Legal Medicine 48 (2021) 101826

Contents lists available at ScienceDirect

Legal Medicine
journal homepage: www.elsevier.com/locate/legalmed

Application and performance of artificial intelligence technology in

forensic odontology – A systematic review
Sanjeev B. Khanagar a, b, *, Satish Vishwanathaiah c, Sachin Naik d, Abdulaziz A. Al-Kheraif d,
Darshan Devang Divakar d, Sachin C. Sarode e, Shilpa Bhandi f, Shankargouda Patil g
a
Preventive Dental Science Department, College of Dentistry, King Saud Bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia
b
King Abdullah International Medical Research Center, Riyadh, Saudi Arabia
c
Department of Preventive Dental Sciences, Divison of Pedodontics, College of Dentistry, Jazan University, Jazan, Saudi Arabia
d
Dental Biomaterials Research Chair, Dental Health Department, College of Applied Medical Sciences, King Saud University, Riyadh 1143, Saudi Arabia
e
Department of Oral and Maxillofacial Pathology, Dr. D.Y.Patil Dental College and Hospital, Dr. D. Y. Patil Vidyapeeth, Pimpri, Pune 411018, Maharashtra, India
f
Department of Restorative Dental Sciences, Divison of Operative Dentistry, College of Dentistry, Jazan University, Saudi Arabia
g
Department of Maxillofacial Surgery and Diagnostic Sciences, Division of Oral Pathology, College of Dentistry, Jazan University, Jazan, Saudi Arabia

A R T I C L E I N F O A B S T R A C T

Keywords: Forensic odontology (FO) mainly deals with the identification of the individual through the remains, which
Artificial intelligence mainly includes teeth and jawbones. Artificial intelligence (AI) technology has proven to be a breakthrough in
Artificial neural networks providing reliable information in decision making in forensic sciences. This systematic review aimed to report on
Convolutional neural networks
the application and performance of AI technology in FO. The data was gathered through searching for the articles
Deep learning
Forensic odontology
in the renowned search engines, which have been published between January 2000 - June 2020. QUADAS-2 was
Forensic dentistry adopted for the risk of bias analysis of the included studies. AI technology has been widely applied in FO for
Machine learning identifying bite-marks, predicting mandibular morphology, gender determination, and age estimation. Most of
these AI models are based on either artificial neural networks (ANNs) or convolutional neural networks (CNNs).
The results of the studies are promising. Studies have reported that these models display accuracy and precision
equivalent to that of the trained examiners. These models can be promising tools when identifying victims of
mass disasters and as an additive aid in medico-legal situations.

1. Introduction determination is up to 90% by analysis of the skull bone [6]. In major

disaster cases, when the victim’s body is mutilated, which is beyond
Forensic odontology (FO) can be defined as a branch of dentistry that visual identification, the skull bones, jawbones, and teeth remains have
mainly deals with the identification of an individual by assessing the proven to be the most valuable source for identifying the individual. The
unique structures of the oral cavity [1]. The application of this science is shape of the mandible is mainly determined by the masticatory forces
mainly for medico-legal investigations, during mass disasters, and that usually vary from males and female [7]. Radiographic estimation of
identifying the accidental remains by analyzing dental records, and also the gender using the radiographs of the jawbones is considered more
for identifying an individual based on the human remains [2]. In FO, workable since they are simple and less destructive methods and can be
identification of the individual is possible through the remains, which applied on both dead and living cases [8].
mainly includes teeth and jawbones. Tooth eruption patterns can be a Various other methods are also employed in forensic odontology for
major source of evidence for estimating the chronological age of the identifying an individual. These include studying of the soft tissues of
victim. The state and wearing of the tooth structure can give an insight the diseased, which mainly include palatal rugae pattern (palatoscopy)
into the eating patterns or any habits related to the oral cavity [2–5]. and the pattern of the lip prints (cheiloscopy). Since they are considered
Size and shape of the jawbone of the victim can be major evidence for being unique in every individual and like fingerprints, they are perma­
estimating the gender of the victim. The accuracy of the gender nent and unchangeable, hence they can be a reliable in identifying an

* Corresponding author at: Preventive Dental Science Department, College of Dentistry, King Saud Bin Abdulaziz University for Health Sciences, King Abdullah
International Medical Research Center, Ministry of National Guard Heath Affairs, Riyadh, Saudi Arabia.
E-mail address: [email protected] (S.B. Khanagar).

