Three-dimensional cephalometric analysis of the maxilla: Analysis of new landmarks.

Introduction: Clinical evaluation of the midface including the paranasal and upper lip regions is highly subjective and complex. Traditional and 3-dimensional cephalometrics were not developed with the clinical appearance of these midfacial areas in mind and are therefore inappropriate surrogates for the clinical appearance of the midface, making them unsuitable as aids in diagnosing dentofacial deformities. The aim of this study was to evaluate traditional as well as newly defined landmarks and measurements and their correlation with clinical appearance of the midface.

Methods: Fifty-two subjects who underwent full-field cone-beam computed tomography were recruited for this study. A single examiner assessed each subject's midfacial region (paranasal and upper lip), and a second examiner obtained traditional and newly defined cephalometric measurements for each subject. Both examiners were blinded to each other's data throughout the study. Statistical analysis was performed to assess the correlations of the traditional and novel cephalometric measurements with clinical midfacial findings. The impact of the soft tissue thickness in the paranasal region was also analyzed. The performance of any classification derived from statistically significant variables was analyzed with the use of micro-F scores and area under the receiver operating characteristic curve (AUC).

Results: Both traditional (SNA) and newly defined measurements (SN, SN, SN, SN) had no statistically significant correlation with clinical paranasal diagnosis. However, in the absence of upper lip procumbency or protrusion, SN and SN had statistically significant correlations with clinical paranasal diagnosis (P = 0.047 and P = 0.003, respectively). For upper lip analysis, both traditional (SNA) and newly defined measurements (SN) had strong correlations with clinical upper lip diagnosis (P < 0.001). All statistically significant cephalometric variables had good intra- and interobserver reliability (correlation coefficients ≥0.972 and ≥ 0.968, respectively) except SNA, which had a low interobserver reliability (correlation coefficient 0.739). Fitted models for paranasal and upper lip analyses showed low micro-F scores, indicating low precision and recall. However, AUC values of 0.7019 and 0.6362 for the paranasal and upper lip analysis, respectively, suggest improved performance of the model when properly trained with a larger sample size.

Conclusions: Newly defined measurements SN and SN correlated with clinical paranasal diagnosis only in the absence of upper lip procumbency and protrusion. SNA and SN were strongly correlated with clinical upper lip diagnosis. However, fitted models based on this study sample yielded low micro-F scores, making the fitted models currently unsuitable for anything besides correlation with clinical findings. A larger sample size will be necessary to further clarify the potential roles of these measurements, especially given the reasonable AUC values. The findings of this study demonstrate the highly subjective and relative nature of midfacial diagnosis and the importance of clinical judgment despite the potential utility of some traditional and new measurements.