J Forensic Sci, 2014 doi: 10.1111/1556-4029.12385 Available online at: onlinelibrary.wiley.com
TECHNICAL NOTE ODONTOLOGY Ashith B. Acharya,1 B.D.S., G.D.F.O.
Forensic Dental Age Estimation by Measuring Root Dentin Translucency Area Using a New Digital Technique*
ABSTRACT: Dentin translucency measurement is an easy yet relatively accurate approach to postmortem age estimation. Translucency area
represents a two-dimensional change and may reflect age variations better than length. Manually measuring area is challenging and this paper proposes a new digital method using commercially available computer hardware and software. Area and length were measured on 100 tooth sections (age range, 19–82 years) of 250 lm thickness. Regression analysis revealed lower standard error of estimate and higher correlation with age for length than for area (R = 0.62 vs. 0.60). However, test of regression formulae on a control sample (n = 33, 21–85 years) showed smaller mean absolute difference (8.3 vs. 8.8 years) and greater frequency of smaller errors (73% vs. 67% age estimates ≤ 10 years) for area than for length. These suggest that digital area measurements of root translucency may be used as an alternative to length in forensic age estimation.
KEYWORDS: forensic science, forensic odontology, age prediction, dentin transparency, area measurement, Adobe Photoshopâ, regression analysis
Postmortem age estimation is an important contributor to establishing identity of the deceased in forensic casework. Age estimation, in conjunction with sex assessment, stature prediction, and population designation, assists in reconstructive identification; alternatively, when law enforcers already have a putative age of the deceased, a forensic age estimate that is close to the presumed age provides the police clarity in their line of investigation. Since it was first proposed by Gustafson (1) as one of six parameters for adult age estimation, and backed by Johanson (2) as the most strongly correlated with age among the six, root dentinal translucency has gained immense interest and used as a single parameter in age estimation (3–7). Apart from its relative accuracy, the simplicity of its evaluation is an added attraction to assessing it. Conventionally, dentinal translucency has been examined using calipers to measure its length on unsectioned or sectioned teeth (4,5). In the last two decades, digital methods have also been proposed to quantify dentinal translucency (8–11) and have been reported as producing more accurate age estimates when compared to caliper-based measurement (11,12). Most of these, however, have focused on measuring the length of translucency, and a simple method to capture translucency area is still lacking. Lorentsen and Solheim (13) found that translucency area had a better correlation than length with chronological age, and also
contributed more often to stepwise multiple regression formulae than length measurements. Based on their results, they recommended measuring translucency area for age estimation. Translucency area has previously been measured manually (4,5), although this may not be an ideal approach—while manual measurements of translucency length are easily taken using calipers, the same is not true for area because it depends on an approximation of a sq. mm grid placed over the translucent zone (e.g., Ref. [5]). Hence, a method that records area more objectively is desirable and digital approaches may offer a solution. While few papers have proposed digital approaches to capture translucency area (9,10,14), they either lack a detailed description of the method (9,14) or require capturing tooth images on a video camera, converting the analog signal to digital format, and subsequent image processing using customized software program (10)—steps that, today, may be perceived as complex and outdated. With advances in computing technology, a simpler method is now feasible using commercially available computer hardware and software. Using these, this paper proposes a new technique to measure the area of root dentinal translucency and compares age estimates obtained from it with those of length measurements.
Methods 1 Department of Forensic Odontology, S.D.M. College of Dental Sciences and Hospital, Sattur, Dharwad, 580009 Karnataka, India. *Funded by an institutional grant from the S.D.M. College of Dental Sciences and Hospital, Karnataka, India. Received 6 Nov. 2012; and in revised form 15 Feb. 2013; accepted 23 Feb. 2013.
© 2014 American Academy of Forensic Sciences
Materials The material consisted of teeth collected from 133 subjects aged 19–85 years (mean age = 48.7 years) encompassing a heterogeneous sample of females and males of different age groups in relatively equal numbers (Table 1). Six subjects had multiple 1
2
JOURNAL OF FORENSIC SCIENCES
TABLE 1––Sample distribution across age groups, the sexes, and tooth classes. Sex Age group (years) 19–30 31–40 41–50 51–60 >60 Total
Tooth Type
Subjects
Sections
M
F
Incisors
Canines
Premolars
22 23 27 25 36 133
22 24 27 29 45 147
6 13 12 15 20 66
16 10 15 10 16 67
11 10 8 7 16 52
– 7 7 7 10 31
11 7 12 15 19 64
teeth (ranging from 2 to 7) contributing to a total of 147 teeth. The teeth were collected from the Department of Oral and Maxillofacial Surgery of two dental schools and two private clinics of this region, spread across a radius of c. 80 km. Only fully erupted permanent teeth extracted for valid clinical reasons such as periodontal disease, malocclusion/orthodontic treatment, and caries were included. Carious teeth were included only when the root, or root dentinal translucency, was unaffected macroscopically by the disease. Tooth Processing and Digitization The extracted teeth were thoroughly cleaned and soft tissue remnants removed from the root surface with a scalpel. Following preservation in 10% formalin, teeth were mounted in autopolymerizing acrylic for sectioning by a hard-tissue microtome (Leica SP 1600, Leica Microsystems GmßH, Wetzlar, Germany). The mounted teeth were sectioned longitudinally in the buccolingual plane to 250 lm, as close as was possible to the central axis of the tooth. This thickness has, repeatedly, been found to be the best for assessing translucency in terms of clarity (2,6). The sections were coded to ensure blind analysis. Tooth sections were digitized using a previously described method (11). Briefly, the sections were placed adjacent to an ABFO No. 2 scale (Lightning Powder Co. Inc., Jacksonville, FL) on a flatbed scanner (HP Scanjet G3010, Hewlett-Packard Co., Palo Alto, CA) and scanned at a resolution of 600 dpi.