https://doi.org/10.1016/j.legalmed.2020.101826
Received 29 August 2020; Received in revised form 19 November 2020; Accepted 4 December 2020
Available online 10 December 2020
1344-6223/© 2020 Elsevier B.V. All rights reserved.
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

individual [9]. Another added advantage is the cost-effectiveness and Table 1

ease in performing the examination on the leaving as well as the dead Description of the PICO (P = Population, I = Intervention, C = Comparison, O =
[10]. Lip prints can be obtained from the cigarettes, glasses and cups and Outcome) elements.
clothing. Lip prints can be used as evidence similar to that of a finger­ Research What are the AI applications and their performance in forensic
prints [11]. Similarly, palatal rugae also holds a unique feature of an question odontology?
individual and can definitely aid in identifying an individual. Palatal Population Patients diagnostic images related to oral and maxillofacial
rugae are protected from trauma and also form extreme temperatures, as regions [clinical images, radiographs, CBCT, confocal laser
they are shielded by the surrounding tissues like the lips, checks, tongue endomicroscopy (CLE) Images, intraoral fluorescence images,
cephalometric radiographs, near-infrared-light transillumination
and teeth [9,12]. Hence can be retained in the same positions through
(NILT) images]
the individuals’ life. They exhibit racial and gender variations and can Intervention AI based models for personal identification, age estimation, and
be useful in identifying an individual [13]. Intraoral photography along gender determination.
with oral impressions is one on the common method of inspection [14]. Comparison Expert opinions, reference standards
When it comes to hard tissues in the oral cavity dentine translucency Outcome Measurable or predictive outcomes such as accuracy, sensitivity,
specificity, ROC = receiver operating characteristic curve, AUC
seems to be one of the most reliable method in predicting the actual age = Area Under the Curve, ICC = Intraclass Correlation Coefficient,
of an individual. It had been considered as a best morpho-histological Positive/Negative Predictive Values (PPV/NPV)
parameter in estimation of the dental age, not only in terms of accu­
racy but also in terms of simplicity. Root dentine transparency develops
due to progressive sclerosing of the tubules, first at the root apex, and because of the duplication of the data. Thus 459 articles were recovered
then advancing coronally. It has been reported that dentin translucency for the second stage of data selection. The following inclusion and
increases as the age advances [15,16]. Cementum annulations may also exclusion criteria were applied to the remaining articles.
be used as a reliable method for morpho-histological traits of adult
skeleton for dental age estimation [17]. 2.3. Eligibility criteria for the studies
The technological advancements which mainly mimic the human
brains have revolutionised in dentistry. These smart mathematical 2.3.1. Inclusion criteria
technologies are named as artificial intelligent models [18–20]. This
artificial intelligence (AI) models can be trained and then applied for 1. The article must be focused on forensic odontology.
problem-solving and decision making. These models can be applied for 2. There should be an explicit mention of the AI technology used in the
various tasks in dental and medical diagnostics [21,22]. The major study model.
advantage of these models is that they provide reasoning for clinical 3. There should be an explicit mention of a measurable or predictive
decision making. The tremendous advancements in AI technology have outcome that can be quantified.
proven to be a breakthrough in providing reliable information in deci­ 4. There should be explicit mention regarding the data sets used for
sion making. Studies have reported on the application of AI technology either training/validating or evaluation of the AI model.
in predicting the age and gender based on the dental remains of the
individuals [18,23–25]. Hence this systematic review aimed to report on 2.3.2. Exclusion criteria
the application and performance of AI technology in FO.
1. Articles that are focused on areas other than AI technology.
2. Materials and methods 2. Articles with only abstracts and no full-text articles.
3. Articles in other than the English language.
2.1. Information sources and search strategy
2.4. Data extraction and management
The present systematic review was conducted according to guide­
lines of preferred reporting items for systematic reviews and meta-ana­ After applying these eligibility criteria’s, the number of articles was
lyses extension for diagnostic test accuracy (PRISMA-DTA) [26]. The further reduced to 9. The journal details and the author’s name were
data search for this systematic review was gathered through searching hidden and circulated among the authors to analyze the selected articles
for the articles reported in the literature in the renowned search engines critically. These articles were critically analyzed after considering the
mainly PubMed, Medline, Embase, Cochrane, Google Scholar, Scopus, guidelines of QUADAS-2 (Quality Assessment and Diagnostic Accuracy
Web of science, and Saudi digital library that have been published be­ Tool), a tool for quality assessment of the studies on diagnostic accuracy
tween (January 2000 - June 2020). The search strategy was mainly [27].
focused on articles that were based on AI applications in FO. A combi­ QUADAS-2 tool is designed to assess the quality of primary diag­
nation of keywords like artificial intelligence, deep learning, machine nostic accuracy studies; this tool does not intend to replace the data
learning, artificial neural networks, convolutional neural networks, extraction process of the review; hence it has to be applied besides
forensic odontology, and forensic dentistry was used for data searching. extraction of the primary data for the review. This tool comprises 4 key
The search strategy was mainly based on the PICO (problem/patient/ domains that mainly discuss patient selection, index test, reference
population, intervention/indicator, comparison, and outcome) elements standard, and flow of patients through the study and timing of the index
(Table 1). tests and reference standard (flow and timing). Each domain is assessed
in terms of the risk of bias, and the first 3 domains are also assessed in
2.2. Study identification and selection terms of concerns about applicability. Signalling questions are included
to help judge the risk of bias; these questions flag aspects of study design
Abstracts and full-text articles found in the preliminary search were related to the potential for bias and aim to help reviewers’ judge risk of
retrieved. Non-electronic source such as hand searching for the related bias [27].
journals and articles was also conducted along with electronic search­ The authors disagreed on 1 article for its inclusion in this systematic
ing. The data required for this systematic review was selected in two review, following this the article was excluded from the final analysis. So
stages. In the first stage, the articles were selected based on the title and finally, 8 articles were included for qualitative synthesis (Fig. 1).
after referring the abstracts related to our research question. The pre­ These articles were read in full and quantified respective of the year
liminary search mainly yielded 727 articles suitable for the aim of this of its publication so we could have developed the trends that had
systematic review. Two hundred sixty-eight articles were disregarded occurred in recent years in AI in FO.

