Novel approach for forensic dental identification using maxillary homologous models

Abstract

Purpose: As three-dimensional (3D) intraoral scans are becoming popular in modern dentistry, they also offer a new category of antemortem data for forensic identification. This study proposes a novel methodology for individual identification based on 3D homologous models of the maxilla.

Methods: Data from 100 maxillary plaster casts (40 monozygotic twin casts, 28 dizygotic twin casts and 32 singleton casts) were used. From the initial sample, 10 casts were randomly chosen and duplicated four times, resulting in 40 duplicates. These were divided into four groups and subjected to the following alterations: molar attrition (n = 10), canine attrition (n = 10), molar and canine attrition (n = 10), and no alteration (n = 10). All the casts were converted to 3D models and then to homologous models using a template with 24 landmarks. The 3D coordinates of each vertex in the homologous model were then calculated and used for statistical comparison of similarity between two given homologous models using the average Hausdorff distance.

Results: All four groups achieved accurate matching with their original maxillary casts, with a minimal average Hausdorff distance.

Conclusion: This method accurately identified individuals, including monozygotic twins, and exhibited robustness against minor tooth attrition, demonstrating its feasibility as an identification method in actual forensic settings.

Introduction

Individual identification in forensic settings is a major challenge, especially when the remains have suffered severe alterations, decomposition, or burns. Dental enamel is the strongest tissue in the human body and often persists even in severely decomposed bodies [1]. Therefore, comparison of antemortem and postmortem dental information, such as restorations, extractions, radiographs, and unique anatomical characteristics, has proven to be a highly reliable method for individual identification in both mass disaster scenarios and individual cases [2,3]. However, its effectiveness can be limited in situations such as postmortem tooth loss [4,5], severe damage to the teeth, or age-related wear and tear.

Previous studies have demonstrated the possibility of using palatal morphology for individual identification, regardless of its close resemblance in cases of monozygotic twins [6,7]. Palatal morphology is highly stable and resilient even in decomposed or charred bodies [8]. Furthermore, palatal dimensions exhibit sexual dimorphism, offering clues for secondary identification [9,10]. It has also been shown that palatal rugae patterns do not change during a lifetime, even in individuals who have undergone fixed orthodontic treatments [11]. Additionally, intercanine and intermolar arch widths do not change or may show only a minimal decrease after eruption of the permanent dentition [12]. This stability and structural complexity of palatal morphology can be leveraged to develop a potential forensic identification tool.

In modern dentistry, three-dimensional (3D) intraoral scans are being used more frequently, gradually replacing the conventional impression-taking process. As 3D scans are becoming more popular in dental practice, they also present a new and valuable category of antemortem data. Forrest (2019) has stated: “the future of comparison technologies in forensic odontology appears to be firmly rooted in the third dimension”. He has further suggested the possibility of automating the 3D model comparison process in the future for rapid and reliable matching of antemortem and postmortem models [3].

Performing accurate mathematical comparisons between 3D objects is challenging. Homologous modeling [13] is a technology that has been refined by the National Institute of Advanced Industrial Science and Technology, Japan, for mathematical analysis of 3D objects. This technology leverages a transforming 3D mesh, known as a template model, which can morph its shape to fit diverse 3D objects, generating individual homologous models. The template and the resulting individual models share the same vertex count. This shared structure enables mathematical comparisons between individual casts by analyzing their corresponding vertices. The homologous modeling technique has been used in a variety of body-shape analysis studies [14,15] and in a few dental studies in Japan [16,17,18,19].

Mathematical comparison of 3D surface data of the palate and maxillary teeth may be useful for personal identification. However, this has not been attempted in previous studies related to dental identification. The present study proposes a novel methodology for individual identification based on homologous models of the upper dental arch and palate. The primary focus was twofold: first, to assess the ability to differentiate between original maxillary casts and their exact replicas, particularly when dealing with identical twins, and second, to investigate how alteration (trimming) of the teeth on duplicate casts affects identification accuracy in comparison with unaltered casts. This novel forensic identification method aims to enhance and improve existing techniques, particularly in situations where traditional dental methods may prove inadequate.

