.jpg)
We are Relu. A young and enthusiastic team with a ‘work hard, meme hard’ mentality. Our mission: to automate digital dental treatment planning. We do this by creating AI-driven software to enhance the treatment planning process. Simple, right?
Based on the confused looks we get from our friends and family, this is not as simple as we think.
You may wonder: ‘Dentists are just the people that check whether we brush our teeth twice a day, so why would they need the software Relu is making? To clarify, the dental industry covers more than just dentists and orthodontists. In fact, there are about 20 subspecialties. So when we refer to the dental industry, we are also talking about various subspecialties that can benefit from our AI such as those focused on treating missing teeth, broken teeth, crooked teeth and even fractured jaws.
So without further ado, it is time to dive into Relu! We will explain what we actually do, how it involves the dental industry and how we use AI to do it.
Let’s use some analogies to bring the work of Relu to life.
Dentists: the architects of your mouth
Building a home is a lot of work. It is often expensive, painful and takes longer than expected. But imagine how much worse it would be if you started building your dream home without any prior planning: no drawings, no architect, and no contractor. It would probably result in a house that looks similar to a 3-year-old’s Lego structure that was partially destroyed when mom or dad tripped over it at 4AM.
So if you want to accomplish your dream home, it is best to plan it first. This also applies to achieving your perfect smile.
A dental practitioner will not start a procedure or treatment without having a plan, and like any architect, they are going to use planning software to optimize the process.
I still don’t get it…
Patience, reader! We promise, it will become clear.
A lot of architects still like to begin with freehand sketches before bringing them to life using computer-aided design (CAD) software. Instead of sketches, dentists use patient scans to get started. They can use a few types of scans depending on what they need to do: x-rays, intra-oral (IO) scans and/or cone-beam computed tomography (CBCT) scans. We are sure you know what an x-ray is, but are you familiar with CBCT and IO scans?
Dentists basically get to play ‘Harry Potter’ when they use an IO scanner. These handheld devices are like wands a dentist uses to create a digital impression of a person’s mouth. They just stick it in the mouth, move it over the teeth and then a nice 3D model pops up on the screen.
And have you ever seen a CBCT scanner? It looks like a piece of fitness equipment you’d find on a space ship.
Unfortunately, your biceps will remain as undefined after your scan as they were before it. CBCT scans are kind of like x-rays, but instead of a 2D image, you get a more detailed 3D image. Unlike x-rays or IO scans, you get an overall view of the entire mouth, jaw and other surrounding structures.
Okay…
Software for designing a home is streamlined with features such as floor plan creation, adding/customizing architectural features and measuring/resizing project elements. Dental treatment planning software does something similar.
Depending on the type of treatment, the software will include specific feature sets to do things like moving teeth, placing prosthetics or measuring structures. It is undeniable that treatment planning software has improved patient care, offering more efficiency and reducing risk. However, these software are still not optimal. They require a lot of manual processes, lack visualizations, and are often not ideal for easy communication between clinicians and patients.
Put the “treat” back in treatment planning
This is where Relu comes in, dressed in a superhero-worthy cape/tights combo. Relu creates AI software that automates some of the time-consuming manual processes performed in treatment planning software and offers a full 3D model of the actual patient.
I think I’m starting to get it, any examples?
Orthodontics has come a long way. When some of us were young, metal braces were the only option. Nowadays, an orthodontist can also use clear aligners to straighten your teeth and fix bite issues. If you aren’t familiar with clear aligners, the concept is simple. They are a series of removable, transparent plastic replicas of your teeth which include slight variations to guide teeth to their ideal position. A majority of
treatments will require 14-26 aligners.
To create a treatment plan, the orthodontist will use their IO scanner to get a 3D view of your mouth, which will load directly into the treatment planning software. Next, they will reposition your digital teeth to design your ideal smile. Once you are both satisfied with the final smile design, they send the treatment plan to a lab run by clear aligner elves (3rd cousins twice removed from Santa’s elves) who will create your custom series of clear aligners to help you achieve that smile.
I’m still not seeing how the Relu AI is relevant…
Cue another analogy.
Would you build your dream home on a parcel of land if you didn’t know what was underneath the surface? If you answered ‘yes’ to this question, please do not become a building contractor.
The correct answer is ‘no’. Why? Because not all soil is suitable to support a building.
The traditional method for planning clear aligner treatments relies on the IO scan, which only shows the crowns of the teeth. That means there is no visibility into what is going on beneath the surface, i.e. with the roots or bone structure. When using the IOS-only method, or combined with model roots, it is difficult to predict how the clear aligner treatment will impact the roots and bone. The likely outcome is that the teeth will move slower, or not as predicted, and certainly lead to a need for treatment refinements, which are both costly and lengthen treatment time.
Are you wondering why the roots and bone aren’t already taken into consideration? It is a fair question. Some would argue that minor cases don’t need them, and that is true, but when you consider the fact that nearly 80% of clear aligner treatments require refinements, it seems like a moot point. A major reason is because the process of manually retrieving them from the CBCT scan takes hours. This is one of the ways our AI can make a big difference. We make this type of work possible in a matter of minutes.
The outcome? Clinicians can see a 3D view of the patient’s crowns merged with their actual roots, allowing for accurate root movement predictions and a need for fewer refinements. And the benefit for you, reader? Potentially shaving 5 months off of an 18-month treatment.