Measurement of Translucency Area For measuring translucency area, the scanned images were opened in Adobe Photoshopâ 7.0.1 image-editing software program (Adobe Systems Inc., Mountain View, CA) installed in a Lenovo ThinkCentre desktop computer (Intelâ CoreTM i3-2100 Processor; 3.10-GHz CPU, 4.00 GB RAM) (Lenovo Group Ltd., Hong Kong, China). The root translucency area was selected using the following steps:
• •
•
•
•
The Zoom Tool in the Toolbox of Photoshopâ was selected to magnify the scanned image to a zoom setting of 66.7% (about 94 magnification). The Magic Wand Tool in the Toolbox was selected and the tolerance level set to 18; the Anti-aliased and Contiguous boxes were checked (both options visible on the menu bar). The Magic Wand Tool cursor was placed within the translucent zone visible in the section’s root and clicked. (Note: The tool selects areas of similar color or grayscale pixel values on the image [15]. Manually setting the tolerance level allows one to specify the pixel range. The tolerance for both grayscale and color has a range from 0 to 255 [15]. A lower tolerance level allows selecting zones with similar pixels values, while a higher level selects a broader range of color.) Ideally, the entire root translucency area should get selected with this action (indicated by the dynamic black and white lines, also referred to as the “marching ants”) (Fig. 1). However, it is likely that either more or less than the desired area of root dentin translucency gets selected because the translucency zones may be of slightly different pixel values or because zones of opaque root dentin separate them (Fig. 2). If more, the selected translucent zone is deselected (Ctrl + D, or Command + D for Macintosh computers), and the tolerance reduced from 18 to, say, 14. The Magic Wand Tool cursor is clicked again within the translucent zone. The action is repeated, with a still lower tolerance (12 or 8 usually suffices) if necessary, until the desired translucent area is selected. If less, the tolerance of 18 is left unchanged; the shift key on the keyboard is held down which allows adding to, or
FIG. 1––The entire area of root dentinal translucency has been selected, which is indicated by the black and white line, or “marching ants;” also, note that the crown and root dentin have been separated by a line using the Line Tool in the Toolbox.
ACHARYA
.
DIGITAL TRANSLUCENCY AREA MEASUREMENT
3
FIG. 2––Multiple selections of root dentin translucency is necessary when it is separated by regions of opaque dentin. Note the Rectangle Tool, beneath which the Line Tool is hidden.
•
•
extending, the selected translucent zone. A tiny “+” appears next to the Magic Wand Tool cursor (indicating the additive mode), and the selection of the desired translucency area is continued (Fig. 2). The additive/multiple selections may also be obtained with varying degrees of tolerance, as desired by the examiner. If translucency extends into the crown, the Line Tool in the Toolbox (Fig. 1), which is hidden below the Rectangle Tool (Fig. 2), may be used to separate the crown from the root by drawing a line at the cemento-enamel junction. This facilitates selecting only the root dentin translucency. However, because drawing the line creates a new layer in the image, go to Layer in the menu bar, click on Flatten Image and proceed from the first step. While selecting the translucent area, however, any visibly thin bands of peripheral translucency (c. 60 Total Subjects Males Females Total Sections Incisors Canines Premolars Total
Training Dataset (%)
Test Dataset (%)
19.0 20.0 18.0 18.0 25.0 100
9.1 9.1 27.3 21.2 33.3 100
49.0 51.0 100
51.5 48.5 100
37.0 21.0 42.0 100
31.9 21.3 46.8 100
based on the standard deviation. Instead, he encourages calculating mean differences, which is more appropriate in real-life situations, produces lower error rates, and is easier to understand (19). Gorard believes that this argument can be extended to SEE versus MAD (S. Gorard, personal communication). Hence, the MAD was used because it has the potential to represent the error more objectively; the SEE was also calculated separately for the test dataset to compare it with the MAD. In addition, the number (and percentage) of estimates with errors ≤ 10 years and ≥ 15 years was also calculated because Solheim and Sundnes (20) have categorized the former as “acceptable” and designated the latter as “unsatisfactory” in forensic age estimation (20; pg. 11). All arithmetic calculations were undertaken on an Excel spreadsheet (Office 2011; Microsoft Corp., Redmond, WA). Results The difference between the primary and repeat measurements of translucency area was statistically insignificant (t-value = 0.615; p > 0.05). The regression analysis revealed that maximum translucency length had the highest correlation with age, followed by average translucency length and translucency area (Table 3). The correlation coefficients for the quadratic functions were greater than those of their linear counterparts. The reference sample SEE was lowest for the quadratic function derived for maximum translucency length, and quadratic functions in general (Table 3). However, test of functions on the control dataset (n = 33) revealed that the quadratic function derived for translucency area produced the smallest SEE, MAD, and the highest number of “acceptable” age estimates (Table 4). The quadratic functions, overall, produced lower SEEs, MADs, and higher “acceptable” age estimates than their linear counterparts (Table 4), the exception being maximum translucency length wherein linear and quadratic functions produced the same level of “acceptable” age estimates. On the other hand, the linear functions produced fewer “unacceptable” age estimates than the quadratic functions, with the exception of translucency area, where it was the same for both (Table 4). Discussion The assessment of translucency is an easy, practical, and relatively accurate method of age estimation, which is usable by experts and novices alike. The 250-lm-thick tooth sections ensured the best possible visualization of translucency. While this study obtained sections using a hard-tissue microtome, similar sections may also be obtained through manual grinding and their thickness verified using a micrometer (e.g., Digimatic Micrometer, Mitutoyo Corp., Kawasaki, Japan). While manual grinding would not require mounting the section in autopolymerizing acrylic, the process of grinding per se would take a considerably longer time. Nevertheless, dentin translucency correlates most closely with chronological age when compared to other microscopically visible age changes on the dental tissues (2,9). Hence, large-scale interest in assessing this parameter resulted in several published papers that evaluated both length and area measurements (5,6,9,14,21). However, in contrast to Lorentsen and Solheim’s findings (13), these studies reported higher age correlations for length than for area, which is reflected in the results of this paper. The smaller diameter and less number per unit area (5) of dentinal tubules toward the root apex and root surface have been
ACHARYA
.
