Bone Age in Endocrinology – Part 1: Methods of Bone Age Assessment

Abstract

Bone age assessment is an essential competency for endocrinology residents, guiding the diagnosis and management of growth disorders, pubertal abnormalities and endocrine conditions. This review is presented in two parts: Part 1 focuses on the methods of bone age assessment, while Part 2 addresses its clinical utility and applications. The Greulich and Pyle (G&P) atlas is widely used for its simplicity, while the Tanner–Whitehouse (TW) method offers a more detailed, scoring-based approach but requires more time. In younger children, proximal ossification centres are more informative, whereas distal centres are more informative in older adolescents. TW3 generally estimates slightly younger ages than TW2, and G&P tends to score younger than TW2. Emerging tools—such as automated image analysis, content-based image retrieval and ultrasonography—promise efficiency gains but face challenges of operator dependence and variable reliability. Advances like BoneXpert have improved standardization and reproducibility, yet traditional G&P and TW methods remain central in endocrine practice due to their accessibility, validated references and widespread familiarity.

INTRODUCTION

Skeletal maturation typically progresses with chronological age but may deviate in certain physiological and pathological conditions. Assessing skeletal maturity is crucial in endocrinology, particularly during the evaluation of growth and pubertal disorders. The most commonly used method is bone age assessment (BAA), typically performed using an X-ray of the left hand. However, several other methods exist.

Skeletal maturity assessment (SMA) is not a recent concept. Historically, the Roman Empire used second molar eruption to determine military readiness in young males.[1] Since 1886, carpal bones have been used to estimate the age of the young.[2] Today, BAA is employed not only in endocrinology but also in forensics, youth sports and immigration to assess biological age.

In endocrine practice, SMA and BAA are often used interchangeably. BAA is primarily used to evaluate pubertal and growth disorders and to predict adult height. The two most widely accepted methods are the Greulich–Pyle (G&P) atlas and the Tanner–Whitehouse (TW) method.[3,4] While the G&P method has remained unchanged since 1958, the TW method has undergone several modifications. Although reliable, both are observer-dependent and subject to inter-and intra-observer variability. To address this, automated computer-based BAA tools have been developed using these classical systems.

Bone growth follows a predictable pattern of bone maturation. Once epiphyseal ossification centres appear, they undergo shape modelling before eventually fusing with the shaft. The capitate and hamate are the first carpal bones to ossify, typically within the first year of life. This is followed by the appearance of the distal radial epiphysis and subsequently the epiphyseal centres of the phalanges and metacarpals in a characteristic sequence: proximal phalanges, metacarpals, middle phalanges and distal phalanges, with the ulna appearing last. Notable exceptions include the distal phalanx of the thumb, which appears along with the metacarpals, and the middle phalanx of the fifth digit, which ossifies last. Although BAA can be performed in infants, its reliability is limited below 2 years of age due to the small number of ossification centres and greater interobserver variability; hence, interpretation in this age group should be undertaken with caution. Skeletal maturation progresses in a predictable, age- and sex-dependent manner and can be broadly categorized into distinct developmental stages based on the most informative ossification centres. In early life, the appearance of carpal bones and the distal radial epiphysis are key indicators, whereas in later childhood and adolescence, changes in the size, morphology and eventual fusion of phalangeal epiphyses become more relevant. The sequence of epiphyseal fusion also follows a defined order, beginning with the distal phalanges and progressing through the metacarpals, proximal phalanges, and finally the middle phalanges, with completion of fusion at the distal radius and ulna marking skeletal maturity.[5,6] A stage-wise summary of skeletal maturation, corresponding key ossification markers and major caveats in interpretation is provided in Table 1.

Table 1.

