INTRODUCTION

The orthodontic specialty has advanced greatly over the past decades. One of the most revolutionary changes in the practice is the invasion and dissemination of digital orthodontics (DO).¹ Despite the wide array of applications of DO, there remains an exclusively practice-changing domain, which is creating a robust decision support system to orthodontists.² This encompasses digitization and validation of diagnostic/therapeutic tools,² followed by data collation that would then enable machine learning.^3,4 Since orthodontic diagnosis and treatment planning is considered a complicated process with a multitude of interacting factors requiring great knowledge, experience and skills, the application of artificial intelligence (AI) might be of great need to aid practitioners in the decision-making process.^3,5 Whether these are expert orthodontists wanting to save time and operate at a higher level of efficiency, or others with less clinical experience who need guidance throughout their trajectory, AI is expected to be of utmost importance.⁶ So far, AI-based models have been successfully applied to automated lateral cephalometric tracing, determination of the cervical vertebral maturational stages, assessment of facial attractiveness, diagnosis of temporomandibular joint osteoarthritis, and the need for orthodontic extractions or orthognathic surgery.^3,6–8 However, it has not yet been applied to the mere diagnosis of specific malocclusions with their various components.^3,5 Being the language of the era, AI is expected to be deliberately imperative, especially with the integration of various efficient models and their permissive application.³

One of the most challenging types of orthodontic problems are those occurring in the vertical dimension.⁹ The involvement and interaction of various contributing factors, and their subsequent impact on treatment considerations and outcomes give them substantial priority over other malocclusions in different dimensions.⁹ The anterior open bite malocclusion (AOB) represents a variant of vertical jaw dysplasia where there is lack of anterior tooth contact.¹⁰ AOB might happen due to muscular factors, increased jaw divergence, under-eruption of anterior teeth, overeruption of posterior teeth or a combination of these factors.^11–14 There are also other local and systemic conditions that might cause or be associated with AOB, such as trauma, degenerative joint disease, inflammatory, or autoimmune diseases.¹² The basic types of AOB are the skeletal and dental variants,^12,13 However, more recently, AOB was further subclassified into distinct phenotypes depicting the co-occurrence of certain contributing components.¹⁵ A systematic approach was described in the form of the open bite checklist and a diagnostic tree to discern the AOB phenotype of each patient.¹⁵ Despite being insightful, the open bite checklist manifesto¹⁵ is thought to be quite tedious and time-consuming. Therefore, it was speculated to design computer software using the idea of the open bite checklist, after its revision and modification.

The concept of digitization is intended to produce an efficient and standardized diagnostic tool for which reliability and validity reflect those of the algorithm built into the software.⁸ This software is aimed to form the foundation of a future AI-based model for distinction of AOB phenotypes and their contributing components. This model is supposed to combine the merits of individualized diagnosis and treatment planning together with the efficiency and reliability of DO. Such a combination would enable the orthodontist to tailor the treatment for each particular patient by merely targeting the distinguished offending components.

Accordingly, the aim of this study was to present the newly designed software for AOB phenotype diagnosis. It also aimed at validating this software in comparison to the conventional method used for AOB diagnosis.

MATERIALS AND METHODS

The sample of this study was adult subjects with AOB malocclusion seeking orthodontic treatment at the outpatient clinic of the Orthodontic Department, Faculty of Dentistry, Cairo University in Egypt. The inclusion criteria were adult patients with AOB, skeletal or dental, with no history of systemic disease, chronic medications, or previous orthodontic treatment. The sample characteristics and demographics are summarized in Table 1. Informed consent was obtained from all patients before starting treatment and included using their records in research and publication. Ethics committee approval at Faculty of Dentistry, Cairo University, was obtained before conducting the study. Since this was a diagnostic study, only the pretreatment records of the patients were used, including photographs, radiographs and clinical examination data.

Table 1. Demographic Data of the Included Sample^a

View Fullscreen

A diagnostic checklist for AOB (Figure 1) was the foundation for constructing the specifically designed software. It was inspired by a previously published checklist,¹⁵ but modified to include some overlooked components such as the extent of lower incisor show on smiling and anterior gumminess. Another amendment to the published checklist was modifying the conditions for obtaining the answers to the questions, as well as the norms provided for the upper/lower/anterior and posterior dentoalveolar heights.¹⁶ The software was developed using Java FX, which is a high-level, class-based, object-oriented programming language. The formulated algorithm was used as a source to feed a future AI-based model in a forward chaining method. The software used the answers of the first four questions of the modified checklist as input, with the output being a diagnostic report (Figure 2) defining AOB phenotype, severity, habitual/functional factors, smile esthetics, skeletal/dental and soft tissue factors pertaining to the presenting malocclusion for that particular patient. Additionally, AOB phenotypes were recast to discern the co-occurrence of different components in both arches. For simplification, the AOB phenotypes were classified into main categories, as well as combination phenotypes (Table 2). The main categories were those having similar components in both arches, while combination phenotypes had different components for either arch.