DIGITAL TRANSLUCENCY AREA MEASUREMENT
5
TABLE 3––Correlation coefficients (R), standard errors of the estimate (SEE), and regression formulae derived for translucency length and area measurements. Variables and Regression Function Avg. T length (ATL) linear Avg. T length (ATL) quadratic Maximum T length (MTL) linear Maximum T length (MTL) quadratic T area (TA) linear T area (TA) quadratic
R
SEE (Years)
0.55 0.60 0.60 0.62 0.52 0.60
13.3 12.8 12.8 12.5 13.6 12.7
Regression Formulae Age Age Age Age Age Age
= = = = = =
35.56 29.91 33.39 28.61 38.36 31.75
+ + + + + +
(3.48 (7.45 (2.81 (5.50 (1.06 (2.97
9 9 9 9 9 9
ATL) ATL) + ( 0.44 9 (ATL 9 ATL)) MTL) MTL) + ( 0.24 9 (MTL 9 MTL)) TA) TA) + ( 0.07 9 (TA 9TA))
T, translucency.
TABLE 4––Comparison of accuracy of the formulae derived for the measurements of translucency length and area. Function Avg. T linear Avg. T quadratic Maximum T linear Maximum T quadratic T area linear T area quadratic
≤ 5 Years (%) 11/33 9/33 14/33 12/33 15/33 14/33
(33.3) (27.3) (42.4) (36.4) (45.5) (42.4)
≤ 10 Years (%) 17/33 22/33 21/33 21/33 17/33 24/33
(51.5) (66.7) (63.6) (63.6) (51.5) (72.7)
≥ 15 Years (%) 6/33 7/33 6/33 8/33 6/33 6/33
(18.2) (21.2) (18.2) (24.2) (18.2) (18.2)
MAD (Years)
SEE (Years)
9.45 8.92 8.91 8.84 8.94 8.27
12.04 11.40 11.82 11.80 11.73 10.97
T, translucency.
considered as reasons for the commencement of translucency from the root tip and periphery. As an extension, it may be suggested that the narrower tubules closer to the root surface vis-a-vis those toward the root canal become translucent more consistently as age progresses, thereby contributing to a vertical linear (i.e., length) increase in root dentinal translucency that is more regular with an increase in age, than vertical and horizontal increase combined (i.e., area). According to Thomas et al. (14), because translucent dentin is not deposited uniformly, its three-dimensional volume may be more accurate in age estimation. However, they concede that the problem of computing three-dimensional volume may preclude its usage. In fact, Lorentsen and Solheim (13) have stated that a method previously developed for measuring translucency volume (viz., Ref. [22]) is a “difficult technique requiring expensive equipment and the time-consuming measurement of a sufficient number of teeth” (pg. 8). Therefore, they believe that measuring area of translucent dentin on tooth sections may be a simpler procedure for estimating age (13). The results herein show that it is repeatable too. Moreover, the higher accuracy of the quadratic formula derived for area measurements in the present study’s test dataset may be an added justification for its usage. The higher correlation and lower error rates of the quadratic functions derived for area and length measurements here confirm those obtained previously for length measurements (4,16). This suggests that the rate of increase in the dimensions of translucency slows down or plateaus in old age, probably because most of the root dentin has become translucent, impeding its further increase (4). In the present sample, this appears to occur just after 60 years of age, as seen by the curve of the quadratic function (Fig. 4). No paper that previously assessed and derived regression formulae for translucency length and area had tested the same on a control dataset. The intention of such an approach was to gauge the possible accuracy and applicability of these formulae in forensic casework, and not just stop at an inference based on correlation coefficients (R) and SEEs. The SEE, which is inversely related to R, has been used regularly in age estimation studies—including translucency assessment (8,12,23)—to depict the
FIG. 4––Scatter plot for the training dataset (n = 100) showing correlation (linear and quadratic regression lines) of translucency area with age. The quadratic regression line begins to curve just after the age of 60 years.
accuracy of methods and variables. However, Giles and Klepinger (18) have stated that it is merely “the square root of the average of the squared errors for the sample” (pg. 1218). Further, Gorard (19) points out that the relative efficiency of measures of dispersion such as the SEE depends “on there being no errors at all in the observations” (19; pg. 421). But for normal distributions with small “contaminations” in the data, the relative advantages which sample deviations (e.g., SEE) may have in uncontaminated situations over mean differences (e.g., MAD) are radically reversed. For example, observed differences tend to be longer-tailed with more extreme values than would be expected under ideal conditions. Unlike in MAD, the errors are squared in SEE, which makes it exponentially (rather than additively) greater, and the act of square-rooting its average “does not completely eliminate this bias” (19; pg. 421). Therefore, the SEE may give misleading answers in these cases, and measures of difference such as the MAD are more appropriate. Hence, the
6
JOURNAL OF FORENSIC SCIENCES
MAD is “more efficient in all life-like situations where small errors will occur in observation and measurement.” These criticisms fueled a need to ascertain error rates in test cases using the MAD, as well as its comparison with the SEE. The test, unsurprisingly, showed that the MAD was always lower than the SEE. It also showed that the formulae derived for area measurements produced the most accurate age estimates in terms of number of “acceptable” and “unacceptable” predictions, as well as MAD (Table 4). This occurred in spite of higher SEE visa-vis length measurements. Therefore, the method proposed here may have an advantage due to its smaller MADs, fewer “unsatisfactory” age estimates, and more “acceptable” age estimates; these suggest that digital translucency area measurements may be used as an alternative to digital translucency length measurements in forensic casework. In addition, the MAD of 8.27 years produced by the quadratic function for area measurements may also justify its categorization as a “moderately good” method, as defined by Schmeling et al. ([24]; pg. 179). Sengupta et al. (9) believe that there still is no absolute method for measuring dentinal translucency. They add that “in practice, each observer will use the measurement of the translucency that they find to be the best predictor of chronological age” (pg. 895). It is recognized that methods developed, in general, are dynamic in nature and their usage dictated by practicality, ease of use, relative accuracy, and specific demands of the case. The method described here is no different. In fact, the delineation of translucency area suggested may be undertaken through at least one other approach—as an alternative to using the Magic Wand Tool, the Magnetic Lasso Tool (which is hidden below the Lasso Tool in the Toolbox) may also be used. The use of the Magnetic Lasso Tool has been suggested in an age estimation paper that measured the decrease in size of the pulp chamber and root canal on digital radiographs (25). However, a potential drawback to its use in demarcating translucency is that it requires the operator to manually define the outline of the translucent area. This contrasts with the use of the Magic Wand Tool, which is semi-automatic in nature and selects the boundary of the translucent area based on the manually designated tolerance levels. This, perhaps, renders it less subjective than the Magnetic Lasso Tool. In conclusion, a method to digitally select and measure root dentin translucent area is described here, and was shown to produce more accurate age estimates in test cases when compared to digital length measurements. The readily and commercially available computer hardware and software for applying the method, its relative simplicity and its accuracy demonstrates the method’s potential for use in routine forensic dental age estimation. Acknowledgments The author expresses his gratitude to Professor Srinath L. Thakur, Principal of this institution, for his continued support to research in forensic odontology. Thanks are also owed to Professor C. Bahsker Rao, former Principal, for approving an institutional grant for the study and his support. The author also thanks Professor Sadashiva Shetty, Principal, Bapuji Dental College, Davangere, India, for providing subsidized access to the hard-tissue microtome at his institution and Professor Mandana Donoghue, Head of Oral Pathology at the College of Dental Sciences, Davangere, for making available sections from 14 subjects. Many thanks are owed to Profs. Anand Tavargeri and Kiran Radder for providing some of the extracted teeth from their private practices.
