Face Anthropometric Pattern Recognition Technology for Computer Aided Diagnosis of Human Genetic Disorders. (DW-6/2007)

• ClinicalTrials.gov Identifier: NCT00705055
• Recruitment Status: Completed
• First Posted: June 25, 2008
• Last Update Posted: August 9, 2017
• Sponsor: Carmel Medical Center
• Collaborators: Soroka University Medical Center; Technion, Israel Institute of Technology
• Information provided by (Responsible Party): Carmel Medical Center

Study Description

Brief Summary:

The hypothesis to be tested: After the construction of a database of anthropometric measurements, the system would extract important features of a given facial surface and be able to match it with existing morphometric figures. A given combination of normal and abnormal measurements will open a "probable diagnosis" and a list of "differential diagnosis" that will be expressed as percent of matching in a descendent order to the examiner.

Condition or disease
Genetic Disorders

Detailed Description:

Summary of Relevant Background Studies: Congenital anomalies play a major role in pediatric care. One of the leading causes of infant mortality in developed countries is the sequelae of these congenital anomalies. In some cases this exceeds the death rate for prematurity, SIDS, and other common causes of infant or neonatal death. The available tools for the assessment of a dysmorphic infant or child are based mainly on the experience of the examiner and his ability to translate findings and measurements from the physical examination to a qualitative and quantitative summary of accepted values plotted for the corresponding age (1,2). Various unusual features are expressed in qualitative terms such as short stature, long fingers, pear-shaped nose, small ears or other terms, which imply a comparison with other body proportions and the subjective impression of the examiner. Following that, an impression of the patient as a 'gestalt' is formed in the examiner's mind.

The databases and most of the written material are descriptive with scarce graphic and photographs, making the comparison of the phenotypic expression of the described subject with the one that needs to be diagnosed a difficult one. Even with the extensive existing data on objective measurements available to characterize a phenotype, many of the physicians involved in the diagnosis of a specific case will base part of their diagnosis on "it looks like" and put that impression in the context of other physical and laboratory findings.

Many syndromes in human pathology are recognized by their unique and distinctive facial and body characteristics. These stereotypic phenotypic characteristics are mostly reproducible using anthropometric measurements.

Charts are available for nor+mal data values of various morphometric variables (1,2). However, some of these figures can be accurately measured only on 3D structures (head, face). The following figure demonstrates the measurement of the angle of the palpebral fissure:

Fig. 1: An upward obliquity to the palpebral fissures is known as a mongoloid slant while downward obliquity is referred to as an antimongoloid slant. In order to obtain such measurements within the uterus a 3D configuration and appropriate image analysis is necessary (Figure from Ref. 2).

Fetal alcohol syndrome (FAS) is an example of a syndrome that underwent characterization by graphic data analysis methods (3). The prevalence of fetal alcohol syndrome (FAS) was determined in a foster care population and evaluated the performance of the FAS Facial Photographic Screening Tool. The authors concluded that the screening tool performed with very high accuracy and could be used to track FAS prevalence over time in foster care population to accurately assess the effectiveness of primary prevention. An expert can recognize facial characteristics and provide accurate analysis. Objective measurements could provide less experienced observers with tools that classify anatomical characteristics of different diseases and syndromes. Facial phenotypic patterns can be extracted from large databases of facial surfaces. These biometric measurements can be used for analysis when evaluated with respect to their "normal" values in the general population.

3- Methods of Study: Following approval of the Helsinki Committee, the project will be performed in several successive steps as follows: A. Newborn scanning: A database of 3D pictures (scans) of the face of newborn infants will be created. The scanning will be performed initially at the Carmel Medical Center, during their hospital stay. The examinees will be scanned one time, in order to build a database based on the data obtained from each scanned picture.

The facial anthropometric patterns of the obtained 3D pictures will be studied off-line using a computerized face pattern recognition system developed and used at the Faculty of Computer Sciences at the Technion. The measurements obtained will be compared to geometric anthropometric data already in use by medical geneticists and clinicians (1-9).

B. Hardware and software description:

3D Image Acquisition: Special hardware specially prepared in our department was developed for 3D image acquisition of newborn (see figure 2).

The hardware consists in:a structured light projector (DLP Projector Casio 350j,a digital video camera (PTGray Flea CCD Camera (Point Grey Research® Inc.( Black and white (640x480), Aluminum projector cage, Special medical stand with wheels,Personal Computer - Pentium 4 - XP,Flat screen 17" with stand mount,Firewire cables,I/O cables.

Systems used for image acquisition: Currently there are two basic technologies. One is a laser scan, where a narrow laser generated light plane scans a face in vertical direction and the 3-D structure of the face is recovered based on the form of the light contour at the intersection between the light plane and the face surface.

The second method is based on the so-called structured light technology (regular light), where one or more specially designed light patterns are projected onto a face, and the 3D structure is recovered based on the position measurements of known pattern elements projections on the face.

Next, the range image is converted to a triangulated surface. The mesh can be possibly sub-sampled in order to decrease the amount of data. The choice of the number of sub-samples is a tradeoff between accuracy and computational complexity. Using this technique image acquisition and reconstruction takes about 2-3 seconds.

