All laptops had a camera connected to cap- ture iris images, and, if needed, the recognition process could be performed at any one of the four stations. During ...
Journal of the American Medical Informatics Association
Volume 13
Number 2 Mar / Apr 2006
233
Case Report n
Using Biometrics for Participant Identification in a Research Study: A Case Report PATRICIA M. CORBY, DDS, MS, TITUS SCHLEYER, DMD, PHD, HEIKO SPALLEK, DMD, PHD, THOMAS C. HART, DDS, PHD, ROBERT J. WEYANT, DMD, DRPH, ANDREA L. CORBY, DDS, WALTER A. BRETZ, DDS, DRPH A b s t r a c t This paper illustrates the use of biometrics through the application of an iris-based biometrics system for identifying twins and their parents in a longitudinal research study. It explores the use of biometrics (science of measuring physical or anatomical characteristics of individuals) as a technology for correct identification of individuals during longitudinal studies to help ensure data fidelity. Examples of these circumstances include longitudinal epidemiological and genetic studies, clinical trials, and multicenter collaborative studies where accurate identification of subjects over time can be difficult when the subject may be young or an unreliable source of identification information. The use of technology can automate the process of subject identification thereby reducing the need to depend on subject recall during repeated visits thus helping to ensure data quality. This case report provides insights that may serve as useful hints for those responsible for planning system implementation that involves participants’ authentication that would require a more secure form of identification. j
J Am Med Inform Assoc. 2006;13:233–235. DOI 10.1197/jamia.M1793.
Case Description Biometrics is the science of measuring physical or anatomical characteristics of individuals. Common biometric approaches include the recognition of fingerprints, hand or palm geometry, the retina, the iris, or facial characteristics.1 Biometric security applications use devices to capture, and computers to process, these characteristics in order to confirm or determine the identity of an individual. In longitudinal research studies and clinical trials, the data collected during enrollment and at different times are used to measure the frequencies of different outcomes across study participants. Observations, measurements, biological specimens, and other measurements of each study participant’s status are collected longitudinally. Thus, participant enrollment and identification are considered to be a crucial step and a primary responsibility of a coordinating center to ensure that all the data collected for a specific study participant Affiliations of the authors: Division of Pediatric and Developmental Dental Sciences (PMC, RJW, WAB), Center for Biomedical Informatics (PMC), Department of Epidemiology (RJW, WAB), Center for Dental Informatics (TS, HS), University of Pittsburgh, Pittsburgh, PA; Department of Oral Biology, Harvard University, Boston, MA (PMC); National Institute of Dental and Craniofacial Research (NIDCR), Division of Intramural Research, Bethesda, MD (TCH); Twins Institute for Genetics Research, Montes Claros, Brazil (ALC). Supported by NIH-NIDCR grants DE15351 and DE14528. The authors acknowledge the contribution of Cyntia C. Bastos and Cosme M. de Figueiredo during iris-scanning procedures. Correspondence and reprints: Patricia M. Corby, DDS, MS, Division of Pediatric and Developmental Dental Sciences, School of Dental Medicine, University of Pittsburgh, 3501 Terrace Street, Pittsburgh, PA 15261; e-mail: . Received for review: 01/09/05; accepted for publication: 11/17/05.
is unique and neither the participant nor the visit (or examination) has been misidentified.2 We conducted an extensive search of the literature and found no evidence of studies reporting success/error rates of traditional methods of participant identification in longitudinal studies or clinical trials that could be compared to iris recognition systems. The objective of this study was to implement and evaluate an iris-based biometric system for participant identification in a genetics study in Brazil. The Twins Institute for Genetics Research was established in 2000 to address the relative contribution of genetic and environmental factors on oral and systemic diseases. In that respect, a variety of phenotypic traits that include oral microbial parameters, salivary proteomics, immunological factors, and disease expression, among others, are being collected longitudinally that will allow for molecular genetic analyses. We present this case to illustrate the potential usefulness of iris-based biometric technology as a method of identification that can provide a greater degree of security than commonly accepted individual’s identification and authentication methods, especially in research settings. It also provides an evaluation of iris-based biometrics for research participant identification, its effectiveness, and shortcomings.