Materials and Methods

The Graduate School of Dentistry at Tohoku University houses a large collection of dental plaster casts, originally obtained between 1960 and 2009 from Japanese junior high school students and undergraduate students at the Faculty of Dentistry, Tohoku University. Ethical approval for the study was obtained from Tohoku University Graduate School of Dentistry Research Ethics committee (approval number 2020-03-009). As the initial sample, 100 maxillary dental plaster casts were selected from the cast collection. The age of the individuals the casts were collected from ranged from 12 to 22 years. Selected casts included monozygotic twins (n = 40), dizygotic twins (n = 28), and singletons (n = 32). Selection criteria included the presence of all adult teeth from 16 to 26, absence of any deciduous teeth, absence of palatally or buccally erupted teeth, and minimal air bubbles on the plaster cast. The casts were labeled from C1 to C100.

From this initial sample of 100 casts, 10 were randomly selected and duplicated using alginate and hard plaster. This selected sample included five monozygotic twin casts, three dizygotic twin casts, and two singleton casts. Each cast was duplicated four times, resulting in 40 duplicates (D1-D40). The 10 originally selected casts and their labels after duplication are shown in Table 1.

The duplicated casts were divided into the following four groups:

1. 1.   No alteration to teeth (D1-D10).
2. 2.   Molar attrition (D11-D20): The occlusal surfaces of the right and left first molars were trimmed in a horizontal plane passing through the deepest point of the central fossa, effectively removing all fossae and fissures.
3. 3.   Canine attrition (D21-D30): The occlusal surfaces of both the left and right canines were trimmed in a horizontal plane passing through the widest two points mesiodistally.
4. 4.   Canine and molar attrition (D31-D40): A combination of trimming from groups 2 and 3, with both the first molars and canines receiving the aforementioned occlusal surface trimming.

Figure 1 illustrates the different levels of trimming performed on duplicated casts.

Each cast (initial 100 + 40 duplicated casts) was converted to a 3D digital model using a 3D scanner (Einscan SE-series; Shinning 3D, Zhejiang, PR China). Then, these 3D models were optimized using MeshLab open-source software [Cignoni P et al., Eurographics Italian Chapter Conference: 129-36, 2008] to improve computational efficiency during the subsequent steps.

Table 1 List of duplicated casts

Twin category	Original cast name	Label after duplication
		no alterations to teeth	molar attrition	canine attrition	canine and molar attrition
Singletons	C85	D1	D11	D21	D31
	C91	D2	D12	D22	D32
Monozygotic twins	C5	D3	D13	D23	D33
	C10	D4	D14	D24	D34
	C30	D5	D15	D25	D35
	C33	D6	D16	D26	D36
	C28	D7	D17	D27	D37
Dizygotic twins	C66	D8	D18	D28	D38
	C55	D9	D19	D29	D39
	C42	D10	D20	D30	D40

Fig. 1 Illustration of alterations made to duplicated casts

a: No alterations done to the teeth; b: Molar attrition group. The occlusal surfaces of the left and right first molars were trimmed in a horizontal plane passing through the deepest point of the central fossa; c: Canine attrition group. The occlusal surfaces of the left and right canines were trimmed in a horizontal plane passing through the widest two points mesiodistally; d: Molar and canine attrition group. This group received both molar and canine trimming as described in b. and c.

Homologous model construction

Homologous models were constructed using the HBM Rugle software (Medic Engineering Corp., Kyoto, Japan). The workflow involved the following steps:

1. 1.   Positioning: In order to ensure consistent alignment across all 3D models, each was positioned three times using visual judgment, anatomical points, and superimposition on an average model. For the last step, the iterative closest point algorithm [20] was used.
2. 2.   Isolating the palate and dentition in the 3D model: The aligned 3D models were trimmed using HBM Rugle and Autodesk Meshmixer (Autodesk Meshmixer, Autodesk, Inc., 2017) to exclude the buccal gingiva and areas beyond the first molar.
3. 3.   Landmark placement: As shown in Fig. 2, 24 landmarks were placed on each 3D model. The landmark placement was done by one researcher to minimize any inter-individual variations of placement.
4. 4.   Homologous model generation: Utilizing a template model, homologous models were generated for each cast. All the generated homologous models had the same number of vertices (19,129) as the template model. The homologous model generation process is illustrated in Fig. 3.