Alright, I’m starting to understand. Any other examples?
Dental implants are a common intervention method for tooth loss. They differ from other types of tooth replacements because they not only replace the crown, but the root as well. We are talking about surgery here, so of course, planning the implant procedure is essential to ensure success.
The building contractor analogy still plays, so we are going with it.
Suppose the contractor assesses that they aren’t building on clay. Good, building on clay is a bad idea. But what else is going on under there? Gas lines, sewage pipes, power lines, internet cables? It is pretty important to know. If a contractor doesn’t find out what is there, or if they do find out and still dig around and damage something, they are responsible. This is the same for dentists.
Armed with radiographic images of the patient, the clinician needs to know what is going on even deeper down, so they will map out the borders (segment) of the different structures and measure the volume of bone and soft tissue. They need to make sure they can safely perform the surgery. What do you think is going through their mind while doing this?
‘Where are you mandibular nerve? Wait… what is THAT? Is that the sinus floor? Does that look like enough bone to support the implant?’
Sounds fun, right? Maybe a little stressful? Definitely time consuming. For a technician who does this job on a daily basis, we can imagine that it becomes tedious after a while.
Relu’s AI can help. The time-saving factor is nice because the AI can automatically segment the different structures vital to planning the surgery such as the teeth, lower jaw and upper jaw.
Also, to have an accurate 3D model of the patient to simulate the implant placement, the clinician normally has to align the IO and CBCT scans manually. This process is like a kindergarten teacher trying to get her students to stand in a straight line: when she finally gets Harold to stop rolling on the floor, Taylor is out of line chasing a butterfly. Although we wish we could claim that our AI could help a kindergarten teacher manage her class, we can’t. But for those dental practitioners out there, we can! Our AI will automatically align the two scans.
From a patient safety perspective, it can also automatically trace the mandibular nerve, which, if impacted during implant placement, could place the patient at risk of reversible or irreversible nerve injuries, such as numbness or pain of the lip and tongue, excessive drooling or loss of taste.
AI and Relu, a match made in … Leuven!
By now, we hope that you have a general understanding of what we are doing, but not really how we are doing it. Pull up a chair, friend. It is time to talk AI.
AI or Artificial Intelligence
AI is a fancy and trendy name for a really smart piece of software. The goal of it is to
think like a human does. It is trying to perceive, learn, reason and problem solve just
like we do. And like a human, it has the ability to train itself to become even better
software. How does it do that? Magic!
Sadly, no. Relu has a team of people constantly working on the AI. They are basically teaching it how to teach itself. To achieve this, they tell the AI when it is wrong, how to validate its results and, most of all, how to learn from them.
What is your team training the AI to do?
Good question unknown person! They are training the AI to (1) look at a patient’s CBCT and IO scan, (2) use them to make a 3D model of the patient, and then (3) automatically identify and label the different structures of the head. As more and more people use Relu’s software, the more accurate and powerful it becomes.
Sorry, readers. No more analogies, but we have some nice visuals to help break it down.
Input:
First, you give the AI a CBCT and IO scan which looks something like the pictures on the right.
AI Model:
Next, the AI puts the scans together to make a digital 3D patient model. Remember the poor kindergarten teacher we talked about earlier? That alignment is happening here! Take a gander at our virtual CEO.
Then the AI will identify and label the 3D model. It looks at each little 3D pixel (voxel) and labels what structure it belongs to. This highlighted red spot, for instance, is labelled as belonging to the lower jaw bone.
We know what you are thinking: but how does it know?
We could ask you the same thing: how do you know? How do you know what teeth are or what a jaw looks like? You were taught. The AI was taught as well.
Output:
And there you have it! The hard work of our AI gives you the results.
A dental practitioner would spend nearly four to five hours to complete this process manually.
Fortunately, our AI can do it in five to fifteen minutes. That is a significant time savings!
We mentioned earlier that existing treatment planning software lacked some functionality,
but with Relu AI, they can:
- Optimize treatment planning and simulation,
- Streamline patient communication, and
- Enhance research, education and training
It seems pretty simple now, right? But think about it. To get to the lovely, colorful 3D model above, the AI needs to know some pretty impressive stuff like:
- How to match teeth and roots between the scans?
- How to create a 3D patient with all the parts aligned?
- What is bone? Nerves? Gums?
In a nutshell
Relu automates and enhances digital dental treatment planning. Our AI is integrated inside of popular dental treatment planning software to empower the clinician. This means that hardware manufacturers of IO/CBCT scanners and companies that develop treatment planning software can incorporate Relu technology to offer more automation to their client’s workflow, reducing the time it takes to complete manual processes, limiting the number of (costly) adjustments needed during surgery or treatment, and ultimately, optimizing patient care. Relu’s AI software has been validated in clinical research and reviewed by experts in the field.
Closing thoughts
Relu’s technology is a tool. Its goal is to help improve the dental industry with time-savings, added precision and more predictability. It is in no way a replacement for a dental/oral professional. And just like any tool, it needs to be in experienced hands to enable its full potential.
FUN FACT
Were you wondering where our name comes from? Relu is an abbreviation of Rectified linear unit. It is an activation function used in artificial neural networks and something we use quite a lot when programming that smart AI of ours!