View Figure

• Download as Powerpoint
• Download Figure

View Figure

• Download as Powerpoint
• Download Figure

Table 2. Phenotypes of Anterior Open Bite^a

View Fullscreen

The algorithm fed into the software was designated to include more variations of patients belonging to each AOB phenotype. The output of the software regarding AOB phenotype was termed digital form output (DFO).

For validation of the software output, a comparison was made to that of a conventional form checklist (CFC) used to manually reach the AOB phenotype (Figure 3). It was formulated based on the experience of two senior orthodontists by a consensus of the factors and measurements commonly used in the assessment of AOB cases, with a background of the formulated AOB phenotypes. The outcome of the CFC was termed conventional form output (CFO). Each patient was assessed using both methods, CFO and DFO, and the results were compared (primary outcome).

View Figure

• Download as Powerpoint
• Download Figure

The secondary outcome of the study was the number of missing diagnostic components in the CFO or DFO and their subsequent impact on the agreement level. The missing components comprised the cephalometric measurements of the anterior and posterior, upper and lower dento-alveolar heights that were not included in the CFO.

RESULTS

There was a statistically significant agreement between the CFO and DFO, with a percentage of agreement of 82.2% (83 cases out of 101), and a kappa coefficient of 0.78, which reflects a good level of agreement (Table 3). The AOB phenotype that showed the greatest level of agreement was the uncompensated skeletal open bite (SOB) (representing 28.7%; 29/101) (Table 3). On the other hand, the overall percentage of disagreement between the CFO and DFO was 17.8% (18 cases out of 101) (Table 4). The greatest disagreement took place between the uncompensated SOB and the combination phenotype of uncompensated SOB in the upper arch, with dental contribution in the lower arch (Table 4).

Table 3. Percentage of Agreement and Kappa Coefficient Between DFO and CFO^a

View Fullscreen

Table 4. Percentage of Disagreement Between DFO and CFO Phenotypes^a

View Fullscreen

The number of missing diagnostic components was determined for all the cases, and the percentages are shown in Table 5 with their association to the agreement and disagreement cases. Seventy-eight percent of all the cases (79 cases out of 101), and 100% of the cases that showed disagreement had missing diagnostic components when using the CFO. The chi-square test showed a statistically significant correlation between the number of missing diagnostic components and the level of disagreement between CFO and DFO.

Table 5. Number of Missing Diagnostic Components and Correlation to Agreement/Disagreement Groups

View Fullscreen

The intra- and inter- observer reliability analyses of the CFO showed kappa coefficients of 0.87 and 0.84, respectively (Table 6), which indicate very good levels of agreement. The DFO method showed a kappa coefficient of 1, which reflects a 100% agreement for intra- and inter-observer reliability analyses (Table 6).

Table 6. Intra- and Interobserver Reliability Analysis of CFO and DFO^a

View Fullscreen

DISCUSSION

Vertical dysplasia of the jaws and, specifically, AOB malocclusion represents a complex dentofacial condition, with functional, esthetic, and biological impact.^12,13 Several reports aimed at analyzing the contributing components of AOB,^14,15,17 and, lately, certain phenotypes have been ascribed, with a checklist and a systematic approach to distinguish each patient’s phenotype.¹⁵ However, upon application of this approach, some modifications were deemed essential to the checklist and the diagnostic approach to include a wider array of patients, while preserving the concept of phenotypic distinction.

Another limitation of the previously published checklist and diagnostic tree was its complexity of application and the presence of overlapping phenotypes when considering the upper and lower arches separately. With this fact, coupled with the rising pace of digital orthodontics, it was decided to create fully customized software to surmount the hurdles of this diagnostic procedure. The software would take the answers of the modified checklist as inputs and create a diagnostic report as the output, which takes only a couple of minutes. This diagnostic report not only distinguishes the AOB phenotype for the particular patient, but also the various soft tissue, skeletal, and dental components. Additionally, the habitual and functional factors are included together as well, with the severity of the AOB, smile esthetics, and other associated factors like crowding/spacing, discrepancies in anteroposterior or transverse dimensions, etc.

For the different AOB phenotypes, the basic distinction between skeletal and dental types depends on certain clinical and cephalometric variables.^18,19 However, it seems pertinent to elucidate whether there is any sort of dental compensation/contribution in cases of SOB as this greatly affects not only the treatment plan, but also the treatment sequence, mechanics, and, hence, the expected outcome. Dental compensation in AOB cases is defined by the vertical eruption of the upper/lower anterior dentoalveolar segments to compensate partly for the jaw divergence, while dental contribution reflects the involvement of anterior segments to the AOB by their under-eruption.¹⁵ So, they basically represent opposite phenomena, and each of them implies a certain intervention based on their extent, other contributing factors, and smile esthetics. Understanding these terms, together with respecting the individualized nature of malocclusions and benefiting from the advances in technology, mapped the foundation of designing this study.