References 1. Gustafson G. Age determination on teeth. J Am Dent Assoc 1950;41: 45–54. 2. Johanson G. Age determination from teeth. Odontol Revy 1971;22(Suppl 21):1–126. 3. Miles AEW. Dentition in the estimation of age. J Dent Res 1963;42(Suppl 1):255–63. 4. Bang G, Ramm E. Determination of age in humans from root dentine transparency. Acta Odontol Scand 1970;28:3–35. 5. Solheim T. Dental root translucency as an indicator of age. Scand J Dent Res 1989;97:189–97. 6. Whittaker DK, Bakri MM. Racial variations in the extent of tooth root translucency in ageing individuals. Arch Oral Biol 1996;41(1):15–9. 7. Olze A, Geserick G, Schmeling A. Age estimation of unidentified corpses by measurement of root translucency. J Forensic Odontostomatol 2004;22(2):28–33. 8. Drusini A, Calliari I, Volpe A. Root dentine transparency: age determination of human teeth using computerized densitometric analysis. Am J Phys Anthropol 1991;85:25–30. 9. Sengupta A, Whittaker DK, Shellis RP. Difficulties in estimating age using root dentine translucency in human teeth of varying antiquity. Arch Oral Biol 1999;44(11):889–99. 10. Mandojana JM, Martin-de las Heras S, Valenzuela A, Valenzuela M, Luna JD. Differences in morphological age-related dental changes depending on postmortem interval. J Forensic Sci 2001;46:889–92. 11. Acharya AB. A new digital approach for measuring dentin translucency in forensic age estimation. Am J Forensic Med Pathol 2010; 31:133–7. 12. Valenzuela A, Martin-de las Heras S, Mandojana JM, Valenzuela M, Villanueva E. Multiple regression models for age estimation by assessment of morphologic dental changes according to teeth source. Am J Forensic Med Pathol 2002;23:386–9. 13. Lorentsen M, Solheim T. Age assessment based on translucent dentine. J Forensic Odontostomatol 1989;7:3–9. 14. Thomas GJ, Whittaker DK, Embery G. A comparative study of translucent apical dentine in vital and non-vital human teeth. Arch Oral Biol 1994;39(1):29–34. 15. Johansen RJ, Bowers CM. Digital analysis of bite mark evidence using Adobe Photoshop. Santa Barbara, CA: Forensic Imaging Services, 2000. 16. Acharya AB, Vimi S. Effectiveness of Bang and Ramm’s formulae in age assessment of Indians from dentin translucency length. Int J Legal Med 2009;123:483–8. 17. Spalding KL, Buchholz BA, Bergman LE, Druid H, Frisen J. Age written in teeth by nuclear tests. Nature 2005;437:333–4. 18. Giles E, Klepinger LL. Confidence intervals for estimates based on linear regression in forensic anthropology. J Forensic Sci 1988;33: 1218–22. 19. Gorard S. Revisiting a 90-year-old debate: the advantages of the mean deviation. Br J Educ Stud 2005;53(4):417–30. 20. Solheim T, Sundnes PK. Dental age estimation of Norwegian adults – a comparison of different methods. Forensic Sci Int 1980;16(1):7–17. 21. Singhal A, Ramesh V, Balamural PD. A comparative analysis of root dentin transparency with known age. J Forensic Dent Sci 2010;2: 18–21. 22. Sognnaes RF, Gratt BM, Papin PJ. Biomedical image processing for age measurements of intact teeth. J Forensic Sci 1985;30:1082–9. 23. Drusini AG. Age-related changes in root transparency of teeth in males and females. Am J Hum Biol 1991;3:629–37. 24. Schmeling A, Geserick G, Reisinger W, Olze A. Age estimation. Forensic Sci Int 2007;165:178–81. 25. Jeevan MB, Kale AD, Angadi PV, Hallikerimath S. Age estimation by pulp/tooth area ratio in canines: Cameriere’s method assessed in an Indian sample using radiovisiography. Forensic Sci Int 2011;204:209. e1–5. Additional Information and reprint requests: Ashith B. Acharya, B.D.S., G.D.F.O. Associate Professor and Head Department of Forensic Odontology S.D.M. College of Dental Sciences and Hospital Sattur Dharwad 580009 Karnataka India E-mail: [email protected]
TECHNICAL NOTE ODONTOLOGY Ashith B. Acharya,1 B.D.S., G.D.F.O.
