The uncovered biases and errors in clinical determination of bone age by using deep learning models.

OBJECTIVES: To evaluate AI biases and errors in estimating bone age (BA) by comparing AI and radiologists' clinical determinations of BA. METHODS: We established three deep learning models from a Chinese private dataset (CHNm), an American public dataset (USAm), and a joint dataset combining the above two datasets (JOIm). The test data CHNt (n = 1246) were labeled by ten senior pediatric radiologists. The effects of data site differences, interpretation bias, and interobserver variability on BA assessment were evaluated. The differences between the AI models' and radiologists' clinical determinations of BA (normal, advanced, and delayed BA groups by using the Brush data) were evaluated by the chi-square test and Kappa values. The heatmaps of CHNm-CHNt were generated by using Grad-CAM. RESULTS: We obtained an MAD value of 0.42 years on CHNm-CHNt; this result indicated an appropriate accuracy for the whole group but did not indicate an accurate estimation of individual BA because with a kappa value of 0.714, the agreement between AI and human clinical determinations of BA was significantly different. The features of the heatmaps were not fully consistent with the human vision on the X-ray films. Variable performance in BA estimation by different AI models and the disagreement between AI and radiologists' clinical determinations of BA may be caused by data biases, including patients' sex and age, institutions, and radiologists. CONCLUSIONS: The deep learning models outperform external validation in predicting BA on both internal and joint datasets. However, the biases and errors in the models' clinical determinations of child development should be carefully considered. KEY POINTS: • With a kappa value of 0.714, clinical determinations of bone age by using AI did not accord well with clinical determinations by radiologists. • Several biases, including patients' sex and age, institutions, and radiologists, may cause variable performance by AI bone age models and disagreement between AI and radiologists' clinical determinations of bone age. • AI heatmaps of bone age were not fully consistent with human vision on X-ray films.

Sagittal suture maturation: Morphological reorganization, relation to aging, and reliability as an age-at-death indicator.

OBJECTIVES: The sagittal suture (SS) is assumed to be an initial site for the commencement of cranial suture closure as well as the most frequent spot of isolated craniosynostosis. The present study aimed to inspect the reorganization of the SS at the microlevel to assess the relation between its closure and aging and to establish whether it could be used as a reliable indicator in age-at-death prediction. MATERIALS AND METHODS: The SS was investigated in 68 dry contemporary adult male skulls of known age-at-death. An additional series of 20 skulls was used for verification. The skulls were scanned using a micro-computed tomography system. The SS closure degree was assessed along the three bone layers on cross-sectional tomograms by using a scoring scale. RESULTS: In the entirely open SS, the bone edges consist of compact bone and are widely separated. With SS maturation, the bone edges come into contact, and the remodeling process leads to a decrease in the sutural area and bone homogenization across all three layers. SS closure is an irregular process roughly related to aging, beginning in the early 20s, reaching its peak at about 30 years of age and abating in the late 40s. DISCUSSION: Although related to aging, SS closure is not a simple function of it. Rather, the underlying factors inducing and managing this process are multifaceted and complex. Although the etiology of SS maturation remains unclear, it is reasonable to use SS closure cautiously and only as a supportive method for age prediction.

Testing inter-observer reliability of the Transition Analysis aging method on the William M. Bass forensic skeletal collection.

OBJECTIVES: Skeletal age estimation is an integral part of the biological profile. Recent work shows how multiple-trait approaches better capture senescence as it occurs at different rates among individuals. Furthermore, a Bayesian statistical framework of analysis provides more useful age estimates. The component-scoring method of Transition Analysis (TA) may resolve many of the functional and statistical limitations of traditional phase-aging methods and is applicable to both paleodemography and forensic casework. The present study contributes to TA-research by validating TA for multiple, differently experienced observers using a collection of modern forensic skeletal cases. MATERIALS AND METHODS: Five researchers independently applied TA to a random sample of 58 documented individuals from the William M. Bass Forensic Skeletal Collection, for whom knowledge of chronological age was withheld. Resulting scores were input into the ADBOU software and maximum likelihood estimates (MLEs) and 95% confidence intervals (CIs) were produced using the forensic prior. Krippendorff's alpha was used to evaluate interrater reliability and agreement. Inaccuracy and bias were measured to gauge the magnitude and direction of difference between estimated ages and chronological ages among the five observers. RESULTS: The majority of traits had moderate to excellent agreement among observers (≥0.6). The superior surface morphology had the least congruence (0.4), while the ventral symphyseal margin had the most (0.9) among scores. Inaccuracy was the lowest for individuals younger than 30 and the greatest for individuals over 60. Consistent over-estimation of individuals younger than 30 and under-estimation of individuals over 40 years old occurred. Individuals in their 30s showed a mixed pattern of under- and over-estimation among observers. DISCUSSION: These results support the use of the TA method by researchers of varying experience levels. Further, they validate its use on forensic cases, given the low error overall.

