The main purpose of this study was to form biologically informative, digital templates of dentofacial deformities, so as to enable rapid, computerized, automatic diagnoses by matching the imaging data of a new patient with the characteristic templates. A comprehensive and accurate classification is critical to the process of template establishment. Subclasses of dentofacial deformities have been described in the literature [20, 23, 25]. However, most such classification systems rely on a single angle measurement or molar relationship. A comprehensive classification of deformity patterns requires subclasses that are not based on a single variable alone. Moyers et al. classified 697 patients with class II malocclusions into six horizontal and five vertical divisions by using cluster analysis, and developed diagnostic definitions and designed treatment plans pertaining to each subclass [20].
In this study, we used cluster analyses with no a priori definitions to establish a classification system of craniofacial deformities. This research was carefully designed to avoid the shortcomings that were limited by sample size or selection bias. We included a sample of 2249 patients with any type of malocclusion and performed a comprehensive analysis of as many cephalometric landmarks as possible. Considering the ethnic diversity, we only recruited Han population into the sample. The cephalometric measurements were obtained by working out the average locations of the selected landmarks. All the landmarks were independently identified and located by three professionally trained orthodontic residents. The data were calibrated, which helped to ensure the accuracy of the results. A total of 21 CMTs were established representing almost all patterns of dentofacial deformities. These cephalometric values and CMTs could not completely represent of an ethnicity, but can still be used for reference.
The most commonly used method to evaluate therapeutic effects in previous studies was choosing one type of malocclusion (mostly based on Angle’s classification), and then statistically analyzing cephalometric measurements before and after treatment. This method is simple, appropriate to understand and widely used. However, there are two obvious flaws in this method. First, the classification of the entire sample is based on a single feature, such as molar relationships or jaw relationships, ignoring the influences of other confounding features. Second, line and angle measurements cannot reflect the overall dentofacial morphology, unlike the main landmarks shown on CMTs (Fig. 1). The CMTs formed from this cluster analysis are based on landmark information, and therefore, the resultant classification based on these CMTs will ensure a high degree of sample similarity within the same subclass. The CMT of a large sample represents the mean morphology of a pattern of similar structures, and the changes in the CMT before and after treatment represent the mean morphological changes in samples with similar structures. This close similarity will greatly increase the reliability of efficacy analysis. CMTs contains much more cranial and maxillofacial deformities than traditional method.
Another obvious advantage of CMTs is that they can be calculated on the computer and display the changes in the position of landmarks. Therefore, the changes before and after treatment can be seen intuitively, which is more conducive to orthodontists’ judgment and doctor–patient communication. Although hand-traced templates have been used to evaluate the dentofacial characteristics of patients [5, 6], predict growth [7] and perform diagnoses [8], few such templates have been used for treatment evaluation. CMT analysis is akin to traditional superimposition analysis, but is more effective than superimposition and better reflects average changes for a class of samples.
For example, we subdivided patients in the CMT_5 subclass, which is a common type in clinical practice, into two groups based on whether or not they underwent tooth extraction. Great improvement in the profile was seen in the extraction group, with significant reductions in the inclination of the upper and lower incisors and in lip protrusion (Fig. 3). In contrast, the soft and hard tissues were not well improved in the non-extraction group, and protrusive incisors and a convex profile persisted after treatment.
The clinical effects of functional appliances are controversial, whereas anterior protraction treatment for skeletal, class III malocclusion is commonly viewed as being efficient [26]. We therefore compared the efficacy of the F-III appliance and anterior protraction in the patients in the CMT_18 subclass, which pertained to skeletal, class III malocclusions (Fig. 4). We found that the ANB angle was more significantly improved in the anterior protraction group than in the F-III functional appliance group, which is consistent with previous studies. Anterior protraction seemed to be more beneficial for skeletal improvement, whereas the F-III appliance seemed to better improve soft tissues. The profile improved well in both groups, without obvious differences. The results seemed to suggest that both methods could be used to treat patients with skeletal, class III malocclusions in mixed dentition or early permanent dentition.
Since the computer system is not yet operational and needs further improvement, there are two main ways to use templates today. First, these templates are a morphological template based on coordinate clustering. Morphological matching could be used quickly in the clinic for reference by these templates, but it is not the most accurate way to use. Second, to get an accurate result, a computer is needed to substitute the coordinate value into the equation. Substitute the coordinate values of the sample into the 21 discriminant equations to see which equation results in a higher value. The sample corresponds to the group with the largest result value of the discriminant equation.
However, the best way to do this is to automate it all. Therefore, the goal in the future is to use software to achieve accurate discrimination. Although we only partially contrasted the effects of several treatment methods in a relatively large number of samples, this is exploratory investigation can provide a basis for future research into the prediction of treatment outcomes in patients with dentofacial deformities. New cases will be gradually accumulated in the informative CMT database, which will be enriched with information about the outcomes of different treatment methods.
In addition to treatment evaluation, this study is superior in many aspects, including the large sample size, cross-validation, and general diversity. Most importantly, a priori definitions were not applied before clustering. In this study, all 2249 patients without a priori pattern definitions were divided by cluster analysis based on the coordinates of their lateral cephalograms rather than the conventional Angle’s classification. In this way, samples of each subclass had integral dentofacial similarity, and represented almost all types of dentofacial deformities.