Forensic Dental Age Estimation by Measuring Root Dentin Translucency Area Using a New Digital Technique*
ABSTRACT: Dentin translucency measurement is an easy yet relatively accurate approach to postmortem age estimation. Translucency area
represents a two-dimensional change and may reflect age variations better than length. Manually measuring area is challenging and this paper proposes a new digital method using commercially available computer hardware and software. Area and length were measured on 100 tooth sections (age range, 19–82 years) of 250 lm thickness. Regression analysis revealed lower standard error of estimate and higher correlation with age for length than for area (R = 0.62 vs. 0.60). However, test of regression formulae on a control sample (n = 33, 21–85 years) showed smaller mean absolute difference (8.3 vs. 8.8 years) and greater frequency of smaller errors (73% vs. 67% age estimates ≤ 10 years) for area than for length. These suggest that digital area measurements of root translucency may be used as an alternative to length in forensic age estimation.
KEYWORDS: forensic science, forensic odontology, age prediction, dentin transparency, area measurement, Adobe Photoshopâ, regression analysis
Postmortem age estimation is an important contributor to establishing identity of the deceased in forensic casework. Age estimation, in conjunction with sex assessment, stature prediction, and population designation, assists in reconstructive identification; alternatively, when law enforcers already have a putative age of the deceased, a forensic age estimate that is close to the presumed age provides the police clarity in their line of investigation. Since it was first proposed by Gustafson (1) as one of six parameters for adult age estimation, and backed by Johanson (2) as the most strongly correlated with age among the six, root dentinal translucency has gained immense interest and used as a single parameter in age estimation (3–7). Apart from its relative accuracy, the simplicity of its evaluation is an added attraction to assessing it. Conventionally, dentinal translucency has been examined using calipers to measure its length on unsectioned or sectioned teeth (4,5). In the last two decades, digital methods have also been proposed to quantify dentinal translucency (8–11) and have been reported as producing more accurate age estimates when compared to caliper-based measurement (11,12). Most of these, however, have focused on measuring the length of translucency, and a simple method to capture translucency area is still lacking. Lorentsen and Solheim (13) found that translucency area had a better correlation than length with chronological age, and also
contributed more often to stepwise multiple regression formulae than length measurements. Based on their results, they recommended measuring translucency area for age estimation. Translucency area has previously been measured manually (4,5), although this may not be an ideal approach—while manual measurements of translucency length are easily taken using calipers, the same is not true for area because it depends on an approximation of a sq. mm grid placed over the translucent zone (e.g., Ref. [5]). Hence, a method that records area more objectively is desirable and digital approaches may offer a solution. While few papers have proposed digital approaches to capture translucency area (9,10,14), they either lack a detailed description of the method (9,14) or require capturing tooth images on a video camera, converting the analog signal to digital format, and subsequent image processing using customized software program (10)—steps that, today, may be perceived as complex and outdated. With advances in computing technology, a simpler method is now feasible using commercially available computer hardware and software. Using these, this paper proposes a new technique to measure the area of root dentinal translucency and compares age estimates obtained from it with those of length measurements.
Methods 1 Department of Forensic Odontology, S.D.M. College of Dental Sciences and Hospital, Sattur, Dharwad, 580009 Karnataka, India. *Funded by an institutional grant from the S.D.M. College of Dental Sciences and Hospital, Karnataka, India. Received 6 Nov. 2012; and in revised form 15 Feb. 2013; accepted 23 Feb. 2013.
© 2014 American Academy of Forensic Sciences
Materials The material consisted of teeth collected from 133 subjects aged 19–85 years (mean age = 48.7 years) encompassing a heterogeneous sample of females and males of different age groups in relatively equal numbers (Table 1). Six subjects had multiple 1
2
JOURNAL OF FORENSIC SCIENCES
TABLE 1––Sample distribution across age groups, the sexes, and tooth classes. Sex Age group (years) 19–30 31–40 41–50 51–60 >60 Total
Tooth Type
Subjects
Sections
M
F
Incisors
Canines
Premolars
22 23 27 25 36 133
22 24 27 29 45 147
6 13 12 15 20 66
16 10 15 10 16 67
11 10 8 7 16 52
– 7 7 7 10 31
11 7 12 15 19 64
teeth (ranging from 2 to 7) contributing to a total of 147 teeth. The teeth were collected from the Department of Oral and Maxillofacial Surgery of two dental schools and two private clinics of this region, spread across a radius of c. 80 km. Only fully erupted permanent teeth extracted for valid clinical reasons such as periodontal disease, malocclusion/orthodontic treatment, and caries were included. Carious teeth were included only when the root, or root dentinal translucency, was unaffected macroscopically by the disease. Tooth Processing and Digitization The extracted teeth were thoroughly cleaned and soft tissue remnants removed from the root surface with a scalpel. Following preservation in 10% formalin, teeth were mounted in autopolymerizing acrylic for sectioning by a hard-tissue microtome (Leica SP 1600, Leica Microsystems GmßH, Wetzlar, Germany). The mounted teeth were sectioned longitudinally in the buccolingual plane to 250 lm, as close as was possible to the central axis of the tooth. This thickness has, repeatedly, been found to be the best for assessing translucency in terms of clarity (2,6). The sections were coded to ensure blind analysis. Tooth sections were digitized using a previously described method (11). Briefly, the sections were placed adjacent to an ABFO No. 2 scale (Lightning Powder Co. Inc., Jacksonville, FL) on a flatbed scanner (HP Scanjet G3010, Hewlett-Packard Co., Palo Alto, CA) and scanned at a resolution of 600 dpi.
