Acquisition of the 3D surface of the palate by in-vivo digitization with Wave

Abstract

An accurate characterization of the morphology of the hard palate is essential for understanding its role in human speech. The position of the tongue is adjusted in the oral cavity, of which the hard palate is a key anatomical structure. Methods for modeling the palate are limited at present. This paper evaluated the use of a thin plate spline (TPS) technique for reconstructing the palate surface from a series of in-vivo tracings obtained with electromagnetic articulography using Wave (NDI). Twenty-four individuals (13 females and 11 males) provided upper dental casts and in-vivo tracings. Models of the palate surfaces were derived from data acquired in-vivo and compared to the scanned casts. The optimal value for the smoothness parameter for the TPS technique, which provided the smallest error of fit between the modeled and scanned surfaces, was determined empirically (the value of 0.05). Significant predictors of the quality of the fit were determined and included the individuals’ palate characteristics such as palate slope and curvature. The tracing protocol composed of four different traces produced the best palate models for the in-vivo procedure. Evidence demonstrated that the TPS procedure as a whole is suitable for modeling the palate surface using a small number of in-vivo tracings.

Introduction

An accurate characterization of the morphology of the hard palate is important for understanding human speech, which broadly encompasses basic motor speech science as well as the studies of disease- and age-related changes in speech production. Such characterizations are important to studies of dentition and feeding as well. Our particular future interest is in understanding the role of tongue position and tongue movements in the production and perception of speech sounds. The mapping of positions of the tongue in the vocal tract for different speech sounds is not trivial, and our understanding of the role of the palate is limited in this process, mostly due to challenges associated with accurate and cost effective ways of modeling the palate surface. The majority of existing palate work has been conducted in developmental conditions associated with abnormal development of oro-facial morphology, such as Down syndrome and cleft palate (Bhagyalakshmi et al., 2007, Hamilton, 1993, Vorperian et al., 2005). The morphology of the palate across other populations is severely understudied (Bhagyalakshmi et al., 2007).

The most common method of studying palates is via the dental cast (e.g., Burlington Growth Project contains nearly 8000 dental casts; Thompson and Popovich, 1977). By-hand measurements as well as digitization of important landmarks have been used to study specific palatal features (Ferrario et al., 2001). Measurements based on palatal casts (see Fig. 1 for an illustration), although accurate and can be viewed as gold standard, are expensive and time consuming to obtain; and casts are space consuming to store. In-vivo cephalographic methods are being introduced in dental practice, which could in principle yield palate models; they are highly accurate but still very expensive (see Kumar et al., 2008). In speech research, where the palate is viewed as a passive articulator, a most complete representation of this structure is accomplished using elecropalatography (EPG). The procedure requires making an artificial palate molded from a palate cast for each subject (Brunner et al., 2009). This procedure is also expensive and time-consuming. In the limit, the palate is represented as a 2D or 3D trace on a midsagittal plane, using technologies as such X-ray microbeam (see Westbury, 1994) and electromagnetic articulography (Perkell, 1998).

In this study we will use a 3D electromagnetic device (Wave, NDI, Canada) to digitize the palate surface in-vivo and propose a method of reconstructing of the palate surface from the in-vivo tracings, using a thin plate spline (TPS) method. TPS is a technique that fits a surface to a cloud of 3D data points, where the fit of the interpolation can range from exact to inexact, as determined by a factor that controls what can be seen as the “rigidity” of the surface or its ability to deform (also referred to as the smoothness factor denoted by lambda, λ). The TPS method was selected, as it is known to work extremely well with sparse and noisy data. This property makes it useful, since this technique is able to reconstruct the palate surface from traces that might not cover the entire area of the palate evenly, and from the traces that are affected by noise arising from the data collection technique. The overall goal of this study was to develop procedures that would optimize the TPS surface derivation technique (and thus the lambda) for a variety of hard palates across a population, characterized by relatively normal palate anatomy. The optimization was achieved by assessing the fidelity of the modeled TPS surface against the existing “gold standard”, the palate cast, which was scanned using a high-resolution ex-vivo method (i.e., a palate cast scanned using a laser scanner).

There is substantial variability in the size and shape of the hard palate across life span. In adults, the length of the palate is between 28.8 and 30.2 mm for females and males on average, as palatal width varies between 33.4 and 37.0 mm between females and males (see Ferrario et al., 2001 for details). The shape of the palate varies among individuals as well, with palates ranging from dome-shaped to flat as measured by the fit of a line in the coronal slice of the palate (Hiki and Iton, 1986, Mooshammer et al., 2004, Perkell, 1998, Weirich and Fuchs, 2011). The palatal shape variability has been linked to sex differences as well as variation in palate size (Fuchs and Toda, 2010). The fidelity of the representation of the palate surface is hypothesized to depend on the characteristics of each palate.

One specific goal of this study was to identify a TPS parameter value (lambda) that would result in an optimal surface fidelity of the digitized palate surface relative to the palate surface obtained by scanning dental casts. Additionally, the effect of variability in the tracing protocol across individuals on the fidelity of the TPS method was investigated. We hypothesized that the measurement error would be negatively correlated with the degree of coverage of the palate surface by the palate probe. Furthermore, we explored the impact of individual characteristics including sex, palate size and shape on the outcome. Our final goal was to propose best practices for in-vivo tracing of the palate that would be applicable across individuals.