INTRODUCTION

The diagnostic value of cephalometric analysis depends on the accurate and reproducible identification of clearly defined landmarks on cephalometric radiographs.1–3 The difficulty in identifying these landmarks is compounded by the variability of the patient's hard and soft tissues, the radiographic quality, and the experience of the clinician. This is why landmark identification constitutes the main source of cephalometric analysis error.4–6 Manual analysis of a cephalometric radiograph is time-consuming, and measuring with a ruler and a protractor can increase significantly the total error of the method.

The introduction of computerized cephalometric analysis gave the possibility of direct digitization of radiographs. This process involves fewer steps, decreases the measurement error because the linear and angular cephalometric variables are automatically calculated, and facilitates the transmission of information between practitioners.7–10 However, in the clinical environment, time pressures can contribute to decreased reliability when the clinician is asked to subjectively define and digitize the position of a series of cephalometric landmarks.11 Advances and affordability in digital radiographic imaging have recently seen a demand for medical professions to automate analysis and diagnosis that were once performed manually.

Many studies have attempted to improve computer-aided landmark identification in the past. In general, their aim was to reduce the intra- and interexaminer variability and save valuable clinical time. Automatic landmark identification was first described by Cohen and Linney12 who applied a computer algorithm to retrieve the landmarks Menton and Sella. Although limited in the number of variables supported, this automatic process was an innovative departure from conventional analysis that relied on subjective decisions. Other studies followed with the same goal to enable the practitioner to scan a radiograph and run software that within seconds could retrieve accurate landmarks.

Implementation of different algorithms, among others, “knowledge-based system,”13 “pyramid method,”14 “grey-scale mathematical morphology,”15 “heuristic image processing,”16 or “active models shape,”17 attempted to locate more landmarks, but with limited success. Other models such as “spatial spectroscopy”11 and “pulse coupled neural networks and genetic programming”18 showed somewhat better results.

Recently, a new version of a cephalometric program was introduced (Viewbox 3.1.1, dHAL Orthodontic Software, Athens, Greece), which offers an innovative feature that can help landmark identification by detecting the edges of anatomical structures. In contrast to previous attempts, which aimed at a completely automatic landmark identification, the objective of this new feature is perhaps less ambitious but it could prove to be a more realistic and clinically useful alternative to conventional manual landmark digitization.

The purpose of this study was to evaluate the validity and the inter- and intraexaminer reliability when on-screen landmark digitization is performed manually or it is computer-assisted by means of the aforementioned cephalometric software feature. The null hypothesis was that there is no difference in both validity and reliability between the two methods.

MATERIALS AND METHODS

Twenty lateral cephalographs were selected from the records of 30 consecutive patients treated at a university postgraduate orthodontic clinic. Exclusion criteria were substandard radiographic quality or the presence of unerupted or partially erupted teeth that could hinder landmark identification. The quality of the cephalographs selected was considered to be sufficient for landmark identification rather than excellent, in an attempt to stay close to the clinical reality.

The radiographs were scanned at a resolution of 300 dpi (Epson Expression 1600 Pro, Epson Deutschland GmbH, Meerbusch, Germany), and a recently introduced version of the software Viewbox™ was used to digitize them. This software is provided with one feature of particular interest called Auto Edge Locking™. This feature provides a cursor that automatically locates the edge between hard and soft tissues. The operator must guide the cursor in the vicinity of the targeted landmark to allow the software to snap the cursor onto the existing edge between structures (Figure 1).

View Figure

• Download as Powerpoint
• Download Figure

All radiographs were digitized four times by two experienced orthodontists using two different methods: conventional on-screen digitization and computer-assisted digitization. A complete set of measurements was performed every week.