The main and combination phenotypes of AOB described in Table 2 are expected to cover most if not all the possible variants of AOB cases. The algorithm built into the software was established with the ability to identify the AOB phenotype, prioritizing the clinical over the cephalometric findings, and providing for variations caused by the multiplicity of involved factors.

After the software was executed, the validation process had to include a comparison to a conventional method for AOB diagnosis. A CFC was created by asking expert orthodontists about the factors they would examine in an open bite patient, being familiarized with the AOB phenotypes. The CFC did not include the upper/lower, anterior and posterior dentoalveolar heights, as they were not commonly used as reported by expert orthodontists. Instead, the clinical variables reflecting these components were included.

The percentage of agreement between CFO and DFO was 82.2%, while the kappa coefficient was 0.78, which reflected a good level of agreement between the two methods, displaying high statistical significance. These findings primarily represent good performance of digitized diagnostic procedures, which would save a lot of precious time and effort for expert orthodontists, and also allow less experienced practitioners to unravel diagnostic information that would otherwise be ambiguous.^2,4

The highest percentage of agreement was for the uncompensated SOB phenotype. The reason behind this finding might have been the greater occurrence of this phenotype and the relative simplicity of its diagnosis, as it neither involves dental compensation nor contribution.¹⁵ On the contrary, the greatest disagreement was reported between the uncompensated SOB and the combination phenotype of uncompensated SOB in the upper arch, with dental contribution in the lower arch. The main difference between these phenotypes is the reduction of the lower anterior dentoalveolar height (LADH) in the latter that is diagnosed clinically or cephalometrically. Since the CFC did not use the dentoalveolar heights in their input, clinical judgment offered the single source of diagnostic information regarding LADH. However, the software gained information regarding LADH from clinical assessment and cephalometric measurements. Therefore, the DFO was found to be more intricate in diagnosing this phenotype.

Since the upper/lower, anterior and posterior dentoalveolar heights were not commonly used by experienced orthodontists and, accordingly, were not included in the CFO, the DFO based on the modified checklist had a better ability to pursue the AOB phenotype. Although the software was designed to prioritize the clinical over the cephalometric findings, Including both in the decision making led to a more robust diagnostic ability of the DFO over the CFO. Interestingly, there was a direct relationship between the number of missing diagnostic components and the occurrence of disagreement between CFO and DFO. This highlights the importance of these components, despite being overlooked even by experienced orthodontists. Nevertheless, a considerable number of cases showed agreement despite having missing cephalometric components. Such a finding might emphasize the need to investigate the reliability of the norms of these variables and, perhaps, study population/age/gender and skeletal classification-specific values that should be based on reliable landmarks. This is because there are suggestions that dentoalveolar heights vary according to the vertical and sagittal skeletal relations.¹⁶ Then, the true contribution of these variables to the AOB phenotypes can be better instituted. Also, the probable inconsistency in lateral cephalometric tracing previously reported²⁰ could be a reason for the agreement in cases having missing diagnostic components.

The reliability analysis performed showed good and very good reliability of the CFO and DFO, respectively. This gives confidence in the expert opinion used in the construction of CFO, and in the algorithm built into the software. The DFO showed 100% reliability, indicating robustness of the included components and the logic built to produce the output. Accordingly, the newly designed software can be considered a reliable, updated digital tool for detailed diagnosis of AOB.

The detailed diagnostic report created by the software is considered the basis for individualized treatment planning. Besides being accurate, fast, and comprehensive; automatically detecting the diseased component “etiological factors” for each patient will allow practitioners to tailor the treatment toward the defective components aiming for a better treatment outcome with enhanced stability and esthetics.⁴ This is particularly intriguing in an era with scarce evidence and lack of methodologically sound trials,^21,22 which might be attributed primarily to the diagnostic shortcomings of the studied samples.

The findings of this study provide a profound standpoint from which data mining could be started, taking a step toward machine learning.^3,5 This process would involve image recognition, digital cephalometric tracing, then automatically filling the checklist on the software to produce the final output.

CONCLUSIONS

• The newly constructed software represents an efficient and valid diagnostic tool for AOB phenotype and its contributing components.

• There was good agreement between the conventional form output and the digital form output, with the latter being more comprehensive and reliable.

• The digitization and validation of the modified open bite checklist manifesto accomplishes the first step toward initiation of the machine learning process concerning the diagnosis of anterior open bite malocclusion.