In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Voi. XXXVIII, Part 7B
618
ANALYSING THE FACIAL MORPHOLOGY WITH A THREE-DIMENSIONAL
GEOMETRICAL FEATURES APPROACH
F. Calignano a ’ ,S. Moos a , E. Vezzetti a *
a Politecnico di Torino, Dipartimento di Sistemi di Produzione ed Economia dell’Azienda, 10129 Torino, Italy -
(flaviana.calignano, sandro.moos, enrico.vezzetti)@polito.it
KEY WORDS: 3D scanner, Shape analysis, Facial morphology, Soft tissue shifts
ABSTRACT:
To obtain the best surgical results in orthognathic surgery, treatment planning and the evaluation of results should be performed. In
these operations it is necessary to provide to the physicians powerful tools able to underline the behaviour of soft tissue. For this
reason, considering the improvements provided by the use of 3D scanners, as photogrammetry, in the medical diagnosis this paper
proposes a methodology for analysing the facial morphology working with geometrical features. The methodology has been tested
over patients affected by malocclusion, in order to analyse the reliability and efficiency of the provided diagnostic results.
1. INTRODUCTION
The assessment of the dimensions and arrangement of facial
soft tissues is important for medical evaluations. Orthodontists
and orthognathic maxillofacial and plastic surgeons often
require quantitative data about the association between soft and
hard tissues (Ferrario 1999, Sforza 2005). For many years, this
information has been obtained from two-dimensional (2D)
radiographs and photos, even if these have been consistently
limited (Chew 2005,Koh 2004). Significant improvements have
been obtained with the use of computer vision algorithms, even
if the use of 2D supports to analyze 3D objects seems to be
quite inadequate. For this reason, many research efforts over the
last 10 years have been directed toward developing computer
vision tools that with the use of 3D scanner devices are able to
provide reliable and more complete data. These systems use
different technologies such as active or passive light reflection
analysis and are able to describe 3D real shapes with a point
cloud, analyzable with 3D software. However, while image-
processing methodologies are well known in the medical
context, the situation for 3D scanners is still quite marginal and
fragmented. Some studies have proposed structured procedures
that could be used for guiding physicians in the use of 3D
scanners in medical diagnosis (Hoffman 2005, Katsumata
2005), but at present no one has succeeded in developing a
standardized and accepted strategy.
Actually, it is possible to move from morphometric tools that
implement statistical shape analysis, such as Generalised
Procrustes Superimposition (GPS) and Principal Component
Analysis (PCA). The first iterative method (GPS) applies
geometrical transformations (scales, translations, rotations, and
reflections) in order to compare reference points (landmarks)
(Mori 2005, Sforza 2004) taken from different point clouds of
the patient’s face. The PCA method evaluates the tendency of
the landmark distributions along the x and y axes, locating a
new working frame centered on the average shape center. The
method creates new variables named principal components
(PCs) that describe how much the landmark configuration of
each sample is different from the average shape.
Moving to the 2D radiographs, Thin-Plate Spline analysis (TPS)
is used on a point set of anatomical landmarks over the pre- and
post-surgery radiograph. Then the postsurgery radiograph is
considered as an infinitely thin metal plate that must be bent in
a direction orthogonal to the plane in order to match its
landmarks to those of the presurgery radiograph while the
bending energy is minimized (Hajeer 2004) If the two shapes
are identical, the bending energy is zero and the plate is flat.
To provide information regarding face morphology in the
regions around the landmarks, the Multisectional Spline method
uses section planes passing through a set of specific reference
points of a point cloud (landmarks) in order to obtain specific
section spline. The shifts of the facial morphology between the
pre- and post-surgery point clouds can be analyzed by
comparing the two section profiles passing through homologous
landmarks and section planes (Soncul 2004).
Working with the entire point cloud instead of only some
portions, with the Clearance Vector Mapping method CVM) the
pre- and post-surgery point clouds are first aligned (iterated
closest point (ICP), CSM) (McIntyre 2003) and then the
magnitude of the 3D shape displacement can be computed by
working on triangulated meshes, following different approaches
(radial, normal) (Harmon 1981). The displacement is shown by
color mapping.
At present, even if the most used methodology for maxillofacial
diagnosis remains the Conventional Cephalometric Analysis
(CCA) (Bookstein 1991), the Multisectional Splines method
seems to be the most reliable and complete methodology
because it is able to provide reliable information about the
tissue shifts, as does the CCA approach, but it is also able to
provide additional global information, e.g., some pathologies
such as lateral asymmetry.
Some significant points need work for development of a
diagnostic procedure that could be accepted by the entire
medical context. It is necessary to define a method that extracts
shape morphology measures, starting from the landmarks as
reference points, to guarantee consistent morphological
comparison, but that also takes into consideration the entire
facial shape (point cloud) in order to consider all useful
* E. Vezzetti, Dipartimento di Sistemi di Produzione ed Economia dell’Azienda, Politecnico di Torino, Corso Duca degli Abruzzi
24, 10129 Torino, Italy, email: enrico.vezzetti@polito.it