Neural Network Model for Dentoalveolar Improvements

Sponsor: School of Dentistry, University of Washington     

 

Malocclusion is the misalignment of the upper and lower teeth when the jaw is closed. Class II malocclusion is one of the three types of the malocclusion where the lower molar is placed behind the upper molar as shown in Figure 1.

Figure 1. Class II malocclusion

 

There are different protocols for treatment of class II malocclusion. Several features are used to measure the effectiveness of each protocol and amount of improvement. Two major features are Peer Assessment Rating (PAR) and the difference between two cephalometric values. Peer Assessment Rating (PAR) scores are computed from clinical exam findings and from scoring dental models. It is a measure to document malocclusion severity and treatment difficulty. The difference between two angular measurements is used to relate the position of the maxilla (mid-face) to the mandible (lower-face); angle ANB in figure 2 represents this measurement.

 

The objective of this study is to create a neural network model that could foresee the dentoalveolar improvements by predicting the values of PAR and angular measurement, based on selected measured features at the beginning of the treatment. This model assists the dentists and the patients to learn how much improvement they shall anticipate at the end of the treatment.

Figure 2. Ceph5b ceph8b is illustrated as angle ANB; the difference between angles SNB (ceph5b) and SNA (ceph8b).

 

Figures 3 and 4 depict some of the preliminary results achieved from the neural network model. This network successfully produced the majority of its outputs within the desired range.  In other words, 92% of PAR and 80% of angular measurement testing values were within the 1.45σ and 1.25σ of their true values.

 

Figure 3. Neural network output of Node 1 (PARDC) and the target (true) values are compared for 50 testing patterns.

 

 

 

 

Figure 4. Neural network output of Node 2 (ceph5b ceph8b) and the target (true) values are compared for 50 testing patterns