Anthropometric approaches to measuring the face: morphics

imaging, and modeling.

By Bradly AliceaFRG, Michigan State University,

Fall, 2006

Classes of Morphological Traits

1. Metric traits

2. Nonmetric traits

Visual recognition of landmarks, easily recognizable landmarks are scored using a binary categorization system.

Point-to-point measurementsof landmarks,sometimes with calipers or other metric device.

Morphological Traits (con't)

Morphological Traits (con't)

1- and 2-D Morphometrics (landmark measurements)

Measurements that can be taken with calipers or flat grids can be regarded as one(or two)-dimensional measurements.

While caliper-based measurements are good for finding the distance between landmarks, they do not completely characterize the complex topology of a face.

Caliper measurements are good for characterizing growth processes or intraspecific variation, however.

1- and 2-D Morphometrics (landmark measurements) (con't)Butler, M.G., Levine, G.J., Le, J.Y., Hall, B.D., Cassidy, S.B. (1995). Photoanthropometric study of craniofacial traits of individuals with Prader-Willi syndrome. American Journal of Medical Genetics, 58(1), 38-45.

* photoanthropometric method, which enables an objective description offacial structures; 37 individuals with Prader-Willi syndrome between the ages of 0 to 12 years.

* facial parameters were measured from strict frontal and profile photographic 35mm slides and compared with other facial measurements from the sameface (e.g., palpebral fissure width to bizygomatic diameter).

1- and 2-D Morphometrics (landmark measurements) (con't)Butler et.al, 1995 used 16 photoanthropometric craniofacial indices establishedby Stengel-Rutkowski et.al, 1984 (Human Genetics, 67, 272-295) and Butler et.al, 1988 (American Journal of Medical Genetics, 30, 165-168).

* none of the parameters were consistently outside of the normal range when compared with photoanthropometric index standards for age established from white control children.

* Several features characterize population: narrow palpebral fissure width in ages 6-12, high midface, broad interalar distance, short back of the nose, prominent high chin, and broad low-set ears.

1- and 2-D Morphometrics (On growth and form…..)

In 1903, D'Arcy Thompsonwrote a book called “On Growth and Form”.

* biological growth and variation is a product of processes that can be captured mathematically.

* “Thompson grids” can beused to capture variation across specimens (using a landmark on one specimen, the grid is “warped” to fit another specimen.

1- and 2-D Morphometrics (On growth and form…..)

1- and 2-D Morphometrics (On growth and form…..)

“On Growth and Form” argues that biological growth is not only geometrically efficient, but also proceeds algorithmically (fractal, Fibonnaci series).

On growth and form......Mitteroecker, P., P. Gunz, and F.L. Bookstein (2005). Heterochrony and Geometric Morphometrics: a comparison of cranial growth in Pan paniscusand Pan troglodytes. Evolution and Development, 7(3), 244.

* variation during development relies on a univariate concept of shape. The authors took a multivariate set of measurements to calculate an “ontogenetic trajectory” for facial growth.

“Appearance” is operationalized as a set of principle component plots representing 1) shape space and 2) size-shape space. There appeared to be variation in heterochrony by region.

Insights from Morphometrics(Heterochrony)

Heterochrony is the "control" of growth during development. In evolution, this is how phenotypic traits have traditionally arisen and varied. In development, this is the mechanism for two phenomena:

* age progression (how "juvenile" or "mature" is a craniofacial trait?)

* proportionality between traits (allometric scaling by age, health)

* when is the onset of growth for a specific trait, and what is the growth rate for a specific trait?

Insights from Morphometrics(Heterochrony) (con't)

Heterochrony ultimately deals with variation in trait size due to timing in growth and devlopment:

Hypermorphosis: expansion of ontogeny (developmental period); this results in more developed (complex, larger) features in adults.

Hypomorphosis: reduction of ontogeny (developmental period); this results in more “juvenile-looking” traits.

Acceleration, retardation: speeding up/slowing down of growth rate (feature appears earlier, later in development).

Allometry: a scaling between the size of two traits (body size, brain volume).

Changes in the trajectory of growth during development changes are called “secular” changes. Secular changes are detected by using longitudinal datasets (data collected over several decades).

* compared to several decades ago, populations in Europe and Asia have shown several trends:

* birth weights and the rate of growth during childhood have both increased. The onset of puberty occurs earlier, while people are both heavier and taller.

* secular changes are usually due to the interaction of environmental and genetic factors (genes provide a range of growth conditions, environment works within these bounds).