The computer-assisted method applies only to those landmarks that lie on the edge of anatomical structures. It was used for the identification of 13 commonly used cephalometric landmarks (Table 1). The position of each of the landmarks was defined with the help of a horizontal (x) and vertical (y) coordinate system (in pixels). However, the mean differences between methods were subsequently expressed in millimeters. Three more landmarks, Sella (S), Nasion (N), and Articulare (Ar), were manually digitized for each radiograph only once at the beginning of the experiment. Their exact position was then transferred (copied) to all subsequent digitization series of the same radiograph. This allowed the calculation of more commonly used cephalometric variables. All 16 landmarks were used to calculate a total of 10 variables: SNA, SNB, Upper incisor to NA, Lower incisor to NB, SN mandibular plane angle, SN Pog, Jarabak ratio, Lower lip to E Line, Lower lip to H Plane, and Subnasale to H Plane. Because of the methodological procedure used, the position of the landmarks S, N, and Ar was identical with repeated digitization of the same radiograph and therefore did not increase the observed variation of the cephalometric variables they codefine.

Table 1.  Cephalometric Landmark Definitions

View Fullscreen

RESULTS

The computer-assisted method did not agree with manual digitization in 7 of 13 landmarks under investigation. Intraoperator reliability was improved by using the computer-assisted method for B point (x-axis; P < .05) and Menton (x- and y-axis; P < .05 and P < .001, respectively).

Intraoperator reliability was decreased for point A (y-axis; P < .05). Interoperator reliability was found to be improved by using the computer-assisted method for B point (x- and y-axis; P < .05 and P < .001, respectively), Soft Labrale Inferior (x- and y-axis; P < .01 and P < .05, respectively), Soft Pogonion (x-axis; P < .05), and Menton (y-axis; P < .001) (Table 2). It was found to be decreased for point A (y-axis; P < .1).

Table 2.  Validity and Reliability in Digitizing the Cephalometric Landmarks Under Study Using the Manual and the Computer-assisted Methods. The Position of Each Landmark is Given Relative to the x and y Coordinate Reference System Used in this Study. All Values are in Millimeters^a

View Fullscreen

The differences observed between the manual and the computer-assisted methods as far as the cephalometric variables are concerned were very small. The two methods did not agree in 5 of 10 variables. However, if a difference of 0.5 mm or degrees is allowed, the two methods agreed for all but one variable (Upper incisor to NA). Both intra- and interoperator reliability was improved in the case of only one variable (SNB, P < .05) (Table 3).

Table 3.  Validity and Reliability for the Cephalometric Variables Under Study Using the Manual and the Computer-assisted Method. All Values are in Millimeters^a

View Fullscreen

DISCUSSION

This study evaluated the validity and reliability of a new edge-based computerized method of cephalometric landmark identification and digitization. This method comes as a new feature (Auto Edge Locking™) of commercially available cephalometric software.

Both cephalometric landmarks and variables were evaluated in this study. Although there was an improvement of the reliability of some landmarks and variables, the measurements performed using this feature were not found to agree in some of the cases with those performed after manual on-screen digitization of the cephalometric radiographs under investigation. These differences were very small, and probably clinically insignificant, but it seems that this promising tool may need improvement. Our study showed, as well, that the edge-based algorithm may be influenced by the individual radiographic image characteristics.

Most cephalometric points lie at the edge of skeletal structures. Although the human eye can determine the edge of a skeletal structure with precision, it is not always easy to trace the intended target with sufficient accuracy. The idea behind the software feature investigated in this study was that an edge-locking cursor could facilitate this hand to eye coordination during on-screen landmark digitization. Using a special algorithm, the software locks on an edge as the operator slides the cursor to the desired point. Therefore, it is not an automatic landmark identification system but provides a computer-assisted digitization feature that aims at increasing reproducibility. Average quality radiographs were used for this study with no particular problems. Although there has been an effort to address the problem of double structure identification using a special double directional cursor, edge-locking did not function in 2 of the 20 radiographs under study for the landmark mandibular tangent posterior.

CONCLUSIONS

• Computer-assisted cephalometric landmark identification led to some improvement of intraoperator reliability compared with manual digitization.

• The edge-based logarithm seems to be a promising tool to facilitate and increase the reliability of on-screen cephalometric analysis for both clinical and research purposes.

• There are difficulties in locating the appropriate edge when artifacts or soft tissue edges are located near the targeted landmark.

• The existence of small but systematic differences between the two digitization methods manifests the need for further improvement of the edge-based algorithm in order.