2
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

Identification
Records identified through database Additional records identified through
searching other sources
(n = 718) (n = 9)

Records identified in total

(n = 727)
Screening

Records after duplicates removed

(n = 459)

Full-text articles assessed for Records excluded after

applying the inclusion and
Eligibility

eligibility
(n = 459) exclusion criteria
(n = 450)

Studies included in Full-text articles excluded,

qualitative synthesis (n = 1)
(n = 9) (Disagreement between
Included

the authors)

Studies included in
qualitative synthesis
(n = 8)

Fig. 1. Flow chart for screening and selection of articles.

3. Results 3.1. A qualitative synthesis of the included studies

A total of 8 research articles were analyzed for quantitative data. The The studies that were included in this review were mainly on the
analysis of the literature showed that the majority of the studies were application of AI in FO for identifying bite-marks [28] predicting the
conducted in the past 10 years. mandibular morphology [29] gender determination [2,30] and age
estimation [23,31,32] (Fig. 2).
Most of these AI-based studies have used artificial neural networks
(ANNs) [2,28,30] and convolutional neural networks (CNNs)

Age estimation

Gender determination

Predicting mandibular morphology

Identifying bitemarks

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

Predicting mandibular
Identifying bitemarks Gender determination Age estimation
morphology
Value 1 1 2 4

Fig. 2. Application of AI technology in forensic odontology.

3
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

[23,31,32]. A study reported of using Bayesian convolutional neural authors also state that the application of this study may not be used to
networks (BCNNs) [33]. These neural networks were designed to assess identify the crime at this stage.
panoramic radiographs (OPG), digital photographs, cephalometric ra­
diographs, lateral cephalometric radiographs (Table 2). 4.2. Mandibular morphology and artificial intelligence

3.2. Risk of bias assessment and applicability concerns Predicting the mandibular morphology is a very crucial aspect of
forensic sciences, especially when there is a need to reconstruct the face
The risk of bias assessment was carried out using the QUADAS-2 tool of the victim with a loss of mandibular bone. This becomes an important
for diagnostic tests. In the risk of bias arm of the tool, 75% of the studies aspect of the identification of an individual through their mortal re­
reported low risk for the patient section. One study each reported high mains, especially during mass disasters. Mandibular prediction can be a
risk [28] and unclear risk [30] (Supplementary Table 1). Since the data very important tool in improvising the facial reconstruction. Sandoval
feeding in AI technology is highly standardized, and there is no effect of et al. [29] reported using an AI-based model for predicting the
flow and time frame in the final output, both the aspects were regarded mandibular morphology through craniomaxillary variables on lateral
as low-risk categories in all the studies. The included studies were based radiographs in patients with skeletal class I, II and III, using automated
on the age estimation and sex determination, which does not require any learning techniques, such as Artificial Neural Networks and Support
gold standard preference test for determination. Hence, reference stan­ Vector Regression The results of the study were quite promising. The
dards in all the studies were reported as low risk, except for the study by ANN model demonstrated high predictability ability, and this model
Stern et al. [31] where unclear risk was identified. Following the risk of could play a key role in facial reconstruction. These results were like the
bias arm of a QUADAS-2 assessment tool, the applicability concern arm results of another study, which showed similar accuracy in predicting
also showed similar results (Supplementary Table 2) (Fig. 3a and b). the mandibular morphology [37,38].

