Let AI do the repetitive work: How machine learning helps dental labs print partial denture frames

Artificial Intelligence (AI) is changing the face of manufacturing. The world is abuzz with talk of AI tools for boosting productivity and saving time. An important way of achieving this is the automation of repetitive, time-consuming tasks.

3D printing, a relatively new technology, is lagging behind more mature manufacturing processes like sheet metal fabrication when it comes to automation. To make one application of additive manufacturing more streamlined, Oqton developed automated segmentation of removable partial denture (RPD) frameworks, a feature that can save dental technicians hours every day. It’s a perfect example of machine learning liberating trained staff from tedious tasks and transforming dental production. 

One of the first labs to trial this capability was Bertram Dental Lab, a large RPD manufacturer in the USA. Automation is among the chief advantages they identified in the Manufacturing OS and they have seen significant headway in RPD segmentation, a process they had always wanted to automate. “It just took somebody who knew what they were doing with AI for dental labs to actually develop that capability,” Tim Bertram, co-owner of the lab explains

Let’s take a closer look at how AI-based segmentation works.

3D printing partial denture frames 

Occasionally a manufacturing technology comes along that is vastly superior to anything used before. Today that is the case with metal 3D printing of RPD frames.  

In some labs, it has displaced hand casting entirely. Its workflow is simpler, and it results in a better-fitting, stronger product that meets aesthetic expectations. 

That said, 3D printing also demands a considerable amount of work from dental technicians, and a huge portion involves repetitive tasks. This inevitably calls to mind support structures. 

Supports are not the most exciting aspect of 3D printing. However, it’s impossible to print some parts successfully without them. At the same time, they lead to an increase in labour hours, material costs, and post-processing work. 

Many dental technicians spend hours on data preparation for 3D printing RPDs, including manually adjusting the supports that software generates.  

Some software companies claim to offer a solution. They come with a standard support strategy which they propose can be applied to dental parts as well as industrial ones. This should enable dental technicians to generate the right supports automatically. 

This is only partially accurate. While these standard strategies do take into account the overhangs and position of the part, this level of automation doesn’t provide satisfactory results for RPD frames.  

The amount of supports you want in an RPD frame heavily depends on the section of the part – clasps require denser supports while the plate requires fewer of them. Technicians understand how to adjust the supports, so they tweak the automatically generated supports. This job can take hours a day.  

In short: support generation for RPD frames is time-consuming, labour-intensive, and costly.

A prime candidate for automation.

AI helps generate supports for 3D printing 

The solution for simpler support generation lies in segmentation. 

The density and amount of supports depend on the region of the RPD frame, so the task that needs to be automated is the identification of all the regions of a frame, which we call segmentation. If we want software to substantially increase efficiency, it’s critical that it segments a partial denture with the same accuracy as a dental technician. 

Oqton has automated RPD segmentation by combining AI with other techniques. This type of task is perfect for AI. Oqton’s engineers Vice Rončević and Anthony Rathé, who developed this feature, explain how the process works. 

“RPD segmentation is a great example of AI working in tandem with other geometry-based techniques,” Anthony explains. “The core of this system is AI-based. It robustly recognizes the basic regions of the RPD – clasps, meshes and plates. However, this doesn’t result in a very clean segmentation and it has limited use. It needs to be complemented with other approaches.”   

The AI predicts the segments of an RPD 3D model uploaded into the Manufacturing OS, which are marked in different colours. 

Since partial dentures need to fit each person’s anatomy, they are extremely variable. Variability is a good candidate for generalization – which is where AI excels.

The first step was to train an AI model on existing segmented RPDs so that it could generalize and predict segments on previously “unseen” samples.  

This resulted in a model that was able to identify some regions of RPD frames – but more nuance was needed. “After the training, the AI was predicting basic regions of the RPD,” Vice explains. "However, there are other regions and sub-regions that require a different support strategy, so the AI results were extended to account for those too."  

The additional regions are crucial. Clasps need to fit well, which means they need to be printed very precisely and require the right support strategy. The regions pictured in orange colour need to be supported and the internal, red, part should not have any supports at all.

To extend the AI predictions, the team used a geometry-based technique called ‘region growing’ in tandem with a post-processing pipeline for the removal of small inaccuracies in predictions. This automated segmentation strategy can identify all the relevant regions. 

The wider potential of automating the 3D model segmentation

Automated segmentation is a feature Oqton developed specifically for 3D printing partial dentures, but its potential extends beyond this area. Segmentation of any part into its constituent features is useful for many downstream applications.  

Orientation, label and pin generation for CNC milling and classification are just a few processes that could benefit from automated segmentation. Because it’s an enabler for a lot of productivity down the line, Oqton’s dedicated AI team will continue to work on further advancing this approach.