A Simplified Method for Assessing Bone Age in Adolescents Authors: Yehuda Limony, Alon Haim, Michael Friger, Emad Alamour, Eyal Kristal, Eli Hershkovitz Published: 2026 ABSTRACT ABSTRACTObjectiveTo develop and evaluate a simplified method for skeletal age assessment in adolescents as a rapid alternative to the Greulich–Pyle (GP) method.DesignRetrospective methodological study comparing the simplified method with the GP atlas, with duplicate readings by experienced and inexperienced raters.PatientsOne hundred and seventeen hand and wrist radiographs (67 boys, 50 girls) from adolescents referred for short stature but otherwise healthy. Radiographs with GP‐assessed bone age ≥ 9 years in girls and ≥ 11 years in boys were included.MeasurementsBone age was scored independently using GP and simplified methods in two rounds by two paediatric endocrinologists and, for the simplified method, two residents. Outcomes included measures of agreement between methods, the proportion of readings within ±1 year of GP, and inter‐ and intra‐rater reliability.ResultsAmong experienced raters, the simplified method showed minimal bias relative to GP, with about 91% of readings within ±1 year. Reliability coefficients for both methods exceeded 0.9. Repeatability of the simplified method was slightly lower than GP but remained within clinically acceptable limits and close to predefined acceptability thresholds for inexperienced raters.ConclusionsThe simplified method offers a transparent, standardised, and reproducible approach to adolescent skeletal age assessment. Although not intended to replace the GP method, it is particularly suited to research involving large datasets or resource‐limited environments and to population‐level skeletal analyses. In busy clinics, it may also serve as a quick check of radiologists' GP‐based interpretations, allowing review of reported bone ages without repeating a full atlas assessment. Introduction The assessment of skeletal maturity, commonly referred to as 'bone age', is crucial in evaluating child growth and sexual development. This is typically done by comparing hand and wrist radiographs with standard reference images, commonly using the Greulich–Pyle (GP) or Tanner–Whitehouse (TW) methods, which are widely regarded as the two conventional reference methods for bone age assessment. These traditional methods require detailed comparisons of each bone in the hand with atlas standards, making the process time‐consuming [1, 2]. Manual bone age assessment requires non‐trivial interpretation time. In a direct comparison, King et al. reported mean assessment times of 7.9 min for the Tanner–Whitehouse 2 (TW2) method and 1.4 min for the Greulich–Pyle (GP) method [3]. Although GP is relatively quick for a single radiograph, the cumulative workload becomes substantial when hundreds of radiographs must be reviewed, particularly when repeat readings or secondary checks are required. This practical limitation motivated the development of our simplified method. We decided not to use the commonly applied rapid modified GP approach, in which the overall appearance of the radiograph is compared with reference radiographs and the nearest match is selected, because although it is faster than the original GP method, it remained too slow for our purposes of rapidly reviewing and interpreting large volumes of radiographs. Additionally, because it relies on overall visual matching rather than a systematic evaluation of specific bones, it is less standardised and may be associated with greater inter‐observer variability [1, 2, 4]. Prior studies have proposed simplified methods for pubertal children that focus on subtle changes in individual phalangeal epiphyses [5, 6], but the small size of these bones and the subtlety of the changes make them difficult to assess reliably. These limitations prompted the development of our present simplified method, which uses a single, clearly recognisable bone or epiphysis per age group to enhance speed, consistency, and ease of learning. Non‐computerised methods rely on a bone age atlas, limiting efficiency in high‐volume settings. A method based on easily memorised descriptions of a few key bones could eliminate the need for an atlas and expedite interpretation, especially in large research datasets or when reviewing radiologists' reports in busy clinics. This study introduces a simplified bone age method, evaluates its accuracy against the GP method, and assesses reliability among both experienced and inexperienced raters. The primary aim of the new method was to support research involving large cohorts, where bone age is interpreted mainly at the group level, for example when estimating mean or median bone age at specific pubertal stages. Accordingly, the key performance requirement was minimal systematic bias relative to the Greulich–Pyle (GP) method, so that group‐level results obtained with the new method would remain comparable