Early Maturity as the New Normal: A Century-long Study of Bone Age.

BACKGROUND: Epiphyseal fusion (EF) marks the completion of longitudinal bone growth, a critical milestone monitored during treatment of skeletal growth and/or developmental disorders. Recently, a trend toward accelerated skeletal maturation in children has been documented. Because current methods for assessing skeletal maturation include children in their reference populations born as early as the 1930s, the timing of EF events in contemporary patients may differ substantially from those standards. QUESTIONS/PURPOSES: (1) Do children today initiate the process of EF in the hand and wrist earlier than past generations on which maturity standards are based? (2) Do children today complete EF in the hand and wrist earlier than past generations on which maturity standards are based? METHODS: A total of 1292 children (665 males, 627 females) participating in the Fels Longitudinal Study, born between 1915 and 2006, were included in this retrospective, observational study. Each participant had between one and 39 serial left hand-wrist radiographs during childhood obtained specifically for research purposes. Main outcomes were the chronological age at the first sign of EF initiation (EF-I) and the first chronological age when EF was complete (EF-C) in the radius and ulna, and metacarpals and phalanges of the first, third, and fifth rays according to criteria of the Fels method. EF is a reliable metric with an average κ agreement statistic of 0.91. Penalized B-splines were used to model the changes in EF-I and EF-C ages and to identify changes across continuous birth years with major comparisons between children born in 1935 and 1995. RESULTS: Approximately half of the epiphyses of the hand and wrist examined exhibited earlier EF-I and/or earlier EF-C in children born in 1995 compared with those born in 1935. The age at each milestone (EF-I and EF-C) decreased by as much as 6.7 and 6.8 months in males and 9.8 and 9.7 months in females, respectively. This change occurred gradually over the past century. The more proximal traits (EF of the distal radius, distal ulna, and metacarpals) were more likely to experience a shift in timing, whereas timing of EF in the phalanges remained relatively stable across birth years. CONCLUSIONS: A trend has occurred over the past century in the timing of EF, in both initiation and completion of the process, for many of the bones of the hand and wrist. Earlier EF reflects modern population advances in both skeletal and sexual maturation. Shifts in the timing of EF have the potential to influence treatment strategies for skeletal growth and/or developmental disorders such as scoliosis or leg length inequality, moving treatment windows to earlier ages. Earlier EF-I and EF-C identified in this study signals a need to reevaluate the timing of maturational milestones and current standards for skeletal assessment. LEVEL OF EVIDENCE: Level II, prognostic study.

Ossification of the Medial Clavicular Epiphysis on Chest Radiographs: Utility and Diagnostic Accuracy in Identifying Korean Adolescents and Young Adults under the Age of Majority.

The aim of our study was to evaluate the utility and diagnostic accuracy of the ossification grade of medial clavicular epiphysis on chest radiographs for identifying Korean adolescents and young adults under the age of majority. Overall, 1,151 patients (age, 16-30) without any systemic disease and who underwent chest radiography were included for ossification grading. Two radiologists independently classified the ossification of the medial clavicular epiphysis from chest radiographs into five grades. The age distribution and inter-observer agreement on the ossification grade were assessed. The diagnostic accuracy of the averaged ossification grades for determining whether the patient is under the age of majority was analyzed by using receiver operating characteristic (ROC) curves. Two separate inexperienced radiologists assessed the ossification grade in a subgroup of the patients after reviewing the detailed descriptions and image atlases developed for ossification grading. The median value of the ossification grades increased with increasing age (from 16 to 30 years), and the trend was best fitted by a quadratic function (R-square, 0.978). The inter-observer agreements on the ossification grade were 0.420 (right) and 0.404 (left). The area under the ROC curve (AUC) was 0.922 (95% CI, 0.902-0.942). The averaged ossification scores of 2.62 and 4.37 provided 95% specificity for a person < 19 years of age and a person ≥ 19 years of age, respectively. A preliminary assessment by inexperienced radiologists resulted in an AUC of 0.860 (95% CI, 0.740-0.981). The age of majority in Korean adolescents and young adults can be estimated using chest radiographs.