Measurement of Translucency Area For measuring translucency area, the scanned images were opened in Adobe Photoshopâ 7.0.1 image-editing software program (Adobe Systems Inc., Mountain View, CA) installed in a Lenovo ThinkCentre desktop computer (Intelâ CoreTM i3-2100 Processor; 3.10-GHz CPU, 4.00 GB RAM) (Lenovo Group Ltd., Hong Kong, China). The root translucency area was selected using the following steps:
• •
•
•
•
The Zoom Tool in the Toolbox of Photoshopâ was selected to magnify the scanned image to a zoom setting of 66.7% (about 94 magnification). The Magic Wand Tool in the Toolbox was selected and the tolerance level set to 18; the Anti-aliased and Contiguous boxes were checked (both options visible on the menu bar). The Magic Wand Tool cursor was placed within the translucent zone visible in the section’s root and clicked. (Note: The tool selects areas of similar color or grayscale pixel values on the image [15]. Manually setting the tolerance level allows one to specify the pixel range. The tolerance for both grayscale and color has a range from 0 to 255 [15]. A lower tolerance level allows selecting zones with similar pixels values, while a higher level selects a broader range of color.) Ideally, the entire root translucency area should get selected with this action (indicated by the dynamic black and white lines, also referred to as the “marching ants”) (Fig. 1). However, it is likely that either more or less than the desired area of root dentin translucency gets selected because the translucency zones may be of slightly different pixel values or because zones of opaque root dentin separate them (Fig. 2). If more, the selected translucent zone is deselected (Ctrl + D, or Command + D for Macintosh computers), and the tolerance reduced from 18 to, say, 14. The Magic Wand Tool cursor is clicked again within the translucent zone. The action is repeated, with a still lower tolerance (12 or 8 usually suffices) if necessary, until the desired translucent area is selected. If less, the tolerance of 18 is left unchanged; the shift key on the keyboard is held down which allows adding to, or
FIG. 1––The entire area of root dentinal translucency has been selected, which is indicated by the black and white line, or “marching ants;” also, note that the crown and root dentin have been separated by a line using the Line Tool in the Toolbox.
ACHARYA
.
DIGITAL TRANSLUCENCY AREA MEASUREMENT
3
FIG. 2––Multiple selections of root dentin translucency is necessary when it is separated by regions of opaque dentin. Note the Rectangle Tool, beneath which the Line Tool is hidden.
•
•
extending, the selected translucent zone. A tiny “+” appears next to the Magic Wand Tool cursor (indicating the additive mode), and the selection of the desired translucency area is continued (Fig. 2). The additive/multiple selections may also be obtained with varying degrees of tolerance, as desired by the examiner. If translucency extends into the crown, the Line Tool in the Toolbox (Fig. 1), which is hidden below the Rectangle Tool (Fig. 2), may be used to separate the crown from the root by drawing a line at the cemento-enamel junction. This facilitates selecting only the root dentin translucency. However, because drawing the line creates a new layer in the image, go to Layer in the menu bar, click on Flatten Image and proceed from the first step. While selecting the translucent area, however, any visibly thin bands of peripheral translucency (c. 60 Total Subjects Males Females Total Sections Incisors Canines Premolars Total
Training Dataset (%)
Test Dataset (%)
19.0 20.0 18.0 18.0 25.0 100
9.1 9.1 27.3 21.2 33.3 100
49.0 51.0 100
51.5 48.5 100
37.0 21.0 42.0 100
31.9 21.3 46.8 100
based on the standard deviation. Instead, he encourages calculating mean differences, which is more appropriate in real-life situations, produces lower error rates, and is easier to understand (19). Gorard believes that this argument can be extended to SEE versus MAD (S. Gorard, personal communication). Hence, the MAD was used because it has the potential to represent the error more objectively; the SEE was also calculated separately for the test dataset to compare it with the MAD. In addition, the number (and percentage) of estimates with errors ≤ 10 years and ≥ 15 years was also calculated because Solheim and Sundnes (20) have categorized the former as “acceptable” and designated the latter as “unsatisfactory” in forensic age estimation (20; pg. 11). All arithmetic calculations were undertaken on an Excel spreadsheet (Office 2011; Microsoft Corp., Redmond, WA). Results The difference between the primary and repeat measurements of translucency area was statistically insignificant (t-value = 0.615; p > 0.05). The regression analysis revealed that maximum translucency length had the highest correlation with age, followed by average translucency length and translucency area (Table 3). The correlation coefficients for the quadratic functions were greater than those of their linear counterparts. The reference sample SEE was lowest for the quadratic function derived for maximum translucency length, and quadratic functions in general (Table 3). However, test of functions on the control dataset (n = 33) revealed that the quadratic function derived for translucency area produced the smallest SEE, MAD, and the highest number of “acceptable” age estimates (Table 4). The quadratic functions, overall, produced lower SEEs, MADs, and higher “acceptable” age estimates than their linear counterparts (Table 4), the exception being maximum translucency length wherein linear and quadratic functions produced the same level of “acceptable” age estimates. On the other hand, the linear functions produced fewer “unacceptable” age estimates than the quadratic functions, with the exception of translucency area, where it was the same for both (Table 4). Discussion The assessment of translucency is an easy, practical, and relatively accurate method of age estimation, which is usable by experts and novices alike. The 250-lm-thick tooth sections ensured the best possible visualization of translucency. While this study obtained sections using a hard-tissue microtome, similar sections may also be obtained through manual grinding and their thickness verified using a micrometer (e.g., Digimatic Micrometer, Mitutoyo Corp., Kawasaki, Japan). While manual grinding would not require mounting the section in autopolymerizing acrylic, the process of grinding per se would take a considerably longer time. Nevertheless, dentin translucency correlates most closely with chronological age when compared to other microscopically visible age changes on the dental tissues (2,9). Hence, large-scale interest in assessing this parameter resulted in several published papers that evaluated both length and area measurements (5,6,9,14,21). However, in contrast to Lorentsen and Solheim’s findings (13), these studies reported higher age correlations for length than for area, which is reflected in the results of this paper. The smaller diameter and less number per unit area (5) of dentinal tubules toward the root apex and root surface have been
ACHARYA
.