Insights from Morphometrics(Development)

3-D morphometrics (skin + bones)

In many cases, we want a volumetric representation of the face.

One way to do this is through medical imaging; in biomechanical research, MRI and CT scans are often used to gain information about morphological structures and their underlying tissues.

* many times, we want to determine the deformational properties of tissuesunderlying the skin's surface.

* medical imaging allows us to acquire large amounts of digitized information which either be mined or used to build models not possible with simpler methods.

Human Biological VariationHuman biological variation is poorly understood in a taxonomic sense. Generally, there are four ways in which a trait can vary:

1) geographically; local populations may share traits as a consequence of both interbreeding and gene flow.

2) specific traits within a lineage (i.e. body types and appendage form among humans; can be specific to a family or several different types passed down in a single family).

2) developmental variability: variation across humans can be expressed differentially (due to lineage-specific traits and environment), but generally follow a specific trajectory across life-history.

Human Biological Variation (con't)

We do know that the human species consists of a series of interconnected “demes” (local breeding populations ~ ethnic groups ~ races) which provide a crucible for locally-adaptive traits.

4) species-wide similarity: specific traits within a range of quantitative values are present in all humans; differ from other species in their configuration, size, location on the head.

The result of this is that we can identify some traits on the basis of nationality (qualitatively) relatively easily; it is much harder to put this variation in a quantitative context based on random samples.

Human Biological Variation (con't)

Human Biological Variation:Neanderthal example

The first Neanderthal specimen (found in 1857).

It turned out that indivudal had severe arthritis (resulting in deformaties, >1 SD phenotypic difference).

* many craniofacial traits are plastic in evolution and development, as opposed to “conserved” features (dental formula, # of vertebrae). Many also coevolve, opening the possibility for modeling.

* training an artificial system on the “normal” range of human variation -- not enough?

1- and 2-D to 3-D measurementsOne potential application of these “big” ideas (i.e. growth and form,heterochrony, intraspecific variation) to computational systems is that we can build predictive models of variation.

Applied specifically to faces, this may be useful in a number of ways:

* age progression software. Many superficial packages exist, but are crude (1-and 2-D landmarks) and not informed by biological reality.

* surgical planning. Advanced knowledge of the face surface as a complex 3-D volume is necessary to repair facial morphology.

* internet-based applications. E-commerce or biometric applications (given the appropriate hardware on the other end), may be the most promising.

Complex 3-D AnthropometryTeschner, M., Girod, S., and Girod, B. (1999). Optimization Approaches for Soft-Tissue Prediction in Craniofacial Surgery Simulation. Medical Image Computing and Computer-Assisted Intervention (MICCAI), Cambridge, UK.

There are two principal methods of planning surgeries based on models:

* one method (stereolithography) is to build physical models generated from CT scans. Improve surgical planning and allow for the accurate manufacture of transplants. Physical models provide information on bone structure, but do not contain knowledge of soft tissues.

* another method is applying imaging (CT) data to computer models, which provide more information by integrating several modalities of information. Multimodal computer models of bone structure and soft–tissue dynamics can be used to simulating facial and plastic surgery.

Complex 3-D AnthropometryKeeve, E., Girod, S., Kikinis, R., and Girod, B. (1998). Deformable modeling of facial tissue for craniofacial surgery simulation. Computer Aided Surgery, 3(5), 228-238.

In craniofacial surgery, the prediction of soft-tissue changes resulting from alterations in the underlying bone structure is critical.

* still based on empirical studies of the relationship between boneand tissue movements.

Complex 3-D Anthropometry (con't)

Vannier, M.W., Marsh, J.L., and Warren, J.O. (1984). Three dimensional CT reconstruction images for craniofacial surgical planning and evaluation.Radiology, 150, 179-184.

3-D reconstruction images of bony and soft tissue surfaces have improved understanding of complex facial deformities.

* manually derived acetate tracings of cephalometric skull radiographs in the anterior-posterior (coronal) and lateral projections.

* equally-spaced, nonoverlapping, high resolution CT scans of the face are routinely collected before and after facial surgeries.

* reformatting in the sagittal, coronal, and oblique planar orientationscan be used to demonstrate complex structural anatomical features.

Complex 3-D Anthropometry (con't)

Contour extraction at other than soft tissue-aim or soft tissue-bone interfacesrequires an image-segmentation scheme more sophisticated than the simple thresholding or level sectioning.

