Using Artificial Intelligence and the analysis of the pubic symphysis to estimate age-at-death
Published on November 23, 2022
Written by Andrea Valsecchi
In this article, we will focus on the estimation of age-at-death, and in particular on the latest efforts toward improving a well-established family of methods using Artificial Intelligence, those involving the analysis of pubic symphysis following the principles of Todd’s method . In these methods, an age phase (i.e. an age interval) is assigned to the individual based on morphological characteristics of the pubic symphysis. Todd’s method  was the first of these kinds of approaches, while later methods  have either modified the pubic bone characteristics being used or the age phases considered.
Todd’s study  was published in 1920. He analyzed 306 skeletons of Caucasian males in the range of 18–60 years to establish 10 development and degenerative stages (Figure 1). To do so, Todd defined a morphological description of the pubic symphysis in each age-at-death phases and accompanied them with photographs to facilitate comparisons. There are clear methodological limitations to Todd’s proposal, starting with the reliability of the sample used, which was composed of unidentified corpses whose age was estimated during the autopsy. In addition, pubic symphysis that did not meet the expected standards was eliminated from the sample. Finally, the obtained results were analyzed only based on his experience, without any statistical analysis. Nevertheless, despite these criticisms regarding the design of the method, the criteria that Todd used to describe the degenerative process of the pubic bone more than 100 years ago are still used by the methodology recommended in the area nowadays.
Revising Todd’s method using AI
In , researchers from the University of Granada (UGR) investigated the use of AI to improve the decision-making of Todd’s method. They used the same morphological characteristics of the pubic symphysis and the same ten age phases as in the original method, but they used machine learning to learn how to assign the correct age phase based on the morphology of the symphysis.
In the first part of this new method, a forensic analyst is required to assess the morphology of the pubic symphysis, just like with Todd’s method, but following a more systematic approach. The assessment has to determine nine different traits of the morphology, each with a specific number of possible values. Figure 2 provides a graphical representation of some of these traits. The nine traits are the following:
- Articular face describes the obliteration process of the epiphysis on the symphyseal surface and has been divided into 6 levels according to the presence of ridges and grooves.
- Irregular porosity describes the extensive erosion of the bone surface, characterized by the progressive appearance of irregular porosity. It is divided into 3 levels.
- Upper symphysial extremity, Lower symphyseal extremity, Dorsal margin and Ventral margin describe the formation of the upper, lower, dorsal and ventral margins, respectively. Each variable has 2 values for the presence or absence of the respective margin, except for variable 9, which includes 3 more levels to describe the degenerative process of the ventral margin.
- Bony nodule describes the formation of the ossification nodule in the upper margin. It considers 2 values for presence or absence.
- Dorsal plateau indicates the presence or absence of texture difference between the dorsal half and the ventral half. It thus considers 2 values.
- Ventral bevel describes the progressive and beveled elevation of the ventral area. It is divided into 3 level
In the second part of the new method, the assessment is used to select the correct age phase. To do so, the researchers used a recent AI technique called NSLVOrd . The result is a set of simple IF-THEN rules (see Figure 3), where the IF part of the rule involves one or more values assigned to the morphological traits, while the THEN part just determines what age phase is going to be assigned to the individual. When multiple rules can be applied to the same individual, all such rules are considered and the final age phase is determined by a weighted average of the age phase provided by each rule.
When AI is used, the quality of the results is affected by both the quality and the quantity of the data available. In this study, the authors used the Pubic Symphysis Collection at the UGR Physical Anthropology Lab. The sample considered results from autopsy studies developed since 1991 in a collaboration with the Institute of Legal Medicine and Forensic Sciences. It includes 837 individuals in the range of 17–82 years, 197 women, and 637 men. Detailed information is available regarding sex, age, cause of death, and, in many cases, additional information such as approximate weight, alcohol or drug consumption, and population origin.
To evaluate the performance of the novel method, a comparison was made against other age estimation techniques based on pubic symphysis. Once the age phase has been estimated, the middle age interval is used as the estimated age. The results are then measured in terms of the average error, simply the average difference between the correct age and the estimated age. Table 1 reports the results of the comparison.
The new method obtained the lowest average error score in the comparison: the effort of applying AI paid off. For further details on this study, please refer to the original article available here.
Further improvements: going beyond age phases
A key aspect of Todd’s method is the use of age phases as opposed to direct estimation of the age of the individual. This made sense for a method that was supposed to be carried out by a forensic analyst: using ten phases provides a good tradeoff between the complexity of the method and the granularity of the estimation. However, once AI is involved in the process, there is no longer the need to use a small number of age phases with a custom design. Instead, age can be assessed directly: a mathematical formula can be created so that the age can be calculated from the values assigned to morphological traits. This is the blueprint for a novel study carried out at the UGR. Again, researchers use AI to find the best-performing formula; this is indeed a classic machine learning problem. Preliminary results show that the average estimation error obtained by the new approach can be as low as 10.8 years. This large improvement in precision provides another success story for the application of AI to Forensic Anthropology. The study is currently undergoing publication. Both methods will be available soon as part of the Biological Profile Estimation module of Skeleton-ID.
 T.W. Todd, Age changes in the pubic bone, Am. J. Phys. Anthropol. 3 (3) (1920) 285–328.
 B. Gilbert, T.W. McKern, A method for aging the female os pubis, Am. J. Phys. Anthropol. 38 (1) (1973) 31–38.
 Gámez, J.C., Irurita, J., González, A., Damas, S., Alemán, I., Cordón, O.: Automating the decision-making process of Todd’s age estimation method from the pubic symphysis with explainable machine learning. Information Sciences (2021).
 J.C. Gámez, D. García, A. González, R. Pérez, Ordinal Classification based on the Sequential Covering Strategy, Int. J. Approximate Reasoning 76 (2016) 96–110.
 D. Slice, B. Algee-Hewitt, Modeling bone surface morphology: A fully quantitative method for age-at-death estimation using the pubic symphysis, J. Forensic Sci. 60 (4) (2015) 835–843.
 D. Stoyanova, B. Algee-Hewitt, D. Slice, An enhanced computational method for age-at-death estimation based on the pubic symphysis using 3D laser scans and thin plate splines, Am. J. Phys. Anthropol. 158 (3) (2015) 431–440.
 D.K. Stoyanova, B.F.B. Algee-Hewitt, J. Kim, D.E. Slice, A computational framework for age-at-death estimation from the skeleton: Surface and outline analysis of 3D laser scans of the adult pubic symphysis, J. Forensic Sci. 62 (6) (2017) 1434–1444