Methods The iris enrollment process begins with the acquisition of a high-resolution image of the eye, illuminated with infrared light. Based on visible characteristics, such as rings, furrows, freckles, and the iris corona, the technology maps the details of the iris and converts them into an IrisCode template.3 This 512-byte template is stored for future identification. To acquire iris images, the system used a small camera (Panasonic Model BM-ET100US, Matsushita Communication Industrial Co., Ltd., Yokohama, Japan). An enrollment
234
CORBY ET AL., Iris Recognition Technology less than 2 minutes, with the actual image capture and template generation taking about 15 seconds. After enrollment, it typically took less than 3 seconds for a person to be authenticated. During the enrollment process, we classified a full enrollment when a participant had acceptable images from eyes, a partial enrollment when a participant had an acceptable image from at least one eye, and a marginal enrollment when a participant had marginal images from both eyes. Unacceptable images were considered a failure to enroll. At baseline, the biometric system successfully enrolled 85% (n 5 986) of the participants (495 adults [94%] and 491 children [76%]). In this group, a total of 929 (94%) participants had acceptable images from at least one eye (472 adults and 457 children) and 57 (6%) participants (23 adults and 34 children) had marginal images from both eyes. Sixty-five participants, mainly young children, required more than one enrollment attempt. The system failed to enroll a total of 184 (15%) participants (155 children and 29 adults). Overall enrollment rates (n 5 1,170) by age and by type of enrollment can be seen in Table 1. With the pediatric population, the major reason for failure (unacceptable enrollments) was the lack of cooperation by young children 1.5 to three years old and system usability constraints related to the difficulties with interaction of the user with the camera device in the process of acquiring the iris image. An analysis of variance showed that there was a statistically significant difference (p , 0.0001) among the mean ages for each enrollment quality level. The average ages according to enrollment quality were (1) acceptable enrollment (n 5 457 [5.4 6 0.07]; (2) marginal enrollment (n 5 34 [4.0 6 0.25]); and (3) unacceptable enrollment (n 5 155 [2.5 6 0.12]).
database containing reference samples of up to 1,200 individuals was created by using commercially available iris recognition software (PrivateID Recognition Demo, V1.5, Iridian Technologies; Moorestown, NJ). This demonstration application performs both enrollment and recognition functions. The enrollment function used PrivateID image capture software (PrivateID V2.1, Iridian Technologies) to capture both the iris image and a face image and stored both the participant’s name and a randomly assigned study ID number in the database. The PrivateID software performed capture, quality assessment, compression, and encryption of the iris images. The demo application was configured to run on as many as five laptop computers that shared a common database through a local Ethernet network. In addition to PrivateID 2.1 and the PrivateID Recognition Demo application, the client software included Microsoft, which was used to manage the demographic database. The server-side product, KnoWho Authentication Server, received images from the ‘‘client,’’ decrypted, unpacked, and used them to generate a standard IrisCode template. This template was then used for either enrollment or recognition. After processing the images collected for enrollment, the application provides feedback on the quality of each captured image on-screen: green for an acceptable image, yellow for a marginal image, and red for an unacceptable image. An enrollment based on an unacceptable image is never saved by the system. Research and biometric data were collected on four laptops, which were used for the enrollment/identification process at registration and subsequently to capture data at the various clinical stations. All laptops had a camera connected to capture iris images, and, if needed, the recognition process could be performed at any one of the four stations. During our baseline study, an identification number was created for each participant. The identification number generated by the biometric system was used as a reference in our demographic database and linked to the participant’s study identification number created at baseline. After a brief explanation of the system functionality to study participants, the images of both eyes were captured and enrollment templates generated. Personal identification cards were created and included the participant’s picture, study ID number, and the biometric ID number generated by the system. Similar procedures were performed one year after baseline measurements.
Our longitudinal data were collected one year later, at which time we performed a second study-related screening of 646 children seen at baseline (only children were screened longitudinally). All participants, prior to screening, were required to perform biometric authentication. The biometric system successfully identified 488 children (out of 491) with acceptable enrollment at baseline. Three participants enrolled at baseline could not be identified successfully by the biometric system because, on them, we were only able to capture marginal images from both irises during the initial enrollment. However, 31 individuals with marginal initial enrollment were identified successfully. Once the identification process was completed, the identity displayed by the system was cross-checked with each individual participant’s study identification card. In addition, 77 of the 155 children, who failed to enroll at baseline, were successfully re-enrolled during the second appointment.