C. Morphometric parameters and their computation: In order to compute common morphometric parameters like inner and outer cantal distance, interpupillary distance, etc., there is a need to recognize various points of interest in the 3D face. This will be done using various pattern recognition algorithms. At the initial stage a manual procedure will be used to mark features on the projection of the facial surface.

Based on the results of the first phase automatic methods will be developed to detect features using statistical and algebraic algorithms. After having the relevant anchor points secured, simple 3D geometry will be used to compute common morphometric data. Parameters include outer canthal distance, interpupillary distance, palpebral fissure length, palpebral fissure angle, nasal-labial (philtrum) length, ear length, ear height, etc. The 3D data available can be used to try and search for other parameters that might be considered as statistically meaningful indicators.

Another avenue of research is to examine the importance of other metrics for distance estimation. One option is to check the contribution of geodesic distances as indicators. Geodesic distance is a distance map computed on the surface itself (Riemannian metric). A minimal geodesic path is the shortest path on the surface connecting two points.

An efficient method for computing the minimal geodesic distances on a triangulated domain was developed by Kimmel and Sethian (10). As the face is a deformable surface it is important to use such a representation for the facial surface that the measurements performed on it would be invariant to possible deformations (i.e. various facial expressions). In this case a bending invariant surface representation introduced by Elad and Kimmel will be used (11).

D. Statistical methods will be used for detecting the best independent significant morphometric variables which significantly correlate with the various syndromes: Discriminating scores will be constructed using the regression coefficients of the multivariate analysis tests and best cutoff points will be found, predicting between different genetic anomalies. Testing of the method and of the results will be done using a validation group of patients and healthy controls, by independent observers. The syndromatic examined newborn infants will be assessed by a geneticist and confirmation of the diagnosis will be made by laboratory tests when appropriate.

E. Statistical Power and number of patients: Many morphometrical variables will be assessed, based on our 3D reconstructing methods. Only after applying the multivariate analysis on the results, the relative diagnostic importance of each variable will be revealed. Thus, no single variable can be considered at this point, as an absolute discriminator between normal and abnormal value. However, if considering for example only one 3D morphometric variable, such as the degree of palpebral fissure slanting, in order to discriminate between "Trisomy 21" and "normal" in the Caucasian population, the following power analysis can be computed: The average and SD values of the slanting eye angle in "normal" is: 3.5 (degrees) ± 1.5. In a Trisomy 21 patient, there is an upward shift of this value. In order to detect a shift of more than 2 SD's (i.e. of more than 3 degree) and supposing that the SD will be larger than 3 degrees in the pathological population we need a minimal number of 21 patients and controls, in order to obtain a statistical power of 90%.

Total number of patients: Our purpose is to obtain scans from 800 newborn infants during the two year study period.

Study Design

• Study Type: Observational
• Actual Enrollment: 1000 participants
• Observational Model: Cohort
• Time Perspective: Prospective
• Official Title: Face Anthropometric Pattern Recognition Technology, Based on 3D Reconstruction Technology, for Computer Aided Diagnosis of Human Genetic Disorders. Multicenter Collaborative Study.
• Study Start Date: November 2007
• Actual Primary Completion Date: December 30, 2013
• Actual Study Completion Date: December 30, 2013

Groups and Cohorts

Group/Cohort
1; Normal Males
2; Normal Females
3; Abnormal Males
4; Abnormal Females

Outcome Measures

Primary Outcome Measures :

1. To create the bases for a "normal patterns" database

Eligibility Criteria

• Ages Eligible for Study: 1 Hour to 2 Weeks (Child)
• Sexes Eligible for Study: All
• Accepts Healthy Volunteers: Yes
• Sampling Method: Non-Probability Sample

Study Population

Newborn infants born at Carmel Medical Center or at Soroka Medical Center

Criteria

Inclusion Criteria:

• all newborn infants born at the Carmel Medical Center following parental consent, and at Soroka Medical Center.

Exclusion Criteria:

• No parental consent; Facial deformation not related to chromosomal or genetic anomalies; babies transferred to the neonatal intensive care unit that need ventilatory support.

Contacts and Locations

Locations

Israel

Department of Neonatology, Soroka University Medical Center

Beer Sheva, Israel

Department of Neonatology, Carmel Medical Center

Haifa, Israel

Sponsors and Collaborators

Carmel Medical Center

Soroka University Medical Center

Technion, Israel Institute of Technology

Investigators

• Principal Investigator: Dan Waisman, MD; Department of Neonatology, Carmel Medical Center

More Information

• Responsible Party: Carmel Medical Center
• ClinicalTrials.gov Identifier: NCT00705055
• Other Study ID Numbers: CMC-07-0018-CTIL; DW 6/2007 ( Other Identifier: Carmel Medical Center, Dr. Dan Waisman, 2007 )
• First Posted: June 25, 2008
• Last Update Posted: August 9, 2017
• Last Verified: November 2013

Keywords provided by Carmel Medical Center:

Anthropometry; face recognition; genetic disease; newborn infant

Additional relevant MeSH terms:

Genetic Diseases, Inborn