Stages of skeletal maturation: Key ossification markers and interpretative caveats

Stage	Age Range (Females)	Age Range (Males)	Key Skeletal Indicators	Major Caveats in Interpretation
Infancy	Birth to 10 months	Birth to 14 months	Appearance of carpal bones (capitate, hamate) and distal radial epiphysis	Limited reliability due to small number of centres and high inter-individual variability; absence of standardized references
Toddler	10 months to 2 years	14 months to 3 years	Number and sequence of epiphyseal ossification centres in long bones; typically first in middle finger, last in fifth finger (proximal phalanges → metacarpals → middle phalanges → distal phalanges)	Variability in timing of ossification; still fewer markers compared to older children
Pre-puberty	2 to 7 years	3 to 9 years	Increase in size and thickness of phalangeal epiphyses; gradual widening to match metaphyseal diameter	Generally most stable phase; minimal pubertal influence
Early and Mid-Puberty	7 to 13 years	9 to 14 years	Continued enlargement and maturation of phalangeal epiphyses	Influenced by onset and tempo of puberty; variability in timing of peak height velocity
Late Puberty	13 to 15 years	14 to 16 years	Progressive epiphyseal fusion in long bones of the hand (distal phalanges → metacarpals → proximal phalanges → middle phalanges)	Narrow window before complete fusion; subtle changes may be difficult to interpret
Post-Puberty	15 to 17 years	17 to 19 years	Completion of epiphyseal fusion, including distal radius and ulna	Hand/wrist BA not useful; clavicle or other sites required

BAA depends on the expertise of the person (observer-dependent) and could be a reason for variation in reporting. BA is being reported by the radiologists. In contrast, most of the endocrinologists prefer assessing BA by themselves, using the G&P system. Use of the TW system takes a lot of time and expertise, which are limiting factors during busy clinical hours. Sticking to one system at one centre would be better as the interchange between the two systems for one child at different age periods is not advisable.

METHODS OF BONE AGE ASSESSMENT

Many methods are available based on the use of technology to generate images or calculate BA. The main methods are

• (1) X-ray-based methods

• (2) USG-based methods

• (3) MRI-based methods.

X-ray-based

BA is assessed using an X-ray of the left hand, compared against standard reference methods. A high-quality radiograph is essential for accurate assessment, and key precautions are listed in Table 2. The G&P and TW methods are the most widely used in clinical and research settings.[3,7]

Table 2.

Suggested precautions for taking good X-ray for bone age[8]*

• Left-hand radiograph is to be used only.

• The radiograph should include complete fingers, including tip, and at least 1–2 inches of distal radius.

• The hand should be placed with “palm down” flat, extended, in contact with the cassette and palm pressed gently over the cassette.

• The axis of the middle finger should be in a direct straight line with hand, wrist and forearm.

• Rotation of radius and Ulna can increase the variability of results and should be avoided all the time. Rotation is more likely to affect results in TW methods as it may change the appearance of bone density.

• Preferably, the upper arm and forearm should be in the same horizontal plane.

• The fingers should not touch each other; they should spread with a clear gap between the two fingers.

• The thumb should be rotated out in a way that is so natural that the position of the thumb should be at 30 degrees to the index finger.

• For younger children who are not able to follow instructions, the hand may be strapped with a bandage or tape over the cassette.

• The X-ray tube should be centred either over the head of the third metacarpal or midway between the tips of the fingers and the distal end of the radius. The X-ray beam should be perpendicular to the cassette.

• The exposure characteristics—should be told to the X-ray technician—Tube distance—30–36 inches, Exposure details—100 MA, 45–50 KV.

*These instructions were given for the initially conventional X-ray method which has been replaced by the digital X-ray method, so a few instructions may not be applicable today. TW: Tanner–Whitehouse

Greulich and Pyle method (G&P Method)