Example The present study had 1,170 participants (524 adults and 646 children), all of whom consented to be enrolled through iris scanning. On average, the time required for enrollment was
Table 1
j
Enrollment Rates by Age and Type of Enrollment
Age 1.5–3 yr old 3–6 yr old 7–8 yr old Adults Total
Full Enrollment* 69 239 86 431 825
(5.90%) (20.43%) (7.35%) (36.84%) (70.51%)
Partial Enrollment* 28 29 6 41 104
(2.39%) (2.48%) (0.51%) (3.50%) (8.89%)
Marginal Enrollment* 16 17 1 23 57
(1.37%) (1.45%) (0.09%) (1.97%) (4.87%)
Failed to Enroll* 144 10 1 29 184
(12.31%) (0.85%) (0.09%) (2.48%) (15.73%)
Total* 257 295 94 524 1,170
(21.97%) (25.21%) (8.03%) (44.79%) (100%)
Prob . x2: ,0.0001. Full enrollment 5 a participant had acceptable images from both eyes; partial enrollment 5 a participant had an acceptable image from at least one eye; marginal enrollment 5 a participant had marginal images from both eyes; failure to enroll 5 unacceptable images. *Number (%).
Journal of the American Medical Informatics Association
Volume 13
Discussion Although the iris-based biometric system implemented in this study had several weaknesses, especially with the pediatric subjects (we failed to enroll 24% of the children), it demonstrated outstanding performance during the process of recognition, especially in avoiding misidentification with identical twins. Reliable identification was particularly important in this study, as correctly matching the genetic information to the individual was a stringent requirement for the validity of our research data. In the longitudinal analysis, the system proved to be a reliable identification tool. Most difficulties in using the system arose as a result of the young age of our study population and not as a result of any inherent technical limitations. Fifty percent of the children at baseline were younger than four years old, and it was in this age group where most problems occurred. Capturing the iris image in those children was problematic because they could not reliably position and then hold their head still in front of the camera. Thus, an initial conclusion based on this study is that iris-based biometrics is not practical in children younger than approximately three years old. Some participants could not open their eyes wide enough and thus the camera could not capture complete images of their irises. In our study, we adjusted the threshold of the system to allow marginal images to be captured during our baseline enrollment and consequently were able to increase our enrollment rates. Although 34 of the pediatric subjects enrolled in our database had ‘‘marginal’’ images from both eyes, only two of them were not recognized with success during our longitudinal study one year later. At follow-up visits, participants were more familiar with the system and usability problems were almost nonexistent. In instances where we made many attempts to enroll a participant, we observed that those individuals tended to have very dark eyes. We hypothesized that the scanner had difficulty reading the iris in such instances. There is evidence suggesting that the iris texture is affected after cataract procedure surgery, consequently affecting the process of recognition due to loss of large areas of Fuchs’
Number 2 Mar / Apr 2006
235
crypts, circular and radial furrows, and pupil ovalization.4 Some participants can be identified successfully and others cannot. However, in these cases, no false match can occur; in other words, no participant will be identified as another individual because of iris textural changes. The recommendation is to ask subjects to re-enroll in the biometrics system for the creation of a new iris template.4 We narrowed this report to the ‘‘twins’’ aspect since this was the focus of our research and at the same time a reliable population for testing such system. The benefits of biometrics usage and deployment in this case report were derived from having a high degree of certainty regarding an individual’s identity. Nevertheless, the underlying principle of biometrics may be well suited to any population where secure methods of identification are required. We are not aware of any previously published study on iris recognition success/error rates for participant identification in cross-sectional and longitudinal research. We envision future applications of biometrics in comprehensive longitudinal cohort studies and multicenter clinical trials. The use of this technology can automate the process of subject identification, thereby reducing the need to depend on subject recall during repeated visits, thus helping to reduce misclassification errors or missing data. Biometrics can be a valuable aid to ensure correct matching of research data to individuals, and therefore, it may potentially elevate research data quality. References
j
1. Nanavati S, Thieme M, Navanati R. Biometrics. Identity verification in a networked world. New York: John Wiley & Sons, Inc., 2002, p 9–22. 2. Gassman J, Owen W, Kuntz T, Martin J, Amoroso W. Data quality assurance, monitoring, and reporting. Control Clin Trials. 1995; 16:104–36. 3. Daugman JG. High confidence visual recognition of persons by a test of statistical independence. IEEE Trans Pattern Anal Machine Intell. 1993;15:1148–61. 4. Roizenblatt R, Schor P, Dante F, Roizenblatt J, Belfort R. Iris recognition as a biometric method after cataract surgery. Biomed Eng Online. 2004;3:2.