4. Discussion 4.3. Gender determination and artificial intelligence

FO mainly deals with the identification of the victims of mass di­ Gender determination is of prime importance when identifying an
sasters, where a visual identification alone is reported to have high er­ individual during mass disasters, during accidental remains and medico-
rors [34]. Forensic odontologist deals with the dental remains of the legal cases. Skeletal bones are one of the strongest bones of the human
victims and matches them with the dental records of the missing person body, and they play a vital role in identifying the individual’s gender
[35] most times these are of very much importance in estimating the age [22,39,40].
of the victims [36]. Gender identification and age estimation are of Human teeth which are the part of the human skull can be measured
major concerns when recognizing a victim of such disasters, and also for in both living and also in the remains of the deceased individuals.
estimating the chronological age of an undocumented individual. In the permanent dentition, there is a difference in the genders,
AI is a technological advancement which has been widely used in mainly because of hormonal changes, which affect the size and shape of
forensic medicine. It has proven to be very efficient in estimating the teeth [41]. Canine teeth are known for demonstrating the highest degree
biological age and predicting the gender of the living and the diseased. of sexual dimorphism in humans. It is also a known fact that canine teeth
Despite limited number confirmative studies, it is has been stated as the are least affected teeth with dental caries and periodontal diseases
use of AI in forensic medicine is progressing beyond conventional henceforth are retained for longer when compared to any other teeth in
practice [18].This could be attributed to the automated technology with the oral cavity [42].
minimum errors in identifying the unknown patterns [19]. A study conducted by Fidya et al.[30]reported on a relatively new AI-
In this systematic review, we have analyzed the studies that have based technology for identifying sexual dimorphism in canines. In this
used AI-based models in forensic odontology. The chief advantage of study the authors quantified the respective accuracy of the Naive Bayes,
using the AI-based modes are, they are designed to identify an individual decision tree, and multi-layer perceptron methods in identifying sexual
through dental records, oral radiographs, and dental casts. These AI- dimorphism in canines the artificial intelligence methods produced a
based models are mainly designed and developed to overcome the highly accurate identification process relating to the gender determi­
variability in allocation caused by the human eye and mind [23]. nation of canine teeth. The most appropriate method was the multi-layer
perceptron with a higher accuracy rate.
4.1. Bite mark and artificial intelligence Another study conducted by Patil et al. [2] reported on the appli­
cation of an AI-based model for gender determination. This model was
Bite marks are considered as crucial evidence during sexual assaults/ highly accurate in identifying the gender of an individual using pano­
attacks, sexual homicide, rape, and sexual abuse and also in child abuse. ramic radiographs. This model was compared with two standard gender
Analyzing human bite marks is very much challenging for forensic determination methods discriminant analysis and logistic regression
odontology. A study reported by Mahasantipiya et al. [28] reported on which have often showed excellent results in gender determination.
using artificial neural networks for bite mark identification. This was These two techniques were analyzed along with artificial neural net­
mainly designed to develop a model with high accuracy in performance works. But the outcome of the study results substantiated that the AI
and to overcome human bias during the process of human identification. model was better than the other two models.
This model was trained by selecting some specific features of the bite
marks, following which the trained networks gave a reasonable result 4.4. Age estimation and artificial intelligence
for the matching accuracy in this initial study. Even though the neural
network used for the bite mark identification did not show a high level of Age estimation is mainly required for estimating the chronological
accuracy, it shows that this approach has potential and needs further age of an individual, for medico-legal issues, especially when there is a
study to improve the performance. Application and training with new legal confirmation required [43]. The age estimation of an individual is
features might be needed to improve the performance of this model. The mainly carried out by experts using hand-wrist radiographs and with the

4
 S.B. Khanagar et al.
Table 2
Details of the studies that have reported on AI based models used in forensic odontology for personal identification, age estimation, and gender determination.
Serial Authors Year of Algorithm Objective of the No. of No. of Study factor Modality Evaluation Comparison if Results (+) Outcomes Authors suggestions/
no publication Architecture study images/ images/ accuracy /average any effective, recommendations
photographs photographs accuracy (-) non
for training / for testing effective
validating (N) neutral

1 Mahasantipiya 2011 ANNs Applying the 250 100 Bite marks Digital photographs Average accuracy Not (+) The experimental More studies are
et al28 neural network 82% Mentioned Effective results show needed to improve
approach in bite reasonable level the bite mark
mark of accuracy. identification
identification. accuracy
2 Fidya et al30 2017 ANNs Naive Bayes, 100 50 Upper and Teeth measurements The accuracy of Naive Bayes, (+) The MLP method The use of AI methods
decision tree, and lower MLP method was Decision tree Effective had a highest produced a highly
multi-layer canine 84%. accuracy and an accurate
perceptron (MLP) absolute error identification process
methods in value lower than relating to the gender
identifying sexual that of its determination of
dimorphism in decision tree canine teeth
canines counterpart.
3 Tobel et al23 2017 CNNs An automated 160 40 Tooth Panoramic Mean ICC was 2 observers (+) Deep CNN based Further optimization
technique for radiographs (OPG) 0.95 Effective AI system is required to achieve
staging the demonstrated a fully automated
development of similar results to system for estimating
lower third molar. the results the dental age.
demonstrated by
other trained
examiners.
4 Sandoval et al29 2017 ANNs AI based model for Not clear 229 Anatomical Lateral Coefficients from Support vector (+) This model This model may be
5

predicting the landmarks cephalograms 0.84 until 0.99 regression Effective demonstrated the key for facial
mandibular high reconstruction
morphology predictability
ability
5 Stern et al31 2017 CNNs Automatic multi- Not clear Not clear Skeletal Skeletal and Dental 1.14 \ pm 0.96 Radiologist (+) Improvement in None
factorial age structures MRI years of mean Effective sensitivity of the
estimation absolute error multi-factorial
methods based on system
MRI data
6 Back et al33 2019 BCNNs Bayesian 2400 Not clear Upper and Orthopantomograms Concordance Not (+) This model Initial results are
convolutional lower jaws Correlation Mentioned Effective provides encouraging although
neural networks to Coefficient ccc = quantitative the accuracy is not yet
perform age and 0:91 estimation of at the level that
uncertainty prediction warrants routine
estimation uncertainty, application.
which is
imperative within
a legal context.
7 Patil et al2 2020 ANNs ANN for gender 444 95 Mandible Panoramic Accuracy of 75% 1 experienced (+) ANN proved as a This automated