The reliability of clinical decisions based on the cervical vertebrae maturation staging method.

OBJECTIVES: Of the various techniques used to determine the optimum timing for growth modification treatments, the cervical vertebrae maturation method has great advantages, including validity and no need for extra X-ray exposure. Recently, the reproducibility of this method has been questioned. The aim of this study was to investigate the cause of poor reproducibility of this method and to assess the reproducibility of the clinical decisions made based on it. MATERIALS/METHODS: Seventy lateral cephalograms of Iranian patients aged 9‒15 years were observed twice by five experienced orthodontists. In addition to determining the developmental stage, each single parameter involved in this method was assessed in terms of inter- and intra-observer reproducibility. In order to evaluate the reproducibility of clinical decisions based on this method, cervical vertebrae maturation staging (CVMS) I and II were considered as phase 1 and CVMS IV and V were considered as phase 3. RESULTS: By considering the clinical approach of the CVMS method, inter-observer reproducibility of this method increased from 0.48 to 0.61 (moderate to substantial) and intra-observer reproducibility enhanced from 0.72 to 0.74. LIMITATIONS: 1. Complete visualization of the first four cervical vertebrae was an inclusion criterion, which also limits the clinical application of CVMS method. 2. These results can be generalized when determining growth modification treatments solely for Class II patients. CONCLUSIONS: Difficulty in determining the morphology of C3 and C4 leads to poor reproducibility of the CVMS method. Despite this, it has acceptable reproducibility in determining the timing of functional treatment for Class II patients.

The influence of body size on adult skeletal age estimation methods.

Accurate age estimations are essential to archaeological and forensic analyses. However, reliability for adult skeletal age estimations is poor, especially for individuals over the age of 40 years. This is the first study to show that body size influences skeletal age estimation. The Iscan et al., Lovejoy et al., Buckberry and Chamberlain, and Suchey-Brooks age methods were tested on 764 adult skeletons from the Hamann-Todd and William Bass Collections. Statures ranged from 1.30 to 1.93 m and body masses ranged from 24.0 to 99.8 kg. Transition analysis was used to evaluate the differences in the age estimations. For all four methods, the smallest individuals have the lowest ages at transition and the largest individuals have the highest ages at transition. Short and light individuals are consistently underaged, while tall and heavy individuals are consistently overaged. When femoral length and femoral head diameter are compared with the log-age model, results show the same trend as the known stature and body mass measurements. The skeletal remains of underweight individuals have fewer age markers while those of obese individuals have increased surface degeneration and osteophytic lipping. Tissue type and mechanical loading have been shown to affect bone turnover rates, and may explain the differing patterns of skeletal aging. From an archaeological perspective, the underaging of light, short individuals suggests the need to revisit the current research consensus on the young mortality rates of past populations. From a forensic perspective, understanding the influence of body size will impact efforts to identify victims of mass disasters, genocides, and homicides.

Effect of obesity on the reliability of age-at-death indicators of the pelvis.

During medicolegal investigations, forensic anthropologists commonly use morphological changes in the auricular surface of the ilium and the symphyseal face of the pubis to estimate age. However, obesity may impact the reliability of age estimations based on pelvic joints. Over the past several decades, the prevalence of obesity has dramatically increased in the United States (US). Since the rate of progression through age-related stages of weight-bearing joints may be influenced by excessive body mass, it is important that anthropologists understand how obesity affects age-related morphological changes in the skeleton. This study investigates the effects of obesity on the validity of the estimated age-at-death based on the Buckberry-Chamberlin and Suchey-Brooks methods by comparing US adults considered normal BMI (BMI 18.5-24.9) and obese (BMI ≥ 30). The obese group exhibits overall greater bias (overestimation of age) and inaccuracy, less precision, and lower correlations between estimated and known age than the normal BMI group using both methods, although differences in the pubic symphysis are not statistically significant. For the auricular surface the age of transition from one phase to the next is lower and the standard deviations are greater for the obese as compared to normal weight individuals. This study helps to elucidate how obesity affects the rate of age-related skeletal change of the human pelvis, and shows that the pubic symphysis may be a more reliable indicator of age in obese individuals and that greater standard deviations are needed for obese individuals when estimating age-at-death from the pelvis.

