Materials and methods

Study design, population, and specimens sampling and preparation

This study arises from a collaboration of the Institute of Legal Medicine and the Institute of Radiology of the University of Rome “Tor Vergata”, and was designed as an observational, double-blind reliability and agreement study of Iscan’s method for age estimation using CT reconstructions of the sternal end of the fourth right rib.

A convenience sample was used, comprising right fourth ribs collected at the Institute of Legal Medicine. The final dataset included 112 sternal rib ends from individuals of European origin with documented age at death. Cases were excluded in the presence of:

anterior chest wall trauma,history or macroscopic evidence of chronic respiratory disease,congenital abnormalities or pathological alterations of the anterior chest wall.

Ethical approval was obtained from the Independent Ethics Committee of Policlinico Tor Vergata (protocol no. 71/17), authorizing bone retrieval during routine autopsies.

Ribs sampling followed a methodology already described in one of our previous morphological research in a wider sample [11]. During autopsy, rib segments were removed by incisions approximately 3 cm medially and laterally to the costochondral junction. Each specimen was labeled with a unique identification code to ensure pseudo-anonymization, while demographic data (gender, age, and height) were recorded separately.

Specimens underwent cleaning and preparation according to the described procedure [11] consisting of:

maceration in soap and water to remove soft tissues,brief boiling (10–15 min) to facilitate debridement,manual removal of residual tissues and cartilage,air-drying for approximately 20 days.

Radiological evaluation and data collection

After preparation, all fragments were scanned using using a 128-slice scanner (GE Medical System, Revolution CT) with the following parameters: slice acquisition 1.25 mm, pitch 0.5; rotation time 0.5 s, tube voltage 120 kVp, with adaptive mA. Images were reviewed and reconstructed using the MPR protocol, and 3D images were created using the Volume Rendering protocol for the CT evaluation.

Specimens were positioned anatomically on a radiolucent foam support to avoid reconstruction artifacts (Fig. 1).

Fig. 1: The image shows the placement of the bone specimen on the foam holder with associated laser CT-centering. The addition of the left bone specimen is to show the correct positioning following the anatomical orientation

Images were blindly analyzed by two radiologist observers in two different scoring sessions at a 6-month interval, to avoid memorization bias. Using MPR and VR reconstructions, each rib was classified according to Iscan’s phase system (phases 1–8), based on the CT-assessable morphological parameters:

progressive increase in pit depth,transformation of pit shape (V-shaped to U-shaped),changes in rim and wall morphology,presence and development of osteophytes.

Table 1 summarizes the main characteristics used for classifying each phase and the corresponding age ranges for both sexes, according to Iscan [5, 6].

4th Rib Phase | Pit Depth | Pit Shape | Rim and Wall Morphology | Osteophytes | Iscan Phase-Based Age Males (y.o.) | Iscan Phase-Based Age Females (y.o.)
1 | Shallow | Flat | Regular | - | 17–18 | -
2 | Indented | V-shaped | Thick with rounded margins | - | 18–25 | 16–20
3 | Slightly shallow | Narrow U-shaped | Thick with smooth margins | - | 19–33 | 20–24
4 | Deep | U-shaped | Thin with irregular margins | - | 22–35 | 24–40
5 | Deep | U-shaped | Irregular, thin walls flaring outward | Small | 28–52 | 29–77
6 | Deep | U-shaped | Irregular, thin walls flaring outward | Moderate | 32–71 | 32–79
7 | Deep | U-shaped | Irregular, thin walls flaring outward | Long | 44–85 | 48–83
8 | Wide and deep | U-shaped | Irregular, thin walls flaring outward | Very long | 44–85 | 62–90

Both raw pictures and multiplanar reconstructions were utilized to assess the depth and shape of the pit. To evaluate the rim and wall configuration visually, VR reconstructions were used. Each subject was placed in one of phases 1 through 8. Representative examples illustrating morphological features defining each Iscan phase are shown in Fig. 2.

Fig. 2: Top row: Axial CT images of bone specimens displayed in ascending Iscan phases (bone window). A progressive deepening of the joint fossa is clearly visible as the phase increases. Bottom row: 3D Volume Rendering reconstructions illustrating in detail the progressive appearance and elongation of osteophytes, which further characterize the later phases

During the assessment, it was observed that the samples also displayed variations in the degree of bone sclerosis. The samples were independently re-evaluated by the observers and assigned a sclerosis score on a three-level ordinal scale defined for this study:

0: no sclerosis,1: sclerosis involving < 50% of the articular surface,2: sclerosis involving > 50% of the articular surface.

This simple ordinal classification was designed to provide a reproducible visual assessment of sclerosis extent on CT images.

Sclerosis was identified as cortical hyperdensity of the joint surface (Fig. 3). This parameter is easily quantifiable through direct observation and is highly reproducible across different evaluations. When necessary, common measurement tools can be used for a more precise estimation of sclerosis extent. Figure 3 provides a visual example of the three sclerosis score grades used in this study.