Legal Medicine 48 (2021) 101826

determination radiographs (OPG) oral and Effective good tool for application is
maxillofacial predicting the promising for
radiologist gender and can identifying gender or
be applied in the age with minimal
forensic sciences errors
for near accurate
results.
8 Blanco et al32 2020 CNNs Two fully 2289 537 Oral Panoramic Correlation DANet and (+) DAS Net DASNet can be used
automatic methods Structures radiographs (OPG) between the ages manual age Effective outperforms the to automatically
(continued on next page)
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

help of panoramic X-rays of jawbones [44]. The accuracy and expert

ANNs = Artificial Neural Networks, CNNs = Convolutional Neural Networks, BCNNs = Bayesian Convolutional Neural Networks, ROC = Receiver Operating Characteristic curve, AUC = Area Under the Curve, ICC =
chronological age of a

developing dentitions
Authors suggestions/
opinions need to be very precise in this task. To have a more precise and

especially in young
subject accurately,
recommendations
standard tool for estimating the chronological age of an individual,
several AI-based automated systems have been developed to overcome

subjects with
predict the
the observer subjectivity [45,46].
A study conducted by Tobel et al. [23] reported of applying an AI-
based model for staging the development of lower third molar on
panoramic radiographs. The performance was evaluated in a 5-fold

DASNet in every
cross-validation scenario, using different validation metrics like accu­
racy, Rank-N recognition rate, mean absolute difference, linear kappa
Results (+) Outcomes

coefficient. Deep Learning Convolutional Neural Network approach

aspect
outperformed all other tested approaches. The model was efficient, and
when compared with the trained examiners, it displayed equivalent
(N) neutral

accuracy.
effective,

effective
(-) non

These results were similar to the study conducted by Stern et al. [31]
where the authors reported of using automatic multi-factorial age esti­
mation methods based on MRI data to extend the maximal age range
Comparison if

from 19 years, as commonly used for age assessment based on hand

estimation
methods

bones, up to 25 years, when combined with wisdom teeth and clavicles.

The performance of this model was similar to radiologists.
any

Blanco et al. [32] in his study reported using two fully automatic
accuracy /average

methods to estimate the chronological age of a subject using the OPG

estimated by the

AUC above 0.92

85.4% with an
0.87 and 0.90,
DASNet (R2 =

respectively).

images. The first (DANet) consists of a sequential Convolutional Neural

DASNet was
Accuracy of
DANet and
Evaluation

Network (CNN) path to predict the age, while the second (DASNet) adds
accuracy

a second CNN path to predict the sex and uses sex specific features with
the aim of improving the age prediction performance. The results
showed that the DASNet outperformed the DANet in every aspect.
DASNet can automatically predict the chronological age of a subject
accurately, especially in young subjects with developing dentitions.
These results were like the results of the study conducted by Back et al.
Modality

[33] where the authors reported of using Bayesian convolutional neural

networks to perform age and uncertainty estimation using a large data
set of 12; 000 panoramic radiographs of the upper and lower jaws. Initial
Study factor

results are quite encouraging, although the accuracy is not yet at the
level that warrants routine application.

4.5. Future directions

photographs

Intraclass Correlation Coefficient, F = F- measure, CCC = Concordance Correlation Coefficient.

for testing
images/

It is clear from the present systematic review that AI could play an

No. of

important role in various aspects of FO. However, all the studies pub­
lished to date, and included in the present analysis were conducted on
for training /

experimental models, and hence, it cannot give the actual measure of the
photographs

validating

success in real-life scenarios. Future studies are recommended on the use

images/
No. of

of AI modalities in actual real-life incidences like criminal cases, mass

disasters, etc. This enables us the real translational value of AI in FO. As
DANet (Dental Age

per our analysis, it has come to our notice that only certain subdivisions
chronological age
of a subject from
(Dental Age and
Objective of the

Net) & DASNet

of FO such as bitemarks [28] mandibular morphology [29] gender

the OPG image
estimate the
Sex Net), to

determination [2,30] and age estimation [23,31,32] have been used for
testing AI. Although it covers almost every aspect of FO, other aspects
study

such as lip prints, palatal rugae, cemental lines, dentin translucency,

should also be investigated with AI technology. Studies are also rec­
publication Architecture

ommended using AI for racial identification by forensic odontology

Algorithm

parameters.