to those derived from GP assessment. A secondary aim is to provide a rapid review tool for clinical use. For this purpose, we defined acceptable agreement with the GP method as a mean bias not exceeding ± 0.5 years, and differences within ± 1.0 year for at least 90% of readings. These thresholds are not absolute standards but were pragmatically selected based on the known variability of the GP method, with reported inter‐rater differences often reaching up to 2.5 years [1, 7]. This design also facilitates skeletal maturity estimation in non‐clinical datasets, including research samples and historical collections. Materials and Methods Development of the Simplified Bone Age Method The simplified method was developed using images of boys aged 11–19 years from the GP atlas. One bone was selected for each age based on clear maturational changes and clinical experience. These selections are summarised in Table 1 using the original GP atlas terminology. Notes in the table reflect practical insights from clinical use. In addition, an auxiliary diagram (Figure 1) illustrating the hand and wrist bones was created to accompany the table. Together, Table 1 and Figure 1 were used by the examiners during the study and are provided here for potential practical use by clinicians and researchers. Table 1Bone maturational stages and their corresponding bone ages in males and females, based on key ossification features identified in the Greulich and Pyle atlas (Greulich and Pyle 1959).Ref. number in the auxiliary diagramDescriptionMale bone ageFemale bone ageIThe pisiform is now more distinct, its shadow being visible through and between the hamate and triquetral.119IIThe space between the epiphysis of the first metacarpal and the trapezium is reduced further and is very narrow, especially in the ulnar side.11.59.5IIIThe outline of the hook of the hamate is visible.a The distal space between the scaphoid and the lunate is very narrow.b 12.510.5IVThe ossification centre of the sesamoid bone is visible.c 1311VAll of the proximal epiphyses have begun to cap their shafts. The process is more pronounced in the volar than on the dorsal side of the epiphysis of the thumb, and on the radial than on the ulnar side of the epiphysis of the fingers.13.511.5VIThe epiphysis of the radius has now begun to cap its shaft.1412VIIFusion of the epiphysis of the distal phalange of the thumb and third finger1513VIIIFusion of epiphysis of all distal phalanges15.513.5IXFusion of epiphysis of all proximal phalanges1614XFusion of all epiphysis of the middle phalanges1715All the epiphyses except that of the radius have fused with their shafts1816All the epiphyses have fused with their shaft1917 a Both the lateral and medial sides of the hook are visible. b When only this change is evident, and the outline of the hook of the hamate is not visible as described in I, then a bone age of 12 in boys and 10 in girls should be assigned. c The sesamoid bone should be clearly visible at this stage. Bone maturational stages and their corresponding bone ages in males and females, based on key ossification features identified in the Greulich and Pyle atlas (Greulich and Pyle 1959). Both the lateral and medial sides of the hook are visible. When only this change is evident, and the outline of the hook of the hamate is not visible as described in I, then a bone age of 12 in boys and 10 in girls should be assigned. The sesamoid bone should be clearly visible at this stage. Auxiliary diagram of the bones of the left hand and wrist, intended for use alongside Table 1. Roman numerals indicate maturational stages corresponding to those listed in the table. Each pair of numbers shown next to a bone represents the bone age, in years, associated with that stage; the left number refers to males and the right number to females. Because female skeletal maturation typically precedes that of males by approximately 2 years (as clearly shown in standard bone age atlases), we assigned female bone age values by referencing male reference images from boys 2 years older. For example, male images representing bone ages of 11.5, 12.5, and 13.5 years were used to represent bone ages of 9.5, 10.5, and 11.5 years in females, respectively. As a result, unlike assessments made using a standard atlas—where intermediate values such as 9.5 years can be assigned based on subjective interpretation of transitional features between two atlas images—the simplified method assigns discrete values based solely on the appearance of a specific male reference image. Thus, when assigning a bone age of 9.5 years to a female using our method, it means the radiograph corresponds to the characteristics of the male image labelled 11.5 years in the GP atlas. To facilitate rapid scoring, the skeletal ages assigned to the simplified stages were expressed in whole‐ or half‐year units. Where the corresponding GP reference ages fell between these values, they were