The G&P method, an atlas of standard left-hand X-rays by age, is the most widely used globally. It is based on radiographs of growing Caucasian children, obtained between 1931 and 1942, from upper-middle-class, above average in education, born in the United States, but of North European ancestry.[3,4,8,9,10] The atlas includes 31 X-ray plates for boys (0–19 years) and 27 for girls (0–18 years). Each of the plates was selected from 100 films of age- and sex-matched children. All these 100 films were arranged in series from least mature to most mature, and the median was selected. BAA is done by matching the subject’s X-ray image with a standard X-ray plate in the atlas. The changes observed in the phalanges should be prioritized over carpal bones due to greater variability in carpal bone development. The assessment compares phalanges and metacarpals’ size, shape and maturation. Finding a 100% match is practically impossible, so the radiograph’s closest matching plate is selected as the BA.[3,4] Its major advantage is that it is user-friendly and less time-consuming (1–2 min). The cost of the atlas is one-time and is approximately INR 10,000 (available online). Nowadays, E-copy is also available and operable on mobile phones. Its major disadvantage is subjective variation and sometimes it is very difficult to decide on one standard, as the patient’s image may fall between the two standards.

Applicability of G&P method to other ethnicities

Since the G&P atlas was developed using Caucasian population data, its applicability to other ethnic groups remains debated. Asian children, especially boys, show delayed skeletal maturity compared to the G&P atlas (P = 0.048). Its use in Indian children has raised some concerns about skeletal retardation of age.[11,12] Recently published meta-analyses suggested that both G&P and TW3 methods are precise, but racial bias can alter their accuracy in estimating BA among different ethnic groups.[10] In the African-Americans, the difference between CA and BA was statistically significant, though marginal, while skeletal maturation in East and South Asians does not correspond to the G&P atlas.[13] Boys have shown delay in skeletal maturity during early and middle childhood, followed by advancement during adolescence. Significant differences between BA and CA were found in two age categories, 6–9 years and 17 years.[14] The clinical implications of these differences are important during BA assessment of Asian boys in early childhood, where it can be interpreted as delayed BA.[13,14,15]

The Tanner–Whitehouse (TW) method

The TW method is based on studies conducted among children in the United Kingdom. Its first edition was published in the 1930s, while the second edition (TW2) was based on data from the 1950s and 1960s.[4,7,10] It was published in 1983 and updated in 2001 as the third edition (TW3). TW2 is based on the original British population of the TW method, while the TW3 system presents a new standardized RUS score using a North American sample.[16] The TW2 method is much more objective but complex, time-consuming and requires definitive training for BA assessors. They examined a number of longitudinal series of radiographs of normal children in the Harpenden Growth Study (Tanner 1962,1981) taken every 6 months over a period of 12 years.[4,7,9,17]

There are three different TW2 methods: the Radius-Ulna-Short bones (RUS) method for evaluating the 13 long or short bones, the Carpal method for evaluating the seven carpals, and the 20 bones method for evaluating the 13 long or short bones and seven carpals. About 50% of children are at the median age when transitioning between stages, which indicates that they relate to age at the first appearance of a stage.[4,17]

Difference between TW2 and TW3

The Skeletal Maturity Score (SMS) or manual ratings attached to each stage of the bones are the same in TW2 and TW3. The significant difference between the two systems is the SMS centile charts for RUS, which are updated in TW3 based on newer studies. Still, the carpal score chart remains unchanged between TW and TW2 except for minor rescaling of the vertical axis. With revised centiles, the TW3 method estimates slightly younger ages than the TW2 method except at age group of 10- to 12-year where TW3 reference values are about a year ahead of TW2.[7,9,16,18]

Comparison between G&P and TW methods

Although both BA standards are good and comparable, G&P consistently scores children younger than the TW2 method, while TW3 age estimates are younger than TW2.[5,19,20] From age 6 onward, the average G&P child was about nine months ahead of the average British child in the 1950s as defined by the TW2 norms. Thus, on average, children rated on the G&P system will be about nine months less in BA than those rated on the TW2 system. TW3 standards terminate at 16.5 years for boys and 15 for girls, while G&P extends to 19 and 18 years, respectively. The TW method is more complex and time-consuming but offers greater subjectivity and reproducibility than G&P.[9,21] The average time spent on assessments was 1.4 minutes for G&P and 7.9 minutes for the TW2 method.[7,9,12,22]

