Various image manipulations can be applied to make some areas more conspicuous and aid in visualisation. Contrast, brightness and gamma can be adjusted, and histogram techniques can be applied over the whole image or at specific areas [13], as shown in Fig. 4.3.

Multiple digitisation has been recommended as a method to reduce point identification error [2, 23]. Baumrind and Miller [23] reported that in order to reduce this error by half, each point should be digitised four times, and the average of the four attempts should be used as the location of the point. Although this may be excessive [4], even a double digitisation is not possible without a computer system. Cephalometric software allow multiple digitisation on screen (without showing the previous attempts, so as not to bias the user) and calculate the average position. The user can inspect the digitisations and delete outliers. Multiple digitisation may be more important in case reports than large-scale studies, where errors tend to average out [2].

The scanned image can be magnified on-screen to almost any detail in order to facilitate point placement. The limiting factor here is the resolution at which the radiograph was scanned. As mentioned earlier, a resolution of 150 dpi will produce approximately 6 pixels per mm, which far exceeds the usual requirements of cephalometrics.

The precision with which points are located on-screen depends on a number of factors, such as scanning resolution, the zoom setting when digitising and the internal design of the software [15, 24]. The important thing to keep in mind is that the points are located using the mouse, so the movement of the mouse on screen is a major factor in determining the precision. For example, assume that a radiograph is displayed at approximately life size on a computer monitor. Assume a 15″ TFT monitor running at a resolution of 1,024  ×  768 pixels with physical dimensions of 304  ×  228 mm. The dimensions of a screen pixel are easily calculated as 0.297  ×  0.297 mm. Since the mouse can only move from pixel to pixel on the screen, it becomes evident that it is not possible to digitise points with more precision than approximately one-third of a millimetre. If the radiograph is zoomed out to occupy less screen space, this value will worsen. If more detail is required, the software should allow zooming. Assume that the image is zoomed in so that 10 mm is now displayed at a size of 10 cm on screen. This should allow digitisation at an accuracy of 0.03 mm. However, two other factors come into play at this magnification. One is the resolution used during scanning. If this was relatively low, then the image pixels will be apparent, setting a limit to the effective precision; although we can place the mouse and digitise at sub-pixel positions of the image, there is no way to ensure that the digitised location is correct. The final limiting factor is the internal design of the software program, which may not allow unlimited precision.

A practical method to calculate the maximum precision under specific conditions is the following:

One of the reasons that the error of point identification is high for points such as Gonion and Gnathion is that these points are located on curved osseous boundaries. The investigator has the task of locating the most extreme point along this boundary (e.g. the most inferior and posterior point, in the case of Gonion), not an easy task given the absence of anatomical markers. Geometrical constructions can be used as aids, but these may also introduce errors of their own. An alternative is to delegate the task to the computer. Software already exists that can assist in this respect. The user need only draw the outline of the boundary, and then the software automatically locates the points on this boundary following simple geometrical rules of point definition.

As an example, Fig. 4.4 shows the placement of points along the outline of the mandibular symphysis. The computer locates each point according to its definition; Pogonion is placed at the most anterior position, Menton at the most inferior, etc. The anteroposterior direction is defined by the Frankfurt horizontal plane, which has been digitised previously, thus circumventing errors caused by improper head orientation.

Automatic location of points on digitised outlines may remove some of the subjectivity, and therefore error, of point identification, but it may introduce a new source of error, that of tracing the outline. However, because the outline usually represents a well-defined brightness edge in the radiographic image, it is possible to use computer vision techniques to identify it. Edge-detection methods [13, 25] are among the first developed methods in computer vision, and, although not as reliable as one would like, they can be used to good effect. Such methods have only recently been introduced in cephalometric software, and their effectiveness in error ­reduction is beginning to be investigated [26].

Radiation risk is commonly assessed by the “effective dose”. The effective dose is the sum of the doses to each organ exposed to the radiation, weighted by a coefficient, which represents the organ’s sensitivity to radiation exposure. Organs such as the gonad and bone marrow have high weighting factors, whereas skin and neural tissue have low weighting factors. The unit of effective dose is the Sievert (Sv). For dental applications, where dose is low, the mSv (1,000 mSv  =  1 Sv) and the μSv (1,000 μSv  =  1 mSv) are used.