* the authors use a method based on CAD design techniques. After extracting frontal soft tissue and bone contours from the original CT scans using level slicing, the conresponding point data was curve fit using cubic splines.

* 3-D surface geometry of frontal bony contours was encoded in fewer than 2,500 coordinates using 40 CT scans as input for a patient with asymmetric hypentelonism.

* soft tissue contours were separately entered in the system using the same technique. The contours were rotated to show the plane on top view. Aninfenolateral displacement of the left orbit was observed in the frontal views. A left frontal bone cleft was also demonstrated.

Complex 3-D Anthropometry

The above figures show various diagnostic marks of traumas, phenotypic landmarks, and a surgical planning blueprint (circa 1984!)

Towards Biomechanical Modeling

Marchetti, C., Fares, J.E., Sarti, A., Gori, R., and Lamberti, C. (2006). Maxillo -Facial Virtual Surgery from 3D CT Images. Teatro virtuale, Bioengineering. http://www.cineca.it/sap/teatrbioengvirtsurg.htm.

This paper describes an approach for elastic modeling of human tissue based on the use of embedded boundary condition techniques.

* embedded boundary condition models allow to simulate the cranio-facial surgery directly on the grid of the 3D CT image of the patient.

* approach proposed in this paper involves complete 3D modeling of the solid high-detailed structure of the object, starting from the information present in 3-D diagnostic images.

* application of this approach for modeling the elastic deformation of human tissue in response to movement of bones is demonstrated both on the Visible Human Data Set of the National Library of Medicine and on the CT datasets.

Towards Biomechanical Modeling

CT imaging provides resolution to the bones underlying facialanatomy in vivo.

Clockwise from upper left: mandibular fracture, orbital trauma, zygomatic fracture.

Deformations at the level of bone can affect the recognition of surface structures and landmarks.

Biomechanical ModelingDeuflhard, P., Weiser, M., and Zachow, S. (2006). Mathematics in Facial Surgery. American Mathematical Society. http://www.ams.org/notices/200609/fea-surgery.pdf

Authors approach this with biomechanical modeling. Model Maxillofacial (upper jaw, face area) deformity as a mass spring model. Model provides immediate response, but interactivity is obtained at the expense of numerical stability and approximation quality.

Optimal situation: a reliable preoperative prediction of the expected facial appearance on the basis of a detailed mathematical model based on individualanatomical variation.

Biomechanical Modeling (con’t)In mathematical terms, soft tissue deformation can be described by a mapping φ: Ω→ R3 from the undeformed reference domain Ω ⊂ R3 to its deformedcounterpart Ω ⊂ R3. Usually, computations are performed in terms of the displacement u = φ− I.

* relocation of bones prescribes acertain displacement on the Dirichlet interface ΓD ⊂ ∂Ω of bone and soft tissue.

* soft tissue may be modeled as a hyperelastic material. In this model, the deformation is given by minimizing the stored energy f (φ) = Ω W(∇φ) dx. Different material behavior is due to different material laws W.

Digital Human ModelingExample

Paquet, E. and Viktor, H.L. (2005). Anthropometric calibration of virtual mannequins through cluster analysis and content-based retrieval of 3-D body scans. Information and Measurement Technology Conference, Ottawa, ON.

* used the CAESAR database to calibrate virtual mannequins to fit “average”anthropometric data. Used by an e-commerce website; clothing of different sizes could be modeled using the mannequin based on user’s measurements.

* mannequins generated for their aesthetic qualities do not conform to normal human dimensions. 3-D body scans (in way similar to Mocap?) were made along with 49 anthropometric measurements; results were analyzed using a cluster analysis (EM algorithm, CobWeb method, and k-means classifier were all used to fit data).

Digital Human Modeling Example

* virtual mannequins having dimensions closest to cluster of human data (S, M, L, XL, XXL) were retained.

* 3-D body scan in the cluster closest in shape to virtual mannequin in same cluster was retained, used to tailor mannequin.

Other Valuable ResourcesThree tools of interest for anthropometry research:

1) Visible Human Project (National Library Of Medicine – NIH).Full body scans (MRI and CT), large sample sizes. http://www.nlm.nih.gov/research/visible/visible_human.html

2) CAESAR Anthropometric Database (who owns this? - SAE).Comprehensive measurements of 2400 U.S. & Canadian and 2,000 European civilians http://store.sae.org/caesar/.

3) Computerized Anthropometric Research & Design Laboratory (CARD –DoD). http://www.hec.afrl.af.mil/HECP/cardindex.shtml