5. Conclusions
Year of

AI technology has been widely used in FO, and the results are
promising. Studies have reported that these models display accuracy and
precision equivalent to that of trained examiners. These models have an
added advantage of overcoming human errors and being non-invasive.
Table 2 (continued )

Decision making is a very crucial aspect in forensic sciences, and these

Authors

automated models have shown excellent results along with the elimi­
nation of the human bias. These models can be promising tools when
identifying victims of mass disasters and as an additive aid in medico-
Serial

legal situations. Lack of real-life scenario and experimental nature of

no

these included studies is a major limitation of the present review. Future

6
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

QUADAS-2 Domain

Fig.3b.Proportion and number of studies with low, high, or unclea

Fig.3a. Proportion and number of studies with low, high or unclear
CONCERNS regarding APPLICABILITY
RISK of BIAS

Foot note: In Fig. 3b there is no "flow and timing" variable on Y-axis

Fig. 3. Assessment of individual risk of bias domains and concerns regarding applicability. Foot note: In Fig. 3b there is no “flow and timing” variable on Y-axis.

studies are recommended on the use of AI in other FO parameters such as [10] A. Indira, M. Gupta, M. David, Palatal rugae patterns for establishing individuality,
J. Forensic Dent. Sci. 4 (2012) 2, https://doi.org/10.4103/0975-1475.99149.
lip prints, palatal rugae, cemental lines, dentin translucency.
[11] G. Mishra, K. Ranganathan, T.R. Saraswathi, Study of lip prints, J. For. Dent. Sci. 1
(1) (2009) 28, https://doi.org/10.4103/0974-2948.50885.
[12] A. Saraf, S. Bedia, A. Indurkar, S. Degwekar, R. Bhowate, Rugae patterns as an
Declaration of Competing Interest adjunct to sex differentiation in forensic identification, J. Forensic Odontostomatol.
29 (2011) 14–19.
The authors declare that they have no known competing financial [13] GovindRaj Kumar, R. Dhanapal, T.R. Saraswathi, SreenivasaT Bharath, Sex
determination by discriminant function analysis of palatal rugae from a population
interests or personal relationships that could have appeared to influence of coastal Andhra, J. Forensic Dent. Sci. 3 (2) (2011) 58, https://doi.org/10.4103/
the work reported in this paper. 0975-1475.92144.
[14] ShriramC Bansode, MeenaM Kulkarni, Importance of palatal rugae in individual
identification, J. Forensic Dent Sci. 1 (2) (2009) 77, https://doi.org/10.4103/
Acknowledgements 0974-2948.60378.
[15] S. Gupta, A. Chandra, A. Agnihotri, O.P. Gupta, N. Maurya, Age estimation by
The authors extend their appreciation to the Deputyship for Research dentin translucency measurement using digital method: an institutional study,
J. Forensic Dent. Sci. 9 (2017) 42–48, https://doi.org/10.4103/jfo.jfds_76_14.
& Innovation, “Ministry of Education” in Saudi Arabia for funding this [16] JignaS Shah, AshishF Ranghani, PiyushG Limdiwala, Age estimation by assessment
research work through the project number IFKSURP-2020-106. of dentin translucency in permanent teeth, Indian J. Dent. Res. 31 (1) (2020) 31,
https://doi.org/10.4103/ijdr.IJDR_428_18.
[17] T.J. Stein, J.F. Corcoran, Pararadicular cementum deposition as a criterion for age
Appendix A. Supplementary data estimation in human beings, Oral Surg. Oral Med. Oral Pathol. 77 (3) (1994)
266–270, https://doi.org/10.1016/0030-4220(94)90297-6.
Supplementary data to this article can be found online at https://doi. [18] N.M. Kozan, Y.Z. Kotsyubynska, G.M. Zelenchuk, Using the artificial neural
networks for identification unknown person, IOSR J. Dent. Med. Sci. 16 (2017)
org/10.1016/j.legalmed.2020.101826. 107–113, https://doi.org/10.9790/0853-160403107113.
[19] A. Sikka, A. Jain, Sex determination of mandible: a morphological and
References morphometric analysis, Int. J. Contemp. Med. Res. 3 (2016) 1869–1872.