Due to exposure to natural radiation (from the earth minerals, from cosmic radiation, and from radiation within the human body), it is estimated that an individual receives about 3,000 μSv/year. A typical conventional CT scan of the whole head will incur an effective dose of about 2,000 μSv, which is equivalent to 8–12 months of background radiation. In comparison, a cephalometric radiograph entails a dose of about 5 μSv, and a panoramic gives approximately 25 μSv [52–56]. These values show that dental radiography involves low radiation exposure, but incorporation of 3D CT records as a routine orthodontic procedure would increase the dose substantially [57]. Cone-beam computed tomography (CBCT) results in a wide range of effective dosage to the patient, depending on the machine and the parameters of the examination, including field of view size and image resolution. Typical dosage ranges from 70 to 370 μSv [58]. Guidelines for CBCT imaging in dentistry have been established by national societies and a European initiative [59, 60].

The diagnostic value of 3D records is directly related to their validity and precision. The validity of CBCT records has been reported in a number of investigations, by comparing measurements taken from the 3D reconstructions of skulls or cadaver heads with direct measurements using callipers or 3D digitisers [61–70]. It is generally accepted that CBCT measurements are valid and accurate, but the threshold used for 3D reconstruction and demarcation of tissues may be a significant factor that needs attention.

The choice of the threshold for reconstruction and measurement is difficult because tissues may appear more or less dense than expected, depending on the area being examined and the effect of artefacts. Among the most significant artefacts in CBCT imaging are noise, the partial object effect and the partial volume averaging effect [71, 72]. Noise is present due to reduced x-ray energy, purposely set low on CBCT machines in order to reduce patient exposure. The partial object effect arises because the field of view is smaller than the object under investigation; the parts of the patient’s head that lie outside the field of view may significantly alter the density of the voxels, causing inconsistencies between tissue density and voxel density. These inconsistencies do not allow a reliable correspondence between voxel density and Hounsfield values and thus do not allow reliable differentiation between tissues, based solely on voxel values [73–75]. The partial volume averaging effect is evident when the voxels are large relative to the size of the object under investigation, resulting in multiple tissues occupying the space of a single voxel. In such cases, the value of the voxel will represent the average of the multiple tissues, giving a false impression of the voxel’s composition.

From the above it is evident that a global threshold for tissue segmentation and measurement may not be appropriate [76]. Especially prone to systematic errors are thin structures, such as the alveolar covering of incisor roots. Due to reduced resolution of CBCT images [77], alveolar bone is consistently ­underestimated and dehiscences and fenestrations are significantly overestimated [78–81]. However, high-contrast globular structures, such as teeth, also show errors in measurement which may exceed 1 mm, even if images are taken under ideal conditions, mainly because of the inconsistency of voxel values due to the artefacts mentioned above [82].

Reliability of CBCT measurements is also a significant concern. As with conventional 2D cephalograms, point identification seems to be one of the largest sources of error. CBCT images, due to their 3D nature, present new challenges to the user, as multiple views of the dataset may be needed in order to identify a landmark, including 3D reconstructions and sections through the volume. Some points may even require new definitions; for example, the external auditory canal extends in depth and follows an oblique path, so Porion’s traditional 2D definition needs updating [83]. Such difficulties reflect on repeatability; errors of measurement may exceed several degrees or mm, casting doubt on the capability of 3D measurements to reveal small treatment changes [84].

Laser scanning for 3D measurement and analysis of facial shape were introduced in orthodontics by the group of Moss in England [87–90]. Laser scanning of the head involves projecting a stripe of laser light and registering the shape of the stripe as it reflects from the three-dimensional surface. The stripe is usually projected as a vertical line and rotated around the face so that it scans the head from one side to the other. Scanning can be performed by movement of the laser and camera assembly or by rotation of the patient seated on a motorised chair. Typical scans of the face may take 10–20 s. Other manufacturers use a horizontal laser stripe that scans the head vertically. This system is much faster than the rotational scan (less than 1 s) but requires multiple shots to acquire the whole face.

Structured lighting methods do not use a single laser stripe but project a complex light pattern over the whole object and acquire the 3D information from a single image. The main advantage of this method is the reduced exposure time, making it possible to register dynamic movements and facial expressions.