AI-assisted Diagnosis and Therapy Finding for Craniomandibular Dysfunction (CMD)
Initial situation
Axiography with the gamma dental axiograph and Cadiax software is an established method for precisely recording temporomandibular joint movements. In practice, however, these curves are usually evaluated manually, which is time-consuming, requires a high level of experience and leads to different interpretations between practitioners. This can delay the diagnosis of craniomandibular dysfunction (CMD) or lead to inconsistent results.
Despite technical advances, there is no widespread AI-supported system that automatically analyzes real axiography time series and converts them into clear, clinically useful findings. This is due to the complexity of the movement data, inconsistent recording protocols and the lack of large, curated data sets.
Solution idea
The idea was to pre-process the data as part of the Quick Check and develop a diagnostic AI model to detect jaw misalignment. During pre-processing, the protrusion/retrusion, mediotrusion left/right and opening/closing movement data was aligned to eliminate misalignment issues related to data acquisition and hardware placement. This data was then used directly to train a classifier.
Benefit
The greatest benefit lies in the practical integration of AI-based diagnostic support into the existing Cadiax software. As an add-on, the developed model can be integrated directly into the familiar workflow of dentists - without additional devices or complex training. This makes the technology immediately accessible to thousands of users worldwide.
Thanks to the automated evaluation of axiographic data, practitioners benefit from faster, more objective and reproducible diagnoses, while modern sequence learning methods can be integrated in addition to classic feature models. Time series architectures such as 1D-CNNs, GRU/LSTM networks or Temporal Convolutional Networks capture characteristic movement patterns directly from the data points and model complex functional progressions. Similarity methods allow the comparison of curve morphologies between patient groups. Self-supervised »representation learning« - such as contrastive methods or autoencoders - uses larger unlabeled data sets for more robust and precise diagnostic classification.
Implementation of the AI application
The AI application analyses Cadiax condylar paths using a standardized 6 mm alignment of the start of movement in order to make different patient images comparable. Diagnostically relevant geometric features - such as path lengths, inclinations, lateral deviations, symmetry parameters and left-right differences - are extracted from the aligned 3D curves. These compact feature vectors feed lightweight ML models that reliably classify functional patterns and indicate potential craniomandibular dysfunctions.