it is not clear whether after all that effort satisfactory results will be achieved.
Perhaps for this reason, it is not difficult to find products that, sometimes under a modern and advanced external appearance, hides a totally obsolete hardware/ software design. We hope that by reading this article dental professionals will have a clearer idea and sufficient elements of judgment to objectively qualify the quality and adequacy of a dental scanner.