[20] T. Katne, A. Kanaparthi, S. Gotoor, S. Muppirala, R. Devaraju, R. Gantala, Artificial
intelligence: demystifying dentistry – The future and beyondartificial intelligence:
[1] A. Johnson, B. Gandhi, S.E. Joseph, A morphological study of tongue and its role in
demystifying dentistry – The future and beyond, IJCMSR 4 (4) (2019), https://doi.
forensics odontology, J. Forensic Sci. Crim. Inves. 7 (2018) 1–5, https://doi.org/
org/10.21276/ijcmsr.2019.4.4.2.
10.19080/JFSCI.2018.07.555723.
[21] M.R. Brickley, J.P. Shepherd, R.A. Armstrong, Neural networks: a new technique
[2] V. Patil, R. Vineetha, S. Vatsa, D.K. Shetty, A. Raju, N. Naik, N. Malarout, Artificial
for development of decision support systems in dentistry, J. Dentistry 26 (4) (1998)
neural network for gender determination using mandibular morphometric
305–309, https://doi.org/10.1016/S0300-5712(97)00027-4.
parameters: A comparative retrospective study, Cogent Eng. 7 (2020) 1–12.
[22] W.J. Park, J.-B. Park, History and application of artificial neural networks in
https://doi.org/10.1080/23311916.2020.1723783.
dentistry, Eur. J. Dent. 12 (04) (2018) 594–601, https://doi.org/10.4103/ejd.ejd_
[3] Forensic odontology – the forensics library, https://aboutforensics.co.uk/forensic
325_18.
odontology/ (accessed October 10, 2020).
[23] J. De Tobel, P. Radesh, D. Vandermeulen, P.W. Thevissen, An automated technique
[4] S. Sakoda, B.-L. Zhu, K. Ishida, S. Oritani, M.Q. Fujita, H. Maeda, Dental
to stage lower third molar development on panoramic radiographs for age
identification in routine forensic casework: clinical and postmortem investigations,
estimation: a pilot study, J. Forensic Odontostomatol. 2 (2017) 49–60.
Leg. Med. 2 (1) (2000) 7–14, https://doi.org/10.1016/S1344-6223(00)80002-4.
[24] P. Balgi, F. Katge, S. Singh, J. Jayaraman, Comparative assessment to determine
[5] M. Ohtani, T. Chiba, N. Yoshioka, Survey of dental diseases in forensic autopsy
the accuracy of Demirjian and DARLInG methods of dental age estimation in
cases, Leg. Med. 11 (2009) S341–S343, https://doi.org/10.1016/j.
western Indian children, Leg. Med. 42 (2020) 101629, https://doi.org/10.1016/j.
legalmed.2009.02.063.
legalmed.2019.101629.
[6] C.B. More, R. Vijayvargiya, N. Saha, Morphometric analysis of mandibular ramus
[25] B.Y. Roh, W.J. Lee, J.U. Seo, U.Y. Lee, S.S. Lee, Analysis of forensic odontological
for sex determination on digital orthopantomogram, J. Forensic Dent. Sci. 9 (2017)
examinations at the national forensic service of korea from 2011 to 2015, Leg.
1–5, https://doi.org/10.4103/jfo.jfds_25_15.
Med. 32 (2018) 37–42. https://doi.org/10.1016/j.legalmed.2018.02.007.
[7] J.D. Raj, S. Ramesh, Sexual dimorphism in mandibular ramus of south indian
[26] T.A. McGrath, M. Alabousi, B. Skidmore, D.A. Korevaar, P.M.M. Bossuyt, D. Moher,
population, Antrocom Online J. Anthropol. 9 (2013) 253–258.
B. Thombs, M.D.F. McInnes, Recommendations for reporting of systematic reviews
[8] V. Patil, K.M. Pai, N. Naik, A comparative study on determination of pulp/tooth
and meta-analyses of diagnostic test accuracy: A systematic review, Syst. Rev. 6
area ratio using mimics and adobe photoshop for estimation of age by cameriere’s
(2017) 194. https://doi.org/10.1186/s13643-017-0590-8.
method, World J Dent. 9 (2018) 418–422. https://doi.org/https://www.wjoud.
[27] P.F. Whiting, A.W.S. Rutjes, M.E. Westwood, S. Mallett, J.J. Deeks, J.B. Reitsma, M.
com/doi/WJOUD/pdf/10.5005/jp-journals-10015-1573.
M.G. Leeflang, J.A.C. Sterne, P.M.M. Bossuyt, Quadas-2: a revised tool for the
[9] I.M. Caldas, T. Magalhães, A. Afonso, Establishing identity using cheiloscopy and
quality assessment of diagnostic accuracy studies, Ann. Intern. Med. 155 (2011)
palatoscopy, Forensic Sci. Int. 165 (1) (2007) 1–9, https://doi.org/10.1016/j.
529–536, https://doi.org/10.7326/0003-4819-155-8-201110180-00009.
forsciint.2006.04.010.