Bone Age Estimation for Whom by Whom? By Which Method?

Bone (skeletal) age determination is the simplest and most used index for the assessment of developmental and physiological age in healthy children and those with growth disorders. At present the test is done by manual or automated reading of the hand and wrist X-rays, necessitating two visits by the child: to the pediatrician and radiology departments. A newly developed simple quantitative ultrasound technique (QUST) using several hand and wrist bones, which can be performed in the pediatrician's office could combine the child's growth and biological age evaluation in one visit.

Dual-energy X-ray absorptiometry interpretation and reporting in children and adolescents: the revised 2013 ISCD Pediatric Official Positions.

The International Society for Clinical Densitometry Official Revised Positions on reporting of densitometry results in children represent current expert recommendations to assist health care providers determine which skeletal sites should be measured, which, if any, adjustments should be made, reference databases to be used, and the elements to include in a dual-energy X-ray absorptiometry report. The recommended scanning sites remain the total body less head and the posterior-anterior spine. Other sites such as the proximal femur, lateral distal femur, lateral vertebral assessment, and forearm are discussed but are only recommended for specific pediatric populations. Different methods of interpreting bone density scans in children with short stature or growth delay are presented. The use of bone mineral apparent density and height-adjusted Z-scores are recommended as suitable size adjustment techniques. The validity of appropriate reference databases and technical considerations to consider when upgrading software and hardware remain unchanged. Updated reference data sets for all contemporary bone densitometers are listed. The inclusion of relevant demographic and health information, technical details of the scan, Z-scores, and the wording "low bone mass or bone density" for Z-scores less than or equal to -2.0 standard deviation are still recommended for clinical practice. The rationale and evidence for the development of the Official Positions are provided. Changes in the grading of quality of evidence, strength of recommendation, and worldwide applicability represent a change in current evidence and/or differences in opinion of the expert panelists used to validate the position statements for the 2013 Position Development Conference.

Are the new automated methods for bone age estimation advantageous over the manual approaches?

Bone Age Assessment (BAA) is performed worldwide for the evaluation of endocrine, genetic and chronic diseases, to monitor response to medical therapy and to determine the growth potential of children and adolescents. It is also used for consultation in planning orthopedic procedures, for determination of chronological age for adopted children, youth sports participation and in forensic settings. The main clinical methods for skeletal bone age estimation are the Greulich and Pyle (GP) and the Tanner and Whitehouse (TW) methods. Seventy six per cent (76%) of radiologists or pediatricians usually use the method of GP, 20% that of TW and 4% other methods. The advantages of using the TW method, as opposed to the GP method, are that it overcomes the subjectivity problem and results are more reproducible. However, it is complex and time consuming; for this reason its usage is just about 20% on a world-wide scale. Moreover, there are some evidences that bone age assignments by different physicians can differ significantly. Computerized and Quantitative Ultrasound Technologies (QUS) for assessing skeletal maturity have been developed with the aim of reducing many of the inconsistencies associated with radiographic investigations. In spite of the fact that the volume of automated methods for BAA has increased, the majotity of them are still in an early phase of development. QUS is comparable to the GP based method, but there is not enough established data yet for the healthy population. The Authors wish to stimulate the attention on the accuracy, reliability and consistency of BAA and to initiate a debate on manual versus automated approaches to enhance our assessment for skeletal matutation in children and adolescents.

Technical note: false catastrophic age-at-death profiles in commingled bone deposits.

Age-at-death profiles obtained using the minimum number of individuals (MNI) for mass deposits of commingled human remains may be biased by over-representation of subadult individuals. A computer simulation designed in the R environment has shown that this effect may lead to misinterpretation of such samples even in cases where the completeness rate is relatively high. The simulation demonstrates that the use of the Most Likely Number of Individuals (MLNI) substantially reduces this bias.