rounded to the nearest half‐year. This rounding reduces the number of possible categories and simplifies application without meaningfully affecting average agreement with the original GP ages. Study Sample and Design The study included 67 radiographs of adolescent boys and 50 of adolescent girls, randomly selected from our paediatric endocrine clinic at Soroka Medical Centre (2005–2015). These children were referred for short stature evaluation, followed for at least 2 years, and found to have normal growth and no skeletal disorders. Of all radiographs examined, we included only those with a bone age of at least 9 years in girls and at least 11 years in boys, as these are the lower age limits for which the new method is applicable. Bone age for this inclusion criterion was taken from the routine Greulich–Pyle assessment recorded in the clinical chart at the time of evaluation by an experienced paediatric endocrinologist. Two paediatric endocrinologists experienced in bone age interpretation and two paediatric residents who had no prior experience in bone age interpretation participated. In the first round, the experienced examiners interpreted all radiographs twice using the GP method with a washout period of over 1 month between sessions. In the second round, all four examiners assessed the images twice using only Table 1, and they had no prior experience with the simplified method. The interval between readings again exceeded 1 month. Ethics The local ethics committee of Soroka University Medical Centre reviewed and approved the research proposal with a waiver of informed consent. Approved number SOR‐0018‐15. Statistical Methods Normality Testing Normality of the data distribution was assessed using the Shapiro–Wilk test. Bland–Altman Analyses A Bland–Altman plot is a visual method for analysing the agreement between two measurement methods. The x‐axis shows the average of the two methods for each radiograph, while the y‐axis shows the difference between them. If the methods agree well, most points will cluster around zero, without a clear trend. The central solid line represents the mean difference (or bias), and the dashed lines show the limits of agreement, defined as the mean difference ±1.96 times the standard deviation of the differences [7]. Repeatability Within‐subject standard deviation (repeatability) was calculated as the standard deviation of paired differences divided by √2, which represents the expected variability between repeated readings by the same rater. Selection of ICC Models and Rationale To also allow comparison with previous studies that examined inter‐ and intra‐observer agreement, we chose to analyse reliability using the Intraclass Correlation Coefficient (ICC). We included both ICC and Bland–Altman analyses, as they provide complementary information. ICC offers a single numerical estimate of reliability, which facilitates comparison across studies. In contrast, Bland–Altman analysis visualises the differences between methods and helps detect systematic bias or trends in disagreement. However, since Bland–Altman does not yield a single reliability coefficient, it is less suited for direct comparison between studies. We selected intraclass correlation coefficient (ICC) models based on the measurement structure and purpose of agreement assessment. For intra‐observer reliability, we used ICC [1, 4], a two‐way mixed‐effects model for consistency in repeated measurements by the same rater. For inter‐observer reliability, we used ICC [1, 2], a two‐way random‐effects model for absolute agreement between raters. Both models are based on two‐way ANOVA, where the variance is partitioned into components attributable to subjects, raters, and measurement error. The ICC values are calculated using these variance components and reflect either consistency (excluding rater bias) or absolute agreement (including it). These definitions are consistent with those recommended in methodological reviews, including Koo and Li [8]. Sample Size and Power Analysis Because this was a retrospective study based on existing radiographs, the number of available cases was fixed and not determined through prospective sample size planning. Therefore, a post hoc power analysis was performed to evaluate whether the final sample was sufficient to detect clinically meaningful levels of inter‐ and intra‐observer reliability. The dataset comprised 117 radiographs (50 females and 67 males), each assessed by four raters. Using a two‐way ANOVA model and assuming a significance level of α = 0.05, this design provides greater than 99.9% power to detect an intraclass correlation coefficient (ICC) of 0.80, with a confidence interval width of ± 0.10. These results confirm that the available sample size was more than adequate to support the reliability analyses conducted in this study. Agreement Threshold and Acceptability Criterion As noted by Bland and Altman [9] agreement between