DIGITAL TRANSLUCENCY AREA MEASUREMENT
5
TABLE 3––Correlation coefficients (R), standard errors of the estimate (SEE), and regression formulae derived for translucency length and area measurements. Variables and Regression Function Avg. T length (ATL) linear Avg. T length (ATL) quadratic Maximum T length (MTL) linear Maximum T length (MTL) quadratic T area (TA) linear T area (TA) quadratic
R
SEE (Years)
0.55 0.60 0.60 0.62 0.52 0.60
13.3 12.8 12.8 12.5 13.6 12.7
Regression Formulae Age Age Age Age Age Age
= = = = = =
35.56 29.91 33.39 28.61 38.36 31.75
+ + + + + +
(3.48 (7.45 (2.81 (5.50 (1.06 (2.97
9 9 9 9 9 9
ATL) ATL) + ( 0.44 9 (ATL 9 ATL)) MTL) MTL) + ( 0.24 9 (MTL 9 MTL)) TA) TA) + ( 0.07 9 (TA 9TA))
T, translucency.
TABLE 4––Comparison of accuracy of the formulae derived for the measurements of translucency length and area. Function Avg. T linear Avg. T quadratic Maximum T linear Maximum T quadratic T area linear T area quadratic
≤ 5 Years (%) 11/33 9/33 14/33 12/33 15/33 14/33
(33.3) (27.3) (42.4) (36.4) (45.5) (42.4)
≤ 10 Years (%) 17/33 22/33 21/33 21/33 17/33 24/33
(51.5) (66.7) (63.6) (63.6) (51.5) (72.7)
≥ 15 Years (%) 6/33 7/33 6/33 8/33 6/33 6/33
(18.2) (21.2) (18.2) (24.2) (18.2) (18.2)
MAD (Years)
SEE (Years)
9.45 8.92 8.91 8.84 8.94 8.27
12.04 11.40 11.82 11.80 11.73 10.97
T, translucency.
considered as reasons for the commencement of translucency from the root tip and periphery. As an extension, it may be suggested that the narrower tubules closer to the root surface vis-a-vis those toward the root canal become translucent more consistently as age progresses, thereby contributing to a vertical linear (i.e., length) increase in root dentinal translucency that is more regular with an increase in age, than vertical and horizontal increase combined (i.e., area). According to Thomas et al. (14), because translucent dentin is not deposited uniformly, its three-dimensional volume may be more accurate in age estimation. However, they concede that the problem of computing three-dimensional volume may preclude its usage. In fact, Lorentsen and Solheim (13) have stated that a method previously developed for measuring translucency volume (viz., Ref. [22]) is a “difficult technique requiring expensive equipment and the time-consuming measurement of a sufficient number of teeth” (pg. 8). Therefore, they believe that measuring area of translucent dentin on tooth sections may be a simpler procedure for estimating age (13). The results herein show that it is repeatable too. Moreover, the higher accuracy of the quadratic formula derived for area measurements in the present study’s test dataset may be an added justification for its usage. The higher correlation and lower error rates of the quadratic functions derived for area and length measurements here confirm those obtained previously for length measurements (4,16). This suggests that the rate of increase in the dimensions of translucency slows down or plateaus in old age, probably because most of the root dentin has become translucent, impeding its further increase (4). In the present sample, this appears to occur just after 60 years of age, as seen by the curve of the quadratic function (Fig. 4). No paper that previously assessed and derived regression formulae for translucency length and area had tested the same on a control dataset. The intention of such an approach was to gauge the possible accuracy and applicability of these formulae in forensic casework, and not just stop at an inference based on correlation coefficients (R) and SEEs. The SEE, which is inversely related to R, has been used regularly in age estimation studies—including translucency assessment (8,12,23)—to depict the
FIG. 4––Scatter plot for the training dataset (n = 100) showing correlation (linear and quadratic regression lines) of translucency area with age. The quadratic regression line begins to curve just after the age of 60 years.
accuracy of methods and variables. However, Giles and Klepinger (18) have stated that it is merely “the square root of the average of the squared errors for the sample” (pg. 1218). Further, Gorard (19) points out that the relative efficiency of measures of dispersion such as the SEE depends “on there being no errors at all in the observations” (19; pg. 421). But for normal distributions with small “contaminations” in the data, the relative advantages which sample deviations (e.g., SEE) may have in uncontaminated situations over mean differences (e.g., MAD) are radically reversed. For example, observed differences tend to be longer-tailed with more extreme values than would be expected under ideal conditions. Unlike in MAD, the errors are squared in SEE, which makes it exponentially (rather than additively) greater, and the act of square-rooting its average “does not completely eliminate this bias” (19; pg. 421). Therefore, the SEE may give misleading answers in these cases, and measures of difference such as the MAD are more appropriate. Hence, the
6
JOURNAL OF FORENSIC SCIENCES
MAD is “more efficient in all life-like situations where small errors will occur in observation and measurement.” These criticisms fueled a need to ascertain error rates in test cases using the MAD, as well as its comparison with the SEE. The test, unsurprisingly, showed that the MAD was always lower than the SEE. It also showed that the formulae derived for area measurements produced the most accurate age estimates in terms of number of “acceptable” and “unacceptable” predictions, as well as MAD (Table 4). This occurred in spite of higher SEE visa-vis length measurements. Therefore, the method proposed here may have an advantage due to its smaller MADs, fewer “unsatisfactory” age estimates, and more “acceptable” age estimates; these suggest that digital translucency area measurements may be used as an alternative to digital translucency length measurements in forensic casework. In addition, the MAD of 8.27 years produced by the quadratic function for area measurements may also justify its categorization as a “moderately good” method, as defined by Schmeling et al. ([24]; pg. 179). Sengupta et al. (9) believe that there still is no absolute method for measuring dentinal translucency. They add that “in practice, each observer will use the measurement of the translucency that they find to be the best predictor of chronological age” (pg. 895). It is recognized that methods developed, in general, are dynamic in nature and their usage dictated by practicality, ease of use, relative accuracy, and specific demands of the case. The method described here is no different. In fact, the delineation of translucency area suggested may be undertaken through at least one other approach—as an alternative to using the Magic Wand Tool, the Magnetic Lasso Tool (which is hidden below the Lasso Tool in the Toolbox) may also be used. The use of the Magnetic Lasso Tool has been suggested in an age estimation paper that measured the decrease in size of the pulp chamber and root canal on digital radiographs (25). However, a potential drawback to its use in demarcating translucency is that it requires the operator to manually define the outline of the translucent area. This contrasts with the use of the Magic Wand Tool, which is semi-automatic in nature and selects the boundary of the translucent area based on the manually designated tolerance levels. This, perhaps, renders it less subjective than the Magnetic Lasso Tool. In conclusion, a method to digitally select and measure root dentin translucent area is described here, and was shown to produce more accurate age estimates in test cases when compared to digital length measurements. The readily and commercially available computer hardware and software for applying the method, its relative simplicity and its accuracy demonstrates the method’s potential for use in routine forensic dental age estimation. Acknowledgments The author expresses his gratitude to Professor Srinath L. Thakur, Principal of this institution, for his continued support to research in forensic odontology. Thanks are also owed to Professor C. Bahsker Rao, former Principal, for approving an institutional grant for the study and his support. The author also thanks Professor Sadashiva Shetty, Principal, Bapuji Dental College, Davangere, India, for providing subsidized access to the hard-tissue microtome at his institution and Professor Mandana Donoghue, Head of Oral Pathology at the College of Dental Sciences, Davangere, for making available sections from 14 subjects. Many thanks are owed to Profs. Anand Tavargeri and Kiran Radder for providing some of the extracted teeth from their private practices.