Fig. 2 Positions of 24 landmarks on the template model

The same landmarks were placed on each 3D cast.

Fig. 3 Illustration of the homologous model generation process

Landmark numbers 19-24 cannot be seen in this view as they were placed on the labial surface of the two central incisors.

Statistical analysis

The 3D (X, Y, and Z) coordinates of each vertex in the homologous model were calculated using the same software, HBM Rugle. These coordinates served as data points for the subsequent statistical analyses. With three data points (X, Y, and Z) per vertex, a homologous model containing 19,129 vertices yielded a total of 57,387 data points per model. Statistical analysis of shape dissimilarity was performed using the Average Hausdorff distance [21], [Andrea di F. ADM 2021, doi: 10.1007/978-3-030-91234-5_94], calculated from homologous model coordinates. The homologous models were divided into two groups: the original cast group (n = 100) and the duplicate cast group (n = 40). Each duplicate cast homologous model was then compared with each original cast homologous model, and the average Hausdorff distance was calculated. All the calculations were performed using R version 4.3.2 for aarch64-apple-darwin20. The R code for these calculations is provided in Fig. 4.

The average Hausdorff distance between two finite point sets X and Y is defined as

  

$$ d_{AHD} (X,Y) = \left(\frac{1}{X} \sum_{x \in X} \min_{y \in Y} d(x,y) + \frac{1}{Y} \sum_{y \in Y} \min_{x \in X} d(x,y) \right) \Big/ 2 $$

The directed average Hausdorff distance (DAHD) is calculated by finding the minimum distance for each point in X to its nearest neighbor in Y, summing these distances, and normalizing by the number of points in X. The average Hausdorff distance, a relative metric for evaluating the similarity of 3D virtual objects, is calculated by averaging the DAHD computed from both X to Y and Y to X [22]. A value of 0 indicates perfect similarity, while higher values signify greater dissimilarity. It is important to note that the average Hausdorff distance is a relative measure and cannot be expressed in specific units as an absolute measurement.

Fig. 4 R codes used to calculate average Hausdorff distance

Results

The average Hausdorff distances between each duplicated cast and the original cast are shown in Fig. 5. The minimum average Hausdorff distance was used as the criterion for a match. If the minimum average Hausdorff distance was achieved with its own original cast, it was defined as an accurate match.

All four groups, including those with altered casts, were accurately matched with their original maxillary casts, achieving the minimum Average Hausdorff distance in each case. Accurate matches were achieved using this criterion, even in the presence of monozygotic twin casts in the comparison group.

Fig. 5 Average Hausdorff distance between duplicated casts and original casts. The original cast that achieved the lowest Hausdorff distance (best match) is shown in blue. The twin siblings of monozygotic and dizygotic twins are indicated with star marks.

Duplicate casts without any alterations

All ten duplicate casts were correctly matched to their corresponding original casts, achieving the lowest average Hausdorff distance in each case. Notably, even when the duplicate cast belonged to a monozygotic twin, the lowest Hausdorff distance was consistently achieved with the individuals themselves. Furthermore, in four of the five monozygotic scenarios, the second-lowest Hausdorff distance was observed with its twin cast. However, it should be noted that this distinction is relatively weak. In the case of dizygotic twin casts in this group, the second-lowest Hausdorff distance was achieved with their twin in only one out of three cases.