Process of bone age estimation

G&P atlas

BAA is done by comparing the X-ray film with standard plates given in the atlas; if it matches the given plate, age of that plate (as mentioned in the atlas) is regarded as the age of the child. It begins with a preliminary selection of one or two probable plates and is then followed by detailed comparison of individual bones and epiphyses for final decision.[3]

In prepubertal children, greater emphasis is on the size and shape of epiphyses in the distal phalanges, followed by middle and proximal phalanges. If distal phalanges appear similar between different ages, then the assessment is focused on changes in middle followed by proximal phalanges.[8,23]

Once one or two plates are selected, a detailed comparison of individual bones and epiphyses is required.[8,23] If BA appears to be less or more than the given plate, adjacent plates should be used for a detailed study. However, in cases where a patient’s radiograph falls between two plates, the estimated skeletal age assigned is midway between the ages shown by these two standards. For example, if BA falls between 9 and 10 years, the BA assigned may be 9 years and 6 months.

The Tanner–Whitehouse (TW) method

BA is assessed by successive changes occurring in shape and density markings of each bone, starting from first appearance till its adult form, but the absolute size of the bone is ignored.[4] Maturation of each bone from first appearance to final adult form was divided into different stages (marked as A, B, C, D, etc.). The radius, metacarpals, phalanges, hamate and trapezium each have nine stages, while Ulna and the rest of the carpal bones have eight stages. Stage A represents absence of visible bone, while the last stage (H or I) represents the adult stage. The same stages are used for all ethnicities, including boys and girls. Each stage is given a fixed numerical score (between 0 and 1000) with an attempt to minimize overall disagreement between the different bones of individuals. A total score (sum of scores of all bones, SMS) is matched with the provided centile charts or with the given score in a table to assign the BA. The scores of boys and girls differ systematically, so they have been analysed separately. For boys, BA can be assessed between 2 and 16.5 years, while for girls, it is between 2 and 15 years. A score of 1000 suggests complete skeletal maturity.[4,7,17]

Limitations of BA assessment methods

Both G&P and TW methods are based on white populations and show less variation in them, but accuracy decreases in non-white groups, especially African-American and Asian-American children.[9] These methods are less reliable at age extremes with higher standard deviation. No particular age can be assigned once bones are fully matured. The relationship between CA and skeletal maturity is not constant from one population to another or from one time to another, even in a single population. Differences between RUS and Carpal BAs may be of diagnostic significance. For example, CKD patients showed that carpals were significantly more delayed than RUSA[TW3]. RUS-Carpal difference is also important in the context that changes of a secular trend that are universally seen in the RUS bones are not seen in carpals. The carpal bones mature by intramembranous ossification and are not much dependent on growth hormone, making them less suitable for BA.[12]

Both G&P and TW methods show observer-related variability due to subjectivity.[5,9] One study reported average intra-observer variation of 0.74 years for TW2 and 0.96 years for G&P.[22] Another study found a 95% confidence interval between two methods ranged from 2.28 to 1.52 years, with greater intra-observer variation for G&P compared to TW2.[24]

Difference between the left and right hand

Both systems, G&P and TW, use left hand radiographs over right hand for BA assessment[21] because of several reasons.

• •As most people are right-handed, there are more chances of injury to the right hand, which may affect the appearance and maturation of bones.

• •In the earlier two conferences of Physical Anthropologists (Monaco 1906 and Geneva 1912), it was decided to use the left side for physical measurements rather than the right side of the body and the same applied for BA assessment; hence, it has been strictly followed.

• •All the methods of BA assessment used the left hand; we do not have much choice.

• •By the same logic, the right hand may be used for left-handed people, but there is no clear agreement, and none of the methods uses the right hand. So, even in left-handed people, the left hand is being used for BA.

Published studies[25] show minimal developmental differences between left and right hands. Even if the difference is there, it’s not more than six months, which may not be significant. Since all BA references use the left hand, it is preferred. If the right-hand image is taken by mistake, it can be used for clinical practice without much concern.

Other conventional methods of BA assessment are given in the Supplementary Appendix Table S1.

Table S1.

