Understanding Normal Variation
The acquisition and understanding of normative data are the foundations for investigating the relationships between facial variation, health, and disease. A number of extant and evolving geographically dispersed datasets with variable characteristics [Hammond and Suttie, 2012] have been ascertained and include the Perth Face-Space which, when combined with related disorder sets, consists of an expanding set of approximately 1,500 imaged individuals. Optimally, this information would be coordinated in a global nodal network with common data elements; and to initiate this process, existing datasets and infrastructure should be consolidated and coordinated.
To understand normative variation, population groups underrepresented in current datasets would be included. This mirrors the situation with genetic studies that often exclude these populations, such as Indigenous Australians, who thus do not derive the full benefit of biomedical innovation. Addressing these issues may help to avoid the perpetuation of health disparities in these populations [Baynam, 2012; Kowal et al., 2011]. Furthermore, considering developmental considerations, longitudinal studies initiated in early childhood will be important and, by corollary, studies in elderly populations may also provide unique insights.
A Focus on Rare Diseases
Each rare disease (RD) is, by definition, of low prevalence. However, because there are 5,000–8,000 RDs, when taken cumulatively, the number of affected has been estimated at 1 per 10–18 people, or about 6–10% of the population [Donaldson, 2009; Knight and Senior, 2006; Minister of Health and Social Protection et al., 2004]. This cumulative prevalence equates to an estimated 1.4 to 2 million Australians, including more than 400,000 children and 27 million Europeans. Approximately, 80% of RDs are monogenetic or have a strong genetic basis, and some rare genetic syndromes have characteristic facies that have diagnostic utility. Accordingly, a number of studies have objectively documented facial phenotypes of individual RDs [Cox-Brinkman et al., 2007; Hammond et al., 2004, 2012; Tobin et al., 2008]. This work may form the foundation for developing tools that facilitate disease diagnostics, screening, differential diagnosis, and monitoring of therapy.
Normative studies of young individuals and of those affected with RD will be of paramount importance as an estimated 65% of RD appear in early life [Minister of Health and Social Protection et al., 2004] and timely diagnosis is central to optimal outcomes. Factors supporting facial analysis as particularly suitable for very young individuals include (1) it poses no health risk; (2) capture time is fast in comparison to other medical imaging; and (3) if required, image capture can be easily repeated until a suitable image is obtained [Kung et al., 2012].
Novel approaches to 3D facial analysis, including the normal equivalent, have been developed that approach the issue of the rarity of individual RD [Claes et al., 2010b,c; 2012a,b; Hammond and Suttie, 2012]. These techniques allow for a more individualized assessment of facial dysmorphology. In an exploratory study, one of these techniques, dysmorphometrics was used to establish a facial signature for a rare and treatable disorder that had variable facial severity proportional to the clinical disease state. This finding will potentially provide adjunctive evidence for noninvasively monitoring treatment response in this condition [Kung et al., 2012]. In 2011, 33% of marketed innovative medicinal products in the United States were intended for an RD (FDA, 2011). With the expansion of RD therapeutics, developments in treatment monitoring will be required. Equally, for those disorders for which treatments are not currently available these techniques will be important for objective natural history studies. The fine-scale, objective, scalable, cost-efficient, noninvasive, nonirradiating, and relatively portable nature of 3D facial analyses make these technologies unique candidates for these applications. It is likely that similar methodologies can be extended to nonfacial imaging, for example, for assessment of disproportion, and other aspects of habitus, by whole body scanning in those with skeletal dysplasias; or assessment of digital changes in the mucoploysaccharidoses. Additional opportunities are provided by the ability for joint investigations with traditional imaging technologies. For example, paired with MRI for face–brain studies [Hammond and Suttie, 2012] or with cone–beam CT for skull–face assessments [Cheung et al., 2011]. Lastly, the relative inexpensiveness of this technology makes it suitable to provide phenotypic data to existing disease registries.
Understanding individual RDs has provided fundamental insights into basic biological processes and into the causes of common diseases. A recent example is the perspective provided from relationships between a rare lysosomal storage disorder, Gaucher disease, and the more common Parkinson's disease, which may yield a specific therapeutic approach to Parkinson's disease and related conditions [Mazzulli et al., 2011]; 3D facial analysis of RD may ultimately benefit common diseases. Along related lines, by endophenotyping and/or by elucidating biological pathways, a more speculative use of 3D facial analysis may be to assist with drug rescue, that is, research involving discontinued drugs or candidate drugs that are not currently in development, and drug repurposing, that is, research on approved drugs for new indications. Pharmaceutical companies and biotechs might be able to employ 3D facial analysis research to scan their libraries for promising drug compounds or investigate new drug entities. These screening applications may be a cost-effective mechanism compared to current new drug entity discovery which equates to over $1 billion dollars per new drug [Dimasi and Grabowski, 2007]. This approach may be particularly relevant given the incentives and regulatory environment surrounding orphan drug development.