Duplicates with attrition (molar attrition, canine attrition, molar and canine attrition)

Similarly to the unaltered duplicates, all altered duplicate casts accurately matched their original casts, obtaining the lowest Hausdorff distance in each case. Notably, all the cast groups subjected to attrition showed higher Hausdorff distances than the unaltered cast group. The molar attrition group and canine attrition group achieved similar Hausdorff distance ranges, whereas both the molar and canine attrition group exhibited the highest Hausdorff distances. Additionally, in all three attrition groups, four out of five monozygotic scenarios and one out of three dizygotic scenarios showed the second-lowest Hausdorff distance with their respective twin casts.

Discussion

This study utilized advancements in 3D scanning and data science to accurately match duplicated maxillary casts to their originals, in the presence of identical twins. Several critical components were implemented to achieve a more robust and data-driven identification method based on the morphology of the palate and maxillary teeth. First, the homologous modeling technique was leveraged, creating a novel approach for mathematical comparison of 3D casts. This technique has not been employed previously for forensic comparisons of the palate or dentition. Another key element of this approach was use of the Hausdorff distance as a quantitative measure of morphological differences. This objective approach allows more precise evaluation of shape similarity and dissimilarity. Furthermore, the incorporation of twin data provided valuable insights into the influence of genetics on palatal morphology. Ultimately, this facilitates a more accurate assessment of the discriminatory power of the proposed method for actual forensic scenarios.

The present results demonstrated that the unaltered cast group achieved the highest degree of similarity with their original maxillary casts. Monozygotic twin casts in the unaltered group showed the highest similarity with the original casts of themselves, while the second-highest similarity was observed with the original cast of their twin in four out of five cases. This highlights the ability of this method to distinguish between genetically identical individuals. In contrast, dizygotic twin comparisons showed second-highest similarity to their twin’s original cast in only one out of three cases. This suggests that the palatal and upper dental arch morphology of dizygotic twins exhibits greater variability compared to the monozygotic twins.

Furthermore, all three altered duplicate cast groups successfully matched their original counterparts; however, their similarity scores were lower than those of the unaltered group. Notably, the casts with only molar or canine trimming displayed a higher degree of similarity to their original casts than those with both canine and molar trimming. This can be attributed to loss of additional morphological information due to combined trimming in the latter group. These results demonstrate the robustness of this method against potential damage or modifications to the dentition.

In this study, canines and molars were selected for the attrition simulation because canines are exposed to attrition most frequently [23], and similarly first molars are the first permanent teeth to erupt and consequently have the highest dental morbidity [24,25]. The trimming method was chosen to mimic occlusal surface changes, as this was simple and easy to replicate.

However, there were some inherent limitations to the present study. In actual forensic scenarios, dentition with fractured or missing teeth, or restorations including crowns and bridges, is often encountered. Since it was unfeasible to investigate all such possibilities within the scope of the present study, specific selection criteria were established, including the presence of all teeth up to the first molars and absence of any restorations. Therefore, future studies should investigate the applicability of this method to dental casts exhibiting different dental variations encountered in actual forensic settings, including missing teeth, restorations, and other major alterations. Additionally, the present method relies heavily on the morphology of the palate and upper dental arch. Orthodontic treatment, which alters arch dimensions and tooth positions, could significantly impact identification accuracy. Therefore, future studies should investigate the applicability of this approach to individuals who have undergone orthodontic treatment.

As this was the first study of its kind to use maxillae, established protocols for landmark selection were not available. Accordingly, a preliminary approach to landmark selection was used, focusing on areas believed to be important for capturing maxillary shape and size. Previous studies have shown that the morphology of the labial surfaces of the maxillary anterior teeth has a greater impact on individual identification [26]. Although these chosen landmarks produced good results, the impact of reducing the number of landmarks on the central incisors remains unexplored. Further research is required to determine the optimal number and placement of landmarks, especially for central incisors, where reduction might be feasible without compromising the results. Future studies will hopefully benefit from this initial groundwork by developing more standardized approaches for landmark selection.