two measurement methods is inherently limited by their respective repeatabilities. If the reference method—here, the GP approach—shows substantial variability between repeated measurements by the same examiner, then even a more accurate new method may appear to disagree, simply due to the imprecision of the reference itself. Therefore, as a first step, we quantified the repeatability of the GP method through repeated assessments by experienced raters. Based on those results and clinical judgement, we established an a priori benchmark: the simplified method would be considered acceptable if at least 90% of its readings fell within ±1.0 year of the GP estimates. This threshold was chosen as reasonable because the simplified method is not intended to replace the full GP‐based evaluation, but rather to serve as a quick secondary review or screening tool, particularly useful for identifying potentially inaccurate interpretations in high‐volume clinical or research settings. Statistical Software All data were analysed using a computerised statistical software programme (SPSS version 17.0 for Windows; SPSS, Chicago, IL). Results Normality Check We assessed the normality of the differences between the simplified and GP methods using the Shapiro–Wilk test and found no significant deviation (W = 0.981, p = 0.069). Within‐Rater Variability of Bone Age Assessments (Repeatability) Repeatability, defined as the within‐rater (intra‐rater) standard deviation of paired readings divided by √2, was expressed in years to match the units of bone age estimation. Among experienced examiners, the overall repeatability was 0.33 years for the GP method and 0.41 years for the simplified method in females, and 0.31 years (GP) and 0.35 years (simplified) in males. For the inexperienced examiners using the simplified method, repeatability was 0.26 years in females and 0.44 years in males. Among experienced raters, 90% of repeated GP assessments differed by no more than 0.75 years in females and 1.0 year in males, with some raters achieving complete agreement within ±1 year. These results indicate that both methods yield consistent readings within each rater. Slightly higher variability was observed for the simplified method, particularly in female assessments by experienced raters and in male assessments by inexperienced raters. However, the observed differences in repeatability—amounting to less than 3 months—are not clinically significant and fall within an acceptable range for methods intended for rapid review or large‐scale application. Agreement Between Methods (Bland–Altman) The Bland–Altman plots demonstrated general agreement between the simplified and GP methods. For boys, the average difference was negligible (+ 0.02 years), indicating no systematic bias, with limits of agreement ranging from −1.29 to +1.33 years. For girls, the simplified method yielded slightly lower bone age estimates on average (mean difference −0.14 years), with limits of agreement from −1.63 to +1.35 years. Importantly, these limits are consistent with the known repeatability of the GP method, suggesting that the observed variability likely reflects the inherent measurement imprecision of the standard method. Applying a predefined threshold of clinical acceptability (≥ 90% of differences within ±1 year), 91.5% of differences for boys and 91.0% for girls met this criterion, supporting the validity of the simplified method. Visual inspection of the Bland–Altman plots (Figures 2 and 3) confirms that over 91% of observations fall within ±1 year of the GP estimates, as indicated by the green dotted reference lines. Bland–Altman plot showing agreement between the Greulich–Pyle (GP) and simplified methods for skeletal age estimation in females. The solid black line indicates the mean difference (bias), and the dashed red lines indicate the 95% limits of agreement. Bland–Altman plot showing agreement between the Greulich–Pyle (GP) and simplified methods for skeletal age estimation in males. The solid black line indicates the mean difference (bias), and the dashed red lines indicate the 95% limits of agreement. Among the inexperienced examiners, Bland–Altman analysis comparing the simplified and GP methods showed a mean difference of + 0.03 years in females, with limits of agreement from –1.83 to +1.89 years. In males, the mean difference was + 0.04 years, with limits of agreement from –2.03 to +2.11 years. When combining both first and second assessments from the inexperienced examiners, 76.7% of simplified readings in females and 85.4% in males were within ± 1 year of GP estimates. While these results do not comply with our predefined threshold of clinical acceptability, they approach that threshold and may improve further with additional training. Intra‐ and Inter‐Observer Reliability Intra‐rater reliability was assessed using the ICC [1, 4] model. Among experienced raters, both the GP and simplified methods