Human 3D facial analysis, when combined with other technologies, including facial analysis of mice, has been used to investigate disease biology [Tobin et al., 2008]. Between species (e.g., mouse–human) comparative phenotypic assessments will yield further insights into disease biology and such comparisons will be facilitated by Mouse phenotypic, and genetic, resources that are being coordinated by the International Mouse phenotyping consortium (IMPC) and related projects [Chen et al., 2012]. Notably, the IMPC draft phenotyping pipeline includes dysmorphology [Schofield et al., 2012]. By the way of a further example of the potential of cross-species comparisons, recently a study demonstrated large-scale objective association of mouse phenotypes with human symptoms through structural variation indentified in patients with developmental disorders [Boulding and Webber, 2012]. Furthermore, explorations of disease biology, and other potential applications, may be aided by emerging methods that facilitate between-study integration, interrogation, visualization, and exchange of phenotypic and genotypic information [Adamusiak et al., 2012; de Bono et al., 2012; Pan et al., 2012].
It has been suggested that the most important responsibility of the physician is to observe the phenotype of their patients, and this has become particularly relevant in an era where genotype–phenotype relationships are rapidly being clarified by the use of high throughput molecular technologies. Regarding these genotype–phenotype investigations: methods for capturing and analyzing phenotypic data are one of the major bottlenecks in the understanding of human genome biology [Robinson, 2012]; next-generation sequencing (NGS) will require next generation phenotyping (NGP) [Baynam, 2012; Hennekam and Biesecker, 2012]; the full potential of phenome databases, such as OMIM, requires the aggregation of phenotypic information from multiple databases which depends on the availability of fine-grained phenotype ontology and feature frequency data [Oti et al., 2009]; and delineating the phenotypic components of disease allows (1) relationships between gene function and phenotype to be established, where one aspect of the phenotype is common to several genes in the same pathway, and (2) the discovery of interrelationships between disease pathways and genetic networks [Schofield and Hancock, 2012]. NGS has been described as the most powerful diagnostic tool developed since the roentgenogram [Hennekam and Biesecker, 2012]. Coupling NGP methods such as 3D facial analysis with NGS, and other -omics technologies, will enhance the use of these technologies in delineating disease causation. These factors support 3D facial analysis contributing toward reductions in the burden of RDs. In addition, the objective nature of 3D facial analysis may compliment efforts to develop standardized descriptions of human phenotypic variation, such as the Elements of Morphology Project, that, in partner with standardized molecular nomenclatures, aim to facilitate robust genotype-phenotype correlations [Carey et al., 2012]; incorporation of phenotypic information into genomic variation databases may enhance clinical care and research [Riggs et al., 2012].
Synergies and Future Directions
Unique aspects of 3D facial analysis provide the potential to develop traditional and novel collaborative networks. In addition to existing fruitful collaborations between Computer Engineers, Clinical Geneticists, Paediatricians, Dentists and Surgeons, networks could be expanded to include a diverse range of medical professionals and scientists. The engineered bases of these techniques will also benefit from progress within the greater bioinformatic community. Further input may be provided from commercial bodies, for example, noting the similarities between financial risk analysis and diagnostic algorithms, and between geological surface analysis and analysis of human surfaces.
There is a relative dearth of genetic and phenotypic information in some developing nations and amongst some populations, by virtue of geographic isolation or other factors. The low-cost and portability of 3D scanners can address these issues.
When assessing an individual's facial morphology, Clinical Geneticists often “subtract” familial variation obtained from direct observation of a consultand's relatives or familial photographs. The ability to develop objective familial imaging to assess this, that is, “familial morphometrics,” will improve the sensitivity and specificity of 3D facial analysis. In general, morphometric techniques, an object's or organism's form (a concept that encompasses size and shape independent from orientation or position) is the structural information and the conventional observation of interest. These observations, however, are made in isolation and a context-based observation of shape currently does not exist. The introduction of context information as in computational linguistics [Qiu et al., 2011], can lead to an improved understanding of observations made. The idea is to combine contextual information with structural information already coded in the shape of an individual's face through the relationship of that individual's facial morphology with that of its parents. Such a combined contextual and structural model is novel within morphometrics and allows for unique solutions, including for applications to sparse and rare datasets. These analyses will need to be distinct to independent assessments of family members, as recognizing, rather than controlling for, parental facial variation may provide its own insights. For instance, although the significance is currently unknown, atypical facial asymmetry has been reported in unaffected mothers of children with Autistic spectrum disorders [Hammond et al., 2008].