Fig. 3: Visual representation of the sclerosis scoring system. The images illustrate the three degrees of sclerosis observed in the articular facet of the fourth right rib. Score 0: absence of sclerosis, with normal cortical bone appearance. Score 1: sclerosis affecting less than 50% of the joint surface, visible here as spots of cortical hyperdensities. Score 2: sclerosis affecting more than 50% of the joint surface, characterized by diffuse extent of cortical hyperdensity

The measurements were conducted in a randomized sequence, with operators having no access to prior measurement data, clinical information, or any recorded details regarding the specimens. All data were collected using a simple numerical ID system for pseudo-anonymization, ensuring the specimens could be identified by gender while age details — stored in a separate database — remained inaccessible during the analytical phase.

Statistics and reporting

A first descriptive analysis was performed. For each phase, the actual specimen age range was reported using means and standard deviation, as observed by each operator, separately for male and female subjects. To assess the expected observation for each operator, a prior probability was computed for each rib based on its documented age and the corresponding Iscan phase ranges, defined as:

\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:{\mathrm{P}\mathrm{r}\mathrm{i}\mathrm{o}\mathrm{r}}_{\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}}=\:\left\{\begin{array}{c}0\:\:,\:\:age\:\notin\:\:\left[{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}},{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}}\right]\\\:\:\\\:\:\\\:\frac{1}{\sum\:\mathrm{c}\mathrm{o}\mathrm{m}\mathrm{p}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{b}\mathrm{l}\mathrm{e}\:\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}\mathrm{s}},\:\:age\:\in\:\left[{\mathrm{M}\mathrm{i}\mathrm{n}}_{\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}},{\mathrm{M}\mathrm{a}\mathrm{x}}_{\mathrm{p}\mathrm{h}\mathrm{a}\mathrm{s}\mathrm{e}}\right]\:\:\end{array}\right.$$\end{document}

For example, according to Iscan, a 30-year-old male may be correctly classified as phase 3, 4, or 5, giving each of those three phases a prior probability of 1/3, and zero for all others. Prior probability represents the theoretical phase distribution expected from chronological age used as a reference to evaluate operator classifications.

Inter- and intra-operator reliability were assessed using Weighted Cohen’s Kappa. In this analysis, agreement refers to exact phase assignment between observers and does not consider compatibility with the true age range, which was evaluated separately through the success metric.

For each observation, “success” was defined as assigning a phase compatible with the actual age at death; for each rib, the operator’s success rate was calculated as the number of successful classifications divided by the number of evaluations (two). Because the original Iscan phases present overlapping age ranges, different phase assignments may still be compatible with the same chronological age; therefore, this success metric reflects age compatibility rather than strict phase agreement. These success rates were plotted to obtain a curve modeled with a classical logistic function. To evaluate whether the two observers produced similar distributions of success rates across the examined ribs, differences between operators’ success rate distributions were assessed using the two-sample Kolmogorov–Smirnov test for equality of distributions, a non-parametric test that compares entire distributions without assuming normality, allowing detection of differences in the overall shape and dispersion of success rates rather than only differences in central tendency.

For each phase, prior and posterior probability densities were estimated using Kernel Density Estimation (KDE) with a Gaussian smoothing function and plotted. Kernel Density Estimation (KDE) is a non-parametric smoothing technique that provides a continuous and intuitive representation of probability distributions derived from discrete observations. It was chosen because it allows visualization of the overall pattern of the data without relying on arbitrary histogram binning, which may influence the apparent distribution of observations. In the present study, KDE was used to facilitate comparison between prior and posterior phase distributions and to highlight possible discrepancies between the expected and observed classification patterns.

Estimation error between observed and actual age (underestimation or overestimation) was analyzed by gender and age using Kernel Density Estimation (KDE) with a Gaussian smoothing function. This approach provides a smooth visualization of how estimation errors are distributed across age groups, making areas of higher concentration and possible systematic tendencies toward under- or overestimation easier to identify. The estimation error was represented continuously, with deviations from the centerline indicating the magnitude of the error: larger deviations correspond to greater misclassification. Individual observations were overlaid on the KDE to provide direct visual insight into the distribution of errors across different age groups and sexes.

The Guidelines for Reporting Reliability and Agreement Studies (GRRAS) were used for reporting the results, and statistical significance was set at P = 0.05 for all inferential analyses [12].

To incorporate the sclerosis score and potentially narrow the broad age ranges associated with each phase, two linear regression models were fitted, with the subject’s actual age as the dependent variable. The first model includes only the CT Iscan phase (treated as an ordinal variable), and the second model includes both the Iscan phase and the sclerosis score as predictors. Model performance was compared using changes in the coefficient of determination (R²) and mean squared error (MSE). For each phase and sclerosis category, 95% confidence intervals for the estimated ages were computed to determine whether including sclerosis yielded narrower intervals. A significant improvement in R², a reduction in MSE, or narrower confidence intervals in the extended model was considered evidence that the sclerosis score refines age estimation within each phase. The comparison between the two models was intended to evaluate whether the inclusion of the sclerosis score provided additional predictive information beyond the phase classification alone.