Radiographic measurements of hip dysplasia at skeletal maturity--new reference intervals based on 2,038 19-year-old Norwegians.

OBJECTIVE: Normative references for radiographic measurements commonly used in the diagnosis of developmental dysplasia of the hip at skeletal maturity are incomplete. The present study therefore aimed to establish new gender-specific standards for measurements reflecting the acetabular morphology, namely Sharp's angle, the acetabular roof angle of Tönnis (AA) and the acetabular depth-width ratio (ADR), and measurements reflecting the position of the femoral head related to the acetabulum, namely the center-edge (CE) angle of Wiberg, the refined CE angle of Ogata, and the femoral head extrusion index (FHEI). The joint space width (JSW) is also reported. MATERIALS AND METHODS: The population-based 1989 Bergen Birth Cohort (n = 3,935) was invited at age 19 years to a follow-up during 2007-09, of which 2,038 (52 %) attended. A standardized antero-posterior radiograph was assessed. The normative references are presented as mean ± standard deviation (SD) and 2.5-97.5 percentiles with 95 % confidence intervals. RESULTS: A total of 2,011 (841 males, 1,170 females, mean age 18.6 (SD 0.6)) radiographs were analyzed. Sharp's angle was 38.8° ± 3.5° in males and 40.7° ± 3.5° in females, with 97.5 percentiles of 46° and 47°, respectively. The CE angle was 32.1° ± 6.1° in males and 31.0° ± 6.1° in females, with 2.5 percentiles of 21° and 20°, respectively. The FHEI was 86.0 % ± 6.3 % in males and 85.6 % ± 6.6 % in females, with 2.5 percentiles of 74° and 73°, respectively. CONCLUSIONS: Updated gender-specific reference ranges for radiographic measurements commonly used for hip dysplasia at skeletal maturity are reported, similar to or slightly wider than those described in the literature. Statistically significant gender differences have been confirmed for most of the measurements.

Test of a new components method for age-at-death estimation from the medial end of the fourth rib using a modern Spanish sample.

Recently, Verzeletti et al. (Am J Forensic Med Pathol 31:27-33, 2010) developed a new components method for the estimation of age-at-death from the sternal end of the fourth rib. This approach consists of the assignment of numerical scores for several age-related morphological characteristics, which are then inserted into a regression equation for predicting chronological age. The present study tested the accuracy of the new components method on a sample from Spain, consisting of 58 males and 36 females. The results demonstrated that the regression equation devised by the aforementioned authors for an Italian male sample was less accurate when applied to the Spanish population sample. This was true for separate male and female samples, as well as a combined-sex sample. A pair of population-specific regression formulae was then generated from the Spanish sample data in an attempt to improve the accuracy of the age estimate. The new equation which incorporated sex as an independent variable did not yield more accurate age estimates than a non-sex-specific equation. Furthermore, both of these formulae provided only marginally more accurate results for the Spanish sample compared to the original Italian equation. Thus, the standard errors of the estimate associated with these population-specific models (8.1-8.9 years) were still noticeably larger than the 5.2 years observed by Verzeletti and colleagues in their Italian study. Given the high standard errors associated with the age estimates for the Spanish population sample, the new rib components method should be used in conjunction with other adult aging techniques.

Brief communication: radiographic study of metatarsal one basal epiphyseal fusion: a note of caution on age determination.

This study examines radiographs of first metatarsals of 131 individuals from age 17-88 years to determine whether internal basal epiphyseal lines may be visible past the age of metatarsal fusion, which usually occurs between 14 and 16 years of age (Scheuer and Black: The juvenile skeleton. San Diego: Elsevier Academic Press,2004). In 29% (38 out of 131) of the radiographed first metatarsals (MT1s) the basal epiphyseal scar is visible, including in one individual who was 80 years old. Statistically, there was no relationship between the loss of the epiphyseal scar and age. Thus, the presence of the epiphyseal scar does not necessarily indicate subadult age. These data suggest that OH 8's radiographically visible basal epiphyseal line has no bearing on whether it is a subadult or not.

Transition analysis: a validation study with known-age modern American skeletons.