References 1. Gustafson G. Age determination on teeth. J Am Dent Assoc 1950;41: 45–54. 2. Johanson G. Age determination from teeth. Odontol Revy 1971;22(Suppl 21):1–126. 3. Miles AEW. Dentition in the estimation of age. J Dent Res 1963;42(Suppl 1):255–63. 4. Bang G, Ramm E. Determination of age in humans from root dentine transparency. Acta Odontol Scand 1970;28:3–35. 5. Solheim T. Dental root translucency as an indicator of age. Scand J Dent Res 1989;97:189–97. 6. Whittaker DK, Bakri MM. Racial variations in the extent of tooth root translucency in ageing individuals. Arch Oral Biol 1996;41(1):15–9. 7. Olze A, Geserick G, Schmeling A. Age estimation of unidentified corpses by measurement of root translucency. J Forensic Odontostomatol 2004;22(2):28–33. 8. Drusini A, Calliari I, Volpe A. Root dentine transparency: age determination of human teeth using computerized densitometric analysis. Am J Phys Anthropol 1991;85:25–30. 9. Sengupta A, Whittaker DK, Shellis RP. Difficulties in estimating age using root dentine translucency in human teeth of varying antiquity. Arch Oral Biol 1999;44(11):889–99. 10. Mandojana JM, Martin-de las Heras S, Valenzuela A, Valenzuela M, Luna JD. Differences in morphological age-related dental changes depending on postmortem interval. J Forensic Sci 2001;46:889–92. 11. Acharya AB. A new digital approach for measuring dentin translucency in forensic age estimation. Am J Forensic Med Pathol 2010; 31:133–7. 12. Valenzuela A, Martin-de las Heras S, Mandojana JM, Valenzuela M, Villanueva E. Multiple regression models for age estimation by assessment of morphologic dental changes according to teeth source. Am J Forensic Med Pathol 2002;23:386–9. 13. Lorentsen M, Solheim T. Age assessment based on translucent dentine. J Forensic Odontostomatol 1989;7:3–9. 14. Thomas GJ, Whittaker DK, Embery G. A comparative study of translucent apical dentine in vital and non-vital human teeth. Arch Oral Biol 1994;39(1):29–34. 15. Johansen RJ, Bowers CM. Digital analysis of bite mark evidence using Adobe Photoshop. Santa Barbara, CA: Forensic Imaging Services, 2000. 16. Acharya AB, Vimi S. Effectiveness of Bang and Ramm’s formulae in age assessment of Indians from dentin translucency length. Int J Legal Med 2009;123:483–8. 17. Spalding KL, Buchholz BA, Bergman LE, Druid H, Frisen J. Age written in teeth by nuclear tests. Nature 2005;437:333–4. 18. Giles E, Klepinger LL. Confidence intervals for estimates based on linear regression in forensic anthropology. J Forensic Sci 1988;33: 1218–22. 19. Gorard S. Revisiting a 90-year-old debate: the advantages of the mean deviation. Br J Educ Stud 2005;53(4):417–30. 20. Solheim T, Sundnes PK. Dental age estimation of Norwegian adults – a comparison of different methods. Forensic Sci Int 1980;16(1):7–17. 21. Singhal A, Ramesh V, Balamural PD. A comparative analysis of root dentin transparency with known age. J Forensic Dent Sci 2010;2: 18–21. 22. Sognnaes RF, Gratt BM, Papin PJ. Biomedical image processing for age measurements of intact teeth. J Forensic Sci 1985;30:1082–9. 23. Drusini AG. Age-related changes in root transparency of teeth in males and females. Am J Hum Biol 1991;3:629–37. 24. Schmeling A, Geserick G, Reisinger W, Olze A. Age estimation. Forensic Sci Int 2007;165:178–81. 25. Jeevan MB, Kale AD, Angadi PV, Hallikerimath S. Age estimation by pulp/tooth area ratio in canines: Cameriere’s method assessed in an Indian sample using radiovisiography. Forensic Sci Int 2011;204:209. e1–5. Additional Information and reprint requests: Ashith B. Acharya, B.D.S., G.D.F.O. Associate Professor and Head Department of Forensic Odontology S.D.M. College of Dental Sciences and Hospital Sattur Dharwad 580009 Karnataka India E-mail: [email protected]