In conclusion, this study has demonstrated the potential of using 3D homologous models of the palate and upper dental arch for individual identification. This method successfully distinguished between individuals, including monozygotic twins, and exhibited robustness against minor tooth attrition.

Abbreviations

DAHD: directed average Hausdorff distance; 3D: three-dimensional

Ethical Statements

Ethical approval for this study was obtained from Tohoku University Graduate School of Dentistry Research Ethics Committee (approval number 2020-03-009).

Conflicts of Interest

The authors have no conflicts of interest to declare with respect to the authorship and/or publication of this article.

Funding

This work was supported by JST (Japan Science and Technology Agency) Spring, Grant Number JPMJSP2114.

Author Contributions

IK: conceptualization, methodology, investigation, data curation, writing – original draft, funding acquisition. UC: formal analysis, data curation, writing – review and editing. TS: resources, supervision, writing – review and editing. MK: resources, supervision, writing – review and editing. YH: conceptualization, supervision, writing – review and editing. All authors read and approved the final version of the manuscript.

ORCID iD

¹⁾IK: [e-mail], https://orcid.org/0000-0002-4455-3286

^2,3)UC: [e-mail], https://orcid.org/0000-0002-3272-4180

¹⁾TS: [e-mail], https://orcid.org/0000-0003-4669-2471

¹⁾MK: [e-mail], https://orcid.org/0000-0002-2671-3274

^1,4)YH*: [e-mail], https://orcid.org/0000-0002-8227-1191

Acknowledgments

The authors gratefully acknowledge the technical support provided by Mr. Toyohisa Tanijiri of Medic Engineering, Kyoto, Japan.

Data Availability Statements

Data generated during this study are available upon reasonable request to the corresponding author.