Other Bone Age Methods

S. No	Type	Details
A	The Fels method	An objective assessment method, developed by Roche et al. and is based on North American children from the Fels longitudinal study (13,823 serial scans of 355 boys and 322 girls born between 1928 and 1974 in the age group of 1 month to 22 years). The relative maturation levels measured and then score is entered into a computer program, which calculates the SA.[5,9,10]
B	The Gilsanz and Ratib (GR) Atlas	Developed by Vicente Gilsanz and Osman Ratib in 2005 (last update—2012), based on shape, size, morphology and density of ossification centres in hand radiographs of healthy children, spaced at 6-month intervals in the age group of 2–6 years, and at yearly intervals from 7 to 17 years. It has been observed that GP and GR atlas appear to give similar estimation of bone age.[10,23]
C	Bone age by visualization of dental maturity	Commonly used for forensic purposes with limited use in endocrinology.[1,8]
D	Grave and Brown method	Developed in 1976, based on hand and wrist radiographs.[26]

Computerized methods

Computerized automated systems are becoming popular because of their user-friendliness and less variability. There are many computerized methods available [Supplementary Appendix Table S2]; however, we describe the most commonly used method here.[10]

Table S2.

Other Computerized Methods

S. No	System	Description
I	HANDX System (1989)	Micheal and Nelson created the first semi-automated BAA system using the TW database. It segments bones in hand wrist X-rays automatically through preprocessing, segmentation and measurement.[2,27]
II	PROI-Based System (1991)	Pietka and his team introduced a method based on PROI analysis, focusing on phalanges and epiphyses. Preprocessing identifies epiphyseal/metaphyseal regions (EMROIs), followed by feature extraction. These features offer an objective description of skeletal development. The method hasn’t been evaluated on a large scale.[2,28]
III	The CASAS System (1994):	The CASAS (Computer-based skeletal age scoring system) system, introduced by Tanner and Gibbons using the TW2 method with RUS bones, is a semi-automated approach. It digitized X-rays via a light box and monochrome video camera. Though initially found more reliable than manual methods, its need for manual bone identification limited its popularity and use.[2,29]
IV	Neural Network System Based on Linear Distance Measure (1995)	Gross proposed a system where users measure features from hand-wrist X-rays for bone age assessment using a neural network. Its limitation is the lack of morphological features used in GP or TW methods.[2,30]
V	Phalanges Length Based System (1990)	This was the first fully automated system using PACS (Picture Archiving and Communication System), relying on phalangeal length as a marker of skeletal maturity, with data from Garn. However, since phalangeal length is an unreliable indicator, the system saw limited use.
		Other methods include:
		• Mahmoodi Model
		• Middle Phalanx of the Third Finger (Active Shape Model)
		• Neural Network Classifiers using RUS and Carpal Bone Features
		• Neural Network Based on Radius and Ulna
		• CASMAS (Computer-Aided Skeletal Maturity Assessment System)[2,31]
VI	Automated Web Based System Using Histogram (2012)	Mansourvar et al.[2,32] proposed a fully automated BAA system using histogram-based compression and a content-based image retrieval (CBIR) approach. It relies on a database of 1100 hand X-rays, classified by gender and ethnicity. While it avoids segmentation issues using histograms, it performs poorly with abnormal bone structures or low-quality images.

BoneXpert System: This is a new, computerized, 100% automated approach based on the TW3 RUS standards.[33,34] The architecture of BoneXpert divides the processing into three layers. Layer A reconstructs the bone borders of 15 bones, layer B computes an intrinsic BA value for each of 13 bones (radius, ulna and 11 short bones in fingers 1, 3 and 5), while layer C transforms the intrinsic BA values to either TW BA or G&P BA using relatively simple post-processing.[5,35,36]

BoneXpert standards are based on 1559 images and validated on the G&P atlas in the age range 2–17 years (SD: 0.42 years; CI: 0.37;0.47) and on TW-rated images (SD: 0.80 years; CI: 0.68;0.93). Abnormal bone morphology or very poor image quality was automatically rejected by the bone reconstruction method. Its main advantage is high precision (0.17 yrs) and proven utility across ethnic groups.[37]

BoneXpert version 3: This update was released in September 2019 and represents the most significant update since BoneXpert launched ten years ago. This receives children’s hand X-rays in DICOM format and returns an annotated result picture with BA (GP and TW), Carpal BA (v3.x) and Bone Health Index (BHI) in under 15 seconds. It has included the addition of ethnicities and model improvements, all validated in independent studies and peer-reviewed articles.

