An Age Estimation Method to Panoramic Radiographs from Indonesian Individuals

Dental features can be considered as the best candidate feature for post-mortem identification. If ante-mortem data is unavailable, then forensic experts are needed for reducing the search space by creating post-mortem dental profiling. Age is one of important factors in dental profiling. Manual inspection of dental radiographs suffers from two drawbaks, i.e., intraobserver error and interobserver error. This paper proposed a semi-automatic system for age estimation. There are two phases in developing the proposed system, i


Introduction
Forensic radiology is a part of forensic medicine that studies about human identification using postmortem radiographs of human body parts including skeleton, skull, and teeth.The identification is carried out by comparing postmortem data with antemortem data of a subject in order to find similar records.When the identification is performed two weeks after subject's death, a postmortem biometric identifier should be able to handle decay or severe body damage caused by fire or collision [1].Dental features can be considered as the best candidate for postmortem identification.This is due to the strengthness and variation of dental features available.
In traditional methods, dental based identification depends on information such as missing teeth or teeth performance.With the enhancement of dental medicine studies and dental treatment methods by dentists, those traditional methods are unreliable now.Therefore, it is very important to develop new methods using dental features for identification [2], [3].
If antemortem data is not available, then an aid from dental forensic experts is required to reduce the data population space and increase the antemortem search space.This process is generally called as postmortem dental profiling.Information from this process can result in more focused antemortem data searching.
Age is one of many important factors used to develop or define an identity.An age estimation is a procedure commonly performed by anthropologists, archaeologists, and forensic experts.Dental radiographs are inspected and compared with radiograph images in order to produce scores that helps defining subject's age.The inspection process is performed manually.
There are two disadvantages of manual inspection, especially because the characteristics of dental radiograph images that have low contrast.Firstly, the manual inspection may produce an intra observer error, that is, a different inspection analysis result by an expert in two different observation times.Secondly, the manual inspection can result in an error inter observer, that is, a different inspection analysis result by two different experts.
In case of age estimation, some researchers have been using computer aided application for determining parts of dental radiographs (pulp, tooth boundary, root).Examples of the application are AutoCAD2000 [4], Adobe Photoshop [5], etc.However, those applications still need expert decisions such as how to determine pulp area or tooth boundary.Thus, they still suffer from intra observer and inter observer error.
This paper proposes an automatic approach to age estimation using panoramic radiographs.We firstly develop a model for estimation by analyzing dental features on Indonesian individuals.Secondly, using the model from the previous step, we analyze the performance of the model.Lastly, we develop an age estimation system that receives panoramic images and estimate the age of the subject requested.

Research Method
Panoramic radiographs used in this paper were obtained from 31 individuals by the same dentist.The population were from West Java, Indonesia, consisted of 30 women and only 1 man, with ages ranging from 50 to 73 years.
Initially, we developed a model for estimation by analyzing dental features.We measured the length of selected teeth then analyze the measurements using MS Excel 2007 software.We selected teeth from both sides of the jaw and only those teeth presented in [6] were selected.
Figure 1 shows the international dental numbering system which we used as our numbering system in this paper.There are 32 teeth in adult people, sixteen teeth on each jaw.There are two jaws, maxilla and mandible.Each jaw is divided into two groups, left and right.Thus, each group consists of eight teeth comprised of two bicuspid, one cuspid, two premolar teeth, and three molar teeth.In this research we only selected maxillary central and lateral incisors, maxillary second premolar, mandibular lateral incisor, mandibular canine, mandibular first premolar (see Figure 1).

Right Maxilla
Left Maxilla After measuring manually the lengths of selected teeth, we analyze the correlation value each tooth's length and the age.Next, we derived linear regression models using tooth lengths that were significantly correlated with age.
In addition, we developed an automatic system for computing tooth lengths.Given a digital panoramic radiograph, users can determine the top left and the right bottom boundary of a particular tooth.Next, the system will process this Region of Interest (RoI) into a binary image using our previous methods in [7].Using the binary image, we can compute the tooth length using a vertical projection based method [8,9].Using the derived regression model, we estimate the age based on the tooth length.The design of our proposed age estimation system is as shown in Figure 2.

Results and Analysis
The measurements from manual observations of selected tooth length are shown in Table 1.Some images does not contain some teeth, therefore we put NA (not available) and do not take into account those teeth in our next stages.
After we measured the selected tooth length, next we calculate the correlation value between the lengths of each numbered tooth with the chronological age provided in our dataset as in Equation (1).
The correlation between selected tooth length and person's age is as shown in Table 2.We can see that the higher correlation score is achieved by canines in mandibular (left and right) followed by premolars also in mandibular (lower jaw).
Using the highest correlation value as in Next, we computed the estimation age using formula in Equation (2).Our results showed that the average absolute difference between the predicted and the chronological age was relatively small, i.e. 4.2 years.
Based on the derived estimation formula, we developed an automatic system able to estimate age based on a dental panoramic radiography.Firstly, the user provides top-left and bottom-right corners of a canine.Our system will automatically crop the panoramic radiographt  As an illustration, Fig. 2 shows a sample radiography with a known chronological age = 66.Fig. 3 (left) shows a cropped ROI, which is a mandibular canine and it's binary version is as shown in Fig. 3 (right).The computed length was 516 pixels, resulting in an estimated age = 64.6936.Table 3 shows the results of our estimation system on our dataset.The average absolute error of our estimation system was 5.2 years.

Conclusion
The proposed system firstly asks users to select an ROI, i.e the top-left and bottom-right corners of a mandibular canine.Figure 4 and 5 show this process.This process still needs human interaction.Figure 6 shows the resulted ROI and the output of the system.Future works may develop a fully automated system that is able to define a particular tooth automatically.This research uses a limited number and limited race of Indonesian individuals, i.e.Javanese people only.Therefore, in the future, the database should be added by more individuals from different races.
However, our experiments showed a promising estimation result, i.e. an average absolute error of 5.2 years, compared to application of the Kvaal method to panoramic radiographs from Turkish individuals that yields a difference of more that 12 years [6] .

Figure 1 .
Figure 1.A system of dental numbering in adults

Figure 2 .
Figure 2. Design of the proposed age estimation system

Figure 4 . 205 Figure 5 .
Figure 4.A process of defining the top-left corner of an ROI

Figure 6 .
Figure 6.The output of the proposed system.

Table 2 ,
we develop a regression model to estimate age based on mandibular canine (tooth number 27) lengths.The estimation formula derived was as in Equation (2).

Table 1 .
of interest (ROI) based on the corners provided.Next, the system enhances the ROI and transforms it into a binary one.After that, the tooth length can be computed based on it's vertical integral projection.Finally, using a particular age estimation formula, a predicted age can be computed.Manual measurements TELKOMNIKA Vol.11, No. 1, March 2013 : 199 -206 202 into a region

Table 2 .
Correlation Analysis Results An Age Estimation Method to Panoramic Radiographs from Indonesian … (AnnyYuniarti)

Table 3 .
Estimation Results