References

• 1.  Sweet D (2010) Forensic dental identification. Forensic Sci Int 201, 3-4. doi: 10.1016/j.forsciint.2010.02.030
• 2.  Arora KS, Kaur P (2016) Role of forensic odontology in the Indian Armed Forces: an unexplored arena. J Forensic Dent Sci 8, 173. doi: 10.4103/0975-1475.195124
• 3.  Forrest A (2019) Forensic odontology in DVI: current practice and recent advances. Forensic Sci Res 4, 316-330. doi: 10.1080/20961790.2019.1678710
• 4.  Oliveira RN, Melani RF, Antunes JL, Freitas ER, Galvão LC (2000) Postmortem tooth loss in human identification processes. J Forensic Odontostomatol 18, 32-36.
• 5.  Durić M, Rakocević Z, Tuller H (2004) Factors affecting postmortem tooth loss. J Forensic Sci 49, 1313-1318. doi: 10.1520/JFS2004113
• 6.  Taneva E, Evans C, Viana G (2017) 3D evaluation of palatal rugae in identical twins. Case Rep Dent 2017, 2648312. 1-6. doi: 10.1155/2017/2648312
• 7.  Simon B, Lipták L, Lipták K, Tárnoki ÁD, Tárnoki DL, Melicher D et al. (2020) Application of intraoral scanner to identify monozygotic twins. BMC Oral Health 20, 268. doi: 10.1186/s12903-020-01261-w
• 8.  Muthusubramanian M, Limson KS, Julian R (2005) Analysis of rugae in burn victims and cadavers to simulate rugae identification in cases of incineration and decomposition. J Forensic Odontostomatol 23, 26-29.
• 9.  Oliva B, Sferra S, Greco AL, Valente F, Grippaudo C (2018) Three-dimensional analysis of dental arch forms in Italian population. Prog Orthod 19, 34. doi: 10.1186/s40510-018-0233-1
• 10.  Mustafa AG, Tashtoush AA, Alshboul OA, Allouh MZ, Altarifi AA (2019) Morphometric study of the hard palate and its relevance to dental and forensic sciences. Int J Dent 2019, 1687345. doi: 10.1155/2019/1687345
• 11.  Bansode S, Kulkarni M (2009) Importance of palatal rugae in individual identification. J Forensic Dent Sci 1, 77-81. doi: 10.4103/0974-2948.60378
• 12.  Bishara SE, Jakobsen JR, Treder J, Nowak A (1997) Arch width changes from 6 weeks to 45 years of age. Am J Orthod Dentofacial Orthop 111, 401-409. doi: 10.1016/S0889-5406(97)80022-4
• 13.  Mochimaru M, Kouchi M (2000) Statistics for 3D human body forms. Proc Hum Factors Ergon Soc Annu Meet 44 (38), 852-855. doi: 10.1177/154193120004403846
• 14.  Imaizumi K, Bermejo E, Taniguchi K, Ogawa Y, Nagata T, Kaga K et al. (2020) Development of a sex estimation method for skulls using machine learning on three-dimensional shapes of skulls and skull parts. Foren Imag 22, 200393. doi: 10.1016/j.fri.2020.200393
• 15.  Matsumura H, Tanijiri T, Kouchi M, Hanihara T, Friess M, Moiseyev V et al. (2022) Global patterns of the cranial form of modern human populations described by analysis of a 3D surface homologous model. Sci Rep 12, 13826. doi: 10.1038/s41598-022-15883-3
• 16.  Suzuki K, Nakano H, Inoue K, Nakajima Y, Mizobuchi S, Omori M et al. (2020) Examination of new parameters for sex determination of mandible using Japanese computer tomography data. Dentomaxillofac Radiol 49, 20190282. doi: 10.1259/dmfr.20190282
• 17.  Oki M, Shiozawa M, Aoki S, Tsuchida Y, Suzuki T, Takahashi H (2023) Digitally creating three-dimensional standard shapes of complete dentures by using homologous models : a pilot study. J Digital Dent 1, 26-32. doi: 10.57516/jdigitaldent.1.1_26
• 18.  Oki M, Nakayama M, Tsuchida Y, Kamijo S, Aoki K (2024) Objective evaluation of tooth carving using homologous models. J Dent Educ 88, 314-321. doi: 10.1002/jdd.13418
• 19.  Miyazaki J, Kondo S, Tanijiri T, Negishi S (2024) Morphological differences between the first and second maxillary premolar crowns: a three-dimensional surface homologous modeling analysis. J Oral Biosci 66, 20-25. doi: 10.1016/j.job.2024.01.010
• 20.  Chen Y, Medioni G (1992) Object modelling by registration of multiple range images. Image Vis Comput 10, 145-155. doi: 10.1016/0262-8856(92)90066-C
• 21.  Mémoli F (2008) Gromov-Hausdorff distances in Euclidean spaces. 2008 IEEE Workshops, 1-8. doi: 10.1109/CVPRW.2008.4563074
• 22.  Schutze O, Esquivel X, Lara A, Coello CAC (2012) Using the averaged hausdorff distance as a performance measure in evolutionary multiobjective optimization. IEEE TEVC 16, 504-522. doi: 10.1109/TEVC.2011.2161872
• 23.  Liu B, Zhang M, Chen Y, Yao Y (2014) Tooth wear in aging people: an investigation of the prevalence and the influential factors of incisal/occlusal tooth wear in northwest China. BMC Oral Health 14, 65. doi :10.1186/1472-6831-14-65
• 24.  Dukić W, Delija B, Lulić Dukić O (2011) Caries prevalence among schoolchildren in Zagreb, Croatia. Croat Med J 52, 665-671. doi: 10.3325/cmj.2011.52.665
• 25.  Liu M, Xu X, Song Q, Zhang H, Zhang F, Lai G (2022) Caries prevalence of the first permanent molar and associated factors among second-grade students in Xiangyun of Yunnan, China: a cross-sectional study. Front Pediatr 10, 946176. doi: 10.3389/fped.2022.946176
• 26.  Mou QN, Ji LL, Liu Y, Zhou PR, Han MQ, Zhao JM et al. (2021) Three-dimensional superimposition of digital models for individual identification. Forensic Sci Int 318, 110597. doi: 10.1016/j.forsciint.2020.110597