demonstrated excellent reliability, with the GP method showing slightly higher values. Among inexperienced raters using the simplified method, intra‐rater reliability remained high and was even slightly higher for females than for females assessed by experienced raters, reaching statistical significance (z = −3.01, p = 0.003), while no significant difference was found in males (z = −0.64, p = 0.521). Inter‐rater reliability, assessed using the ICC [1, 2] model, was also high for both methods, though slightly lower for the simplified method in females. Overall, these results indicate that the simplified method provides reproducibility that is closely aligned with the GP method, even among raters with limited experience. Full ICC values with 95% confidence intervals are provided in Table 2. Table 2Intraclass correlation coefficients (ICCs) and 95% confidence intervals.Assessment groupMethodRater experienceICC95% CI lower95% CI upperFemales – Intra‐raterGPExperienced0.9700.9560.980Males – Intra‐raterGPExperienced0.9780.9560.988Females – Intra‐raterSimplifiedExperienced0.9150.8580.953Males – Intra‐raterSimplifiedExperienced0.9660.9000.981Females – Intra‐raterSimplifiedInexperienced0.9690.9360.989Males – Intra‐raterSimplifiedInexperienced0.9750.9330.994Females – Inter‐raterGP—0.8770.8070.923Females – Inter‐raterSimplified—0.7800.6790.862Males – Inter‐raterGP—0.9360.8740.964Males – Inter‐raterSimplified—0.9020.8140.945 Intraclass correlation coefficients (ICCs) and 95% confidence intervals. Discussion The comparison with previous studies demonstrates that the variability observed among our examiners using the Greulich–Pyle method is consistent with that reported in the literature. For example, Groell R [10]. reported a wide range of intra‐observer differences, with means ranging from –1.3 to 0.3 years and standard deviations between 0.36 and 0.75 years. Inter‐observer variability in his study ranged from 0.1 to 2.3 years in mean differences, with standard deviations from 0.04 to 0.94 years. Johnson G.F [11]. found intra‐observer mean differences between 0.15 and 0.41 years, with standard deviations ranging from 0.22 to 0.48 years. Inter‐observer mean differences in his study ranged from 0.38 to 0.84 years, with standard deviations between 0.12 and 0.33 years. In a more recent study, Celis‐Moreno et al. [12] reported intra‐observer standard deviations of 0.39 years (less experienced examiner) and 0.28 years (experienced examiner), and an inter‐observer SD of 0.39 years. In our study, experienced examiners applying the GP method yielded intra‐observer SDs of 0.30–0.33 years and inter‐observer SDs of 0.45–0.53 years—results that fall well within the range of previously reported values. This correspondence suggests that the performance of our examiners is comparable to that of other researchers in the field and supports the reliability of the GP‐based assessments conducted in our study. To contextualise the clinical acceptability of the simplified method, we evaluated the repeatability of the GP method itself. These findings confirm that even the reference method demonstrates inherent variability across repeated assessments. Therefore, the simplified method's performance—where 91.0% (females) and 91.5% (males) of readings fell within ±1 year of GP estimates—falls well within the expected range of variability for the standard method. This supports the clinical validity of using a ± 1‐year threshold as a benchmark for agreement. Although the Bland–Altman plots show limits of agreement that exceed ±1 year in both sexes, this reflects the influence of a small number of outliers. Over 91% of cases lie within ±1 year—a threshold derived from the repeatability of the GP method itself. The plots visually confirm this level of agreement and demonstrate that, despite occasional discrepancies, the simplified method aligns closely with the GP method across the typical bone age range. This supports its reliability for clinical screening and large‐scale applications. Normality was confirmed by the Shapiro–Wilk test (see Results), supporting the use of Bland–Altman and ICC analysis. Bone ossification varies among hand bones, and minor discrepancies in maturation may occur when a selected bone matures at a slightly different rate from others. If such variability is common, it is likely already reflected in the observed differences between the simplified and GP methods. If it is less frequent, its overall impact on our findings would be correspondingly smaller. In either case, the potential effect of relying on a single bone does not constitute an additional source of error beyond what has already been captured in our analysis. Moreover, Celis‐Moreno et al. [12] reported statistically significant but generally modest skeletal age differences between individual hand bones during puberty, further supporting the validity of our approach while acknowledging some variation between bones. Although