7
S.B. Khanagar et al. Legal Medicine 48 (2021) 101826

[28] J.S. P.M. Mahasantipiya, U. Yeesarapat, T. Suriyadet, T.Thaiupathump, Bite mark [38] Sudheer B. Balla, Sudhashree Lingam, Alekhya Kotra, Hima Ranjana P,
identification using neural networks: A preliminary study, in: Proc. Int. Karunakar P, Naga Madhuri N, Roberto Cameriere, New regression models for
MultiConference Eng. Comput. Sci., Hong Kong, 2011. http://www.iaeng.org/ dental age estimation in children using third molar maturity index: A preliminary
publication/IMECS2011/IMECS2011_pp65-68.pdf (accessed July 7, 2020). analysis testing its usefulness as reliable age marker, Legal Med. 39 (2019) 35–40,
[29] Tania Camila Niño-Sandoval, Sonia V. Guevara Pérez, Fabio A. González, Robinson https://doi.org/10.1016/j.legalmed.2019.06.003.
Andrés Jaque, Clementina Infante-Contreras, Use of automated learning techniques [39] B.N. Kumar, R. Taanvi, S. Reddy, S.M. Reddy, S.K. Chennoju, K. Pavani, K.N. Sai
for predicting mandibular morphology in skeletal class I, II and III, Forensic Sci. Praveen, Morphometric analysis of mandible for sex determination-A retrospective
Int. 281 (2017) 187.e1–187.e7, https://doi.org/10.1016/j.forsciint.2017.10.004. study, IJSRM. 4 (2016) 115–122. https://doi.org/https://europepmc.org/article/
[30] F. Fidya, B. Priyambadha, Automation of gender determination in human canines med/28584466.
using artificial intelligence, Dent. J. (Majalah Kedokt. Gigi). 50 (2018) 116–120. [40] T. Dosi, S. Vahanwala, D. Gupta, Assessment of the effect of dimensions of the
https://doi.org/10.20473/j.djmkg.v50.i3.p116-120. mandibular ramus and mental foramen on age and gender using digital panoramic
[31] D. Stern, P. Kainz, C. Payer, M. Urschler, Multi-factorial age estimation from radiographs: a retrospective study, Contemp. Clin. Dent. 9 (2018) 343–348,
skeletal and dental MRI volumes, in: Lect. Notes Comput. Sci., Springer Verlag, https://doi.org/10.4103/ccd.ccd_26_18.
2017: pp. 61–69. https://doi.org/10.1007/978-3-319-67389-9_8. [41] Debbie Guatelli-Steinberg, Paul W. Sciulli, Tracy K. Betsinger, Dental crown size
[32] Nicolas Vila-Blanco, Maria J. Carreira, Paulina Varas-Quintana, Carlos Balsa- and sex hormone concentrations: Another look at the development of sexual
Castro, Inmaculada Tomas, Deep neural networks for chronological age estimation dimorphism, Am. J. Phys. Anthropol. 137 (3) (2008) 324–333, https://doi.org/
from OPG images, IEEE Trans. Med. Imaging 39 (7) (2020) 2374–2384, https:// 10.1002/ajpa.20878.
doi.org/10.1109/TMI.2020.2968765. [42] VandanaM Reddy, Susmita Saxena, Puja Bansal, Mandibular canine index as a sex
[33] W. De Back, S. Seurig, S. Wagner, B. Marre, I. Roeder, N. Scherf, Forensic age determinant: a study on the population of western Uttar Pradesh, J. Oral Maxillofac
estimation with bayesian convolutional neural networks based on panoramic Pathol. 12 (2) (2008) 56, https://doi.org/10.4103/0973-029X.44577.
dental X-ray imaging, in: MIDL 2019 Conf. Abstr. Pap., 2019. https://openreview. [43] A. Schmeling, R. Dettmeyer, E. Rudolf, V. Vieth, G. Geserick, Forensic age
net/forum?id=SkesoBY49E (accessed July 8, 2020). estimation: methods, certainty, and the law, Dtsch. Arztebl. Int. 113 (2016) 44–50,
[34] David Sweet, Forensic dental identification, Forensic Sci. Int. 201 (1-3) (2010) 3–4, https://doi.org/10.3238/arztebl.2016.0044.
https://doi.org/10.1016/j.forsciint.2010.02.030. [44] H. Kaarina, Tooth formation age estimated on a few selected teeth: a simple
[35] Anthony J. Hill, Ian Hewson, Russell Lain, The role of the forensic odontologist in method for clinical use, Procc Finn Dent Soc. 70 (1974) 15–19.
disaster victim identification: Lessons for management, Forensic Sci. Int. 205 (1-3) [45] C. Spampinato, S. Palazzo, D. Giordano, M. Aldinucci, R. Leonardi, Deep learning
(2011) 44–47, https://doi.org/10.1016/j.forsciint.2010.08.013. for automated skeletal bone age assessment in X-ray images, Med. Image Anal. 36
[36] Matthew Blenkin, Jane Taylor, Age estimation charts for a modern Australian (2017) 41–51, https://doi.org/10.1016/j.media.2016.10.010.
population, Forensic Sci. Int. 221 (1-3) (2012) 106–112, https://doi.org/10.1016/ [46] Simukayi Mutasa, Peter D. Chang, Carrie Ruzal-Shapiro, Rama Ayyala, MABAL: a
j.forsciint.2012.04.013. novel deep-learning architecture for machine-assisted bone age labeling, J. Digit
[37] T.C.N. Sandoval, G.P. Sonia Victoria, F.A. Gonzalez, A.J. Robinson, C.I. Contreras, Imag. 31 (4) (2018) 513–519, https://doi.org/10.1007/s10278-018-0053-3.
Use of artificial neural networks for mandibular morphology prediction through
craniomaxillar variables, Univ. Odontol. Bogotá. 35 (2016) 1–6.