Transition Analysis-a recent skeletal age-estimation procedure (Boldsen et al.: Paleodemography: age distributions from skeletal samples (2002) 73-106)-is evaluated using 252 known-age modern American males and females from the Bass Donated Collection and Mercyhurst forensic cases. The pubic symphysis worked best for estimating age, followed by the sacroiliac joint and cranial sutures. Estimates based on all skeletal characteristics are influenced by the choice of prior distribution, although its effect is dwarfed by both the inaccuracy and imprecision of age estimates. Age intervals are narrowest for young adults, but are surprisingly short in old age as well. When using an informative prior distribution, the greatest uncertainty occurs from the late 40s into the 70s. Transition Analysis estimates do not perform as well as experience-based assessments, indicating the existing procedure is too narrowly focused on commonly used pelvic and cranial structures.

Bone age in the 21st century: is Greulich and Pyle's atlas accurate for Israeli children?

BACKGROUND: The applicability today of Greulich and Pyle's Radiographic Atlas of Skeletal Development of the Hand and Wrist (G&P) is uncertain. OBJECTIVE: The purpose of this study was to determine whether G&P is accurate in Israeli children today. MATERIALS AND METHODS: Left-hand radiographs of 679 children (375 boys) ranging in age from 1 day to 18 years old were obtained for trauma in the period 2001-2009 and were evaluated for bone age according to G&P. Individual bone age was plotted against calendar age and smoothed to obtain the association between calendar age and bone age. Any difference was assessed with Wilcoxon signed-rank test. RESULTS: In girls, there was no significant difference between bone age and calendar age (P = 0.188). G&P underestimated bone age in boys <15 years old (median difference, 2.3 months; P < 0.0001) and overestimated bone age in boys ≥15 years old (median difference, 2.9 months; P = 0.0043). The largest median difference (5.4 months; P = 0.0003) was seen in boys 6-10 years old. CONCLUSION: The differences between calendar age and bone age according to G&P were relatively small compared with normal variance and are unlikely to be of clinical importance.

Standardization of the Tanner-Whitehouse bone age method in the context of automated image analysis.

BACKGROUND/AIMS: The Tanner-Whitehouse (TW) method for bone age determination has been the basis for many population studies and it is used in many clinics. However, TW bone age raters can differ systematically from each other. The aim of the study was to present a new standard version of TW bone age rating implemented by the automated BoneXpert method and calibrated on the manual TW stage ratings of the First Zurich Longitudinal Study. SUBJECTS: Hand radiographs of 231 children born in 1954-1956 were recorded annually from an average age of 5-20 years. For validation, 76 X-rays of Tanner's original Gold Series from eight boys were used. RESULTS: The root mean square deviation between manual and automated TW ratings in the Zurich data was 0.67 years for boys in the TW bone age range 5-15 years and 0.63 years for girls, 5-14 years. The new standard TW rating differs systematically from two previous TW versions of the automated method, based on different raters. CONCLUSION: The new automated TW ratings show good accuracy relative to the manual ratings of the Zurich data and the Gold Series. There are significant differences between manual TW raters, an effect which is eliminated with the automated method.

Brief communication: the relation between standard error of the estimate and sample size of histomorphometric aging methods.

Histomorphometric aging methods report varying degrees of precision, measured through Standard Error of the Estimate (SEE). These techniques have been developed from variable samples sizes (n) and the impact of n on reported aging precision has not been rigorously examined in the anthropological literature. This brief communication explores the relation between n and SEE through a review of the literature (abstracts, articles, book chapters, theses, and dissertations), predictions based upon sampling theory and a simulation. Published SEE values for age prediction, derived from 40 studies, range from 1.51 to 16.48 years (mean 8.63; sd: 3.81 years). In general, these values are widely distributed for smaller samples and the distribution narrows as n increases--a pattern expected from sampling theory. For the two studies that have samples in excess of 200 individuals, the SEE values are very similar (10.08 and 11.10 years) with a mean of 10.59 years. Assuming this mean value is a 'true' characterization of the error at the population level, the 95% confidence intervals for SEE values from samples of 10, 50, and 150 individuals are on the order of ± 4.2, 1.7, and 1.0 years, respectively. While numerous sources of variation potentially affect the precision of different methods, the impact of sample size cannot be overlooked. The uncertainty associated with SEE values derived from smaller samples complicates the comparison of approaches based upon different methodology and/or skeletal elements. Meaningful comparisons require larger samples than have frequently been used and should ideally be based upon standardized samples.