bone age estimation is traditionally performed by radiologists, human error remains a factor [13]. While automated bone age interpretation tools are becoming increasingly common, they are not without limitations. Most are trained on specific clinical populations and rely on proprietary algorithms that lack transparency. This raises concerns about generalisability to research samples from diverse backgrounds or historical periods. Furthermore, such tools are often inaccessible in field research settings or when working with analogue or archived radiographs. Chang et al. explicitly addressed the risk of automation bias in AI‐assisted bone age assessment. In their randomised crossover study, AI support significantly improved radiologists' accuracy, but the authors warn that 'automation bias happens when users become over reliant on AI support, which reduces vigilance in information seeking and processing', and note that junior radiologists were particularly prone to accepting AI estimates. They further caution that attempts to save time may encourage radiologists to accept AI age predictions without sufficient scrutiny, potentially introducing new types of interpretive error despite the overall gain in accuracy [14]. Therefore, a structured control procedure such as the simplified review method proposed in the present study may provide a practical safeguard against automation bias, by ensuring that AI‐derived bone age estimates remain subject to consistent expert oversight. In contrast, the simplified method presented here offers a transparent and reproducible approach that is well suited for large‐scale studies. Because it relies on qualitative morphological features rather than population‐specific reference distributions, it is conceptually independent of any particular population; however, its empirical performance still requires validation in other settings. This is particularly valuable in settings where interpretive consistency across observers and time periods is critical, and where atlas‐based or AI‐based tools may be unavailable or inappropriate. Recent work by Khadilkar et al. [15] has also emphasised the need to simplify bone age assessment by proposing a method based on three representative bones of the hand and wrist. While their approach seeks to balance the accuracy of traditional methods with greater efficiency, it still relies on identifying stages from multiple bones and may require familiarity with atlas‐based interpretation. In contrast, the present study offers a visually intuitive, single‐bone reference approach. Intra‐ and inter‐rater reliability with the simplified method was consistent among both experienced and inexperienced raters. Study Strengths and Limitations A major strength of this study is its inclusion of both experienced and inexperienced raters, demonstrating that the simplified method maintains high reliability across varying levels of expertise. The method's low cost, speed, and independence from specialised equipment further enhance its applicability in clinical and research settings. One limitation is that the sample consisted exclusively of adolescents referred for short stature evaluation. Although this reflects common practice in paediatric endocrinology, it may limit generalisability to children with average or tall stature. Additional validation in broader paediatric populations is warranted. Another important limitation is that the inexperienced raters assessed the radiographs only using the simplified method, without performing parallel evaluations using the Greulich–Pyle method. As a result, their performance cannot be directly compared to their use of the standard method, and conclusions regarding the simplified method's robustness across experience levels should be interpreted with caution. Finally, to streamline the simplified decision tool, female age 8 years 10 months was rounded to 9 years. This minor adjustment is not expected to influence clinical interpretation meaningfully. Conclusions The simplified method for assessing bone age in adolescents demonstrated high reliability and strong agreement with the Greulich–Pyle (GP) method, despite its reliance on a single bone per age group. Intra‐ and inter‐rater variability remained within clinically acceptable limits, even among inexperienced raters, and more than 90% of readings fell within ±1 year of the GP estimates. These findings support the method's utility as a rapid, low‐cost option for large‐scale research, particularly in studies of adolescent growth, maturation timing, and population‐level variation in skeletal development. While not intended to replace detailed GP assessments, the simplified approach offers a practical tool to enhance consistency and efficiency in skeletal age research and secondary evaluations. Further validation in more diverse paediatric populations is recommended. Funding The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.