The digital nature of 3D facial data facilitates integration with other digital data. In addition to integration with other -omics technologies, there are a number of variably speculative avenues that could be pursued. Firstly, the promulgation of electronic medical records and text mining techniques, such as natural language processing, when combined with human phenotype ontology and 3D facial data, might generate novel diagnostics. Secondly, crowdsourcing may provide novel solutions. The success of foldit (foldit)and phylo (phylo) in obtaining solutions by utilizing a mass of individuals, that largely may have no formal scientific training, to solve complicated scientific problems is encouraging for further applications. For instance, analogous to sequence alignment in phylo, phenotype alignment may be possible. Although privacy concerns may preclude use of data which can visually identify an individual, the use of composite data from groups of patients, or the use of metadata may be suitable approaches. Some individuals have extraordinary facial recognition ability and they have been described as “super-recognizers” [Russell et al., 2009]. Those with inherent facial pattern recognition ability may be passively identified by crowdsourcing approaches. More generally, closing the bioinformatic gap between research and healthcare may be aided by knowledge engineering, that is, integrating knowledge into computer systems to solve complex problems typically requiring a high level of human expertise which may be facilitated by projects such as the proposed I4 health concept [Beck et al., 2012]
There are current emerging clinical applications of 3D facial analysis and a number of translational and primary research avenues of investigations which include those summarized in Table 1. Importantly, the human network remains central to fostering research and translation of research toward clinical outcomes. “The world does not connect by molecules. It connects through ideas, hopes, FACES, dreams, actions, stories and memories,” Barrie Sanford Grieff. The inherent intrigue and commonality of the human face provides a collaborative focus and a node for the development of interdisciplinary and community collaborations. Many successful examples of collaborations centerd around facial research exist and their expansion and interconnection with each other, other physical and intellectual infrastructures, and with active collaborative engagement of clinicians and researchers will be key to maximizing outcomes. Integration of global resources for human genetic variation and disease might proceed with a series of national or international hubs to hold phenotypic and genotypic data, which could be channelled to a central database [Schofield and Hancock, 2012]. From a RD health care perspective, the Orphanet approach to representation of RDs in health information systems is noted [Rath et al., 2012]. Applying coordinated approaches to RD will be a priority given the current international efforts that are supporting such collaborations, the remarkable impact that -omics technologies are making in this area, and given that rare conditions are disease models that help to understand the pathogenesis of diseases for the benefit of individuals with both common and RDs [Rath et al., 2012]
Table 1. Potential Clinical and Translational Applications of 3D Facial Analysis
|Facial differential diagnosis and screeninga||Dysmorphometric facial signatures of diseases/disease groups|
|Facial diagnosisa||Dysmorphometric facial signatures of individual diseases|
|Craniofacial surgical planning||Individualized assessments, for example, using the NE|
|Postsurgical assessment and audit||Individualized assessments pre- and postprocedure|
|Deep phenotyping to facilitate interpretation of genomic information||Paired 3D facial-genomic (e.g., next-generation sequencing) studies|
|Iatrogenic morbidity reduction||Replacement of ionizing radiation, for example, interposition into craniofacial surgical protocols; replacement of invasive investigations, for example, surface imaging as a proxy for polysomnography for OSA, or hematological monitoring of drug response|
|Cost reduction||Replacement of expensive investigations in selected situations; early diagnosis|
|Natural history studies as a foundation for prognostication and therapeutic trial||Longitudinal individualized assessments|
|Treatment monitoring||Longitudinal individualized assessments|
|Drug repurposing||Comparative facial analysis between disorder groups|
|Novel diagnostics using contextual information||Digital integration with imaging and other investigations, for example, paired face-neuroimaging studies; integration with data in medical records utilizing techniques including natural language processing; or familial morphometrics|
|Crowd sourcing||Identify superrecognizers; phenotype alignment|
|Exploration of biological pathways||Comparative (between and within disorder) facial analysis; comparative (between species, e.g., mouse and human) facial analysis|