1. Version 3 has some advantages over version 2. The accuracy of the BA measurement is dramatically improved. The root mean square error (RMSE) of BoneXpert relative to manual BA rating has decreased from 0.72 y in version 2 to 0.63 y in version 3, and when compared to the average of six manual ratings, the RMSE is still lower, that is, 0.45 y. BoneXpert has the same accuracy as the average of four manual ratings.

2. Carpal BA determination: Carpals were ignored in Version 2 because there is a consensus in some countries that carpals are less important. Version 3 now analyses the carpals, so it also provides carpal BA as a separate number.

3. BA and BHI determination down to age zero: Version 2 analysis of hands limited to BA more than approximately 2 years. Version 3 now extends down to newborns. The model is based on the study of 425 normal children from Paris, who were followed longitudinally with hand X-rays at the ages 1, 3, 6, 9, 12, 18, 24, 36 and 48 months.

4. Fewer rejections: Version 2 rejected some images due to postprocessing (edge enhancement), which is common in modern digital detectors, while version 3 is much better adapted to these variations and rejects far fewer images. Version 3 also locates the bones more accurately than version 2 (SD = 0.1 mm).

USG-based methods

The use of ultrasonographic devices is based on the principle that the speed of sound of ultrasonic waves propagating through cartilage and bone is different; the width of the cartilage or growth plate is measured relative to the width of the epiphysis of distal radius cartilage and used to predict maturation of epiphysis, thereby calculating BA. This method uses two transducers; one produces ultrasonic waves (750 kHz) whereas the other is a receiver. The studies show wide interobserver variability and low reliability due to operator dependence.[5,15]

MRI-based methods

The advantage of MRI-based methods is the ability to differentiate between soft tissue and bone. By measuring the characteristics of cartilage and bone, BA can be predicted. However, the cost and time of investigations make BA determination an unpopular choice.[15]

Artificial Intelligence (AI) in bone age assessment

Recent advances in BAA have incorporated AI-based systems, including automated tools and deep learning models, to improve accuracy and reproducibility. Systems such as BoneXpert demonstrate high precision, with mean absolute errors (MAE) of ~0.5–0.7 years and strong agreement with expert readers. However, performance varies across platforms.[38] In a recent study using BoneView, AI showed accuracy comparable to the Greulich–Pyle (GP) method (MAE 1.38 vs 1.30 years) and outperformed Tanner–Whitehouse 2 (TW2) (MAE 2.86 years), with minimal bias and good correlation (r = 0.857).[39] Despite improving consistency and reducing observer variability, AI does not consistently outperform expert assessment. Limitations include dependence on training datasets, limited generalizability across populations and reduced reliability in abnormal skeletal morphology or poor-quality imaging. Thus, AI serves as a valuable adjunct that enhances standardization and efficiency but should complement, not replace, clinical interpretation.

CONCLUSION

Skeletal maturation assessment efficiently guides diagnosis and prognosis while minimizing unnecessary tests. G&P and TW methods remain the cornerstone in endocrine practice, with newer technologies serving as adjuncts to enhance efficiency.

Author contributions

RK: Conceptualization, Writing – Original Draft, Writing – Review and Editing.

KKG: Writing – Original Draft, Writing – Review and Editing.

EN: Writing – Review and Editing.

All authors contributed equally to the refinement and final approval of the manuscript.

Conflicts of interest

There are no conflicts of interest.

Use of artificial intelligence

No.

Funding Statement

Nil.