Bone Age Assessment Method Research Articles

Leveraging bone age assessment via a novel joint decomposition teacher–student learning paradigm from X-ray images

ABSTRACT Advancements in technology has accelerated the evolution of bone age assessment (BAA) methodologies, one of which is deep learning algorithms, which overcome the drawbacks of conventional approaches. In spite of excellent effectiveness of deep neural networks in detection of the correct class for bone age, they have a significant degree of complexity due to the numerous parameters they employ for each region of interest (ROI). In this paper, we propose a BAA method using a hybrid knowledge distillation (KD) paradigm in order to conquer this difficulty by mapping different ROIs into a single ROI. In this regard, the student receives knowledge from a teacher network that has been pre-trained on six ROIs including bones of five fingers and the wrist, transfers the knowledge of its final response layer and internal layers to the student. Then, six student models each of which is constructed based on just one of these ROIs, while receiving the information of the teacher model. Empirical results on digital hand atlas report that our student model trained on one ROI obtains 95% accuracy on 19 classes of bone age.

The automated Greulich and Pyle: a coming-of-age for segmental methods?

The well-known Greulich and Pyle (GP) method of bone age assessment (BAA) relies on comparing a hand X-ray against templates of discrete maturity classes collected in an atlas. Automated methods have recently shown great success with BAA, especially using deep learning. In this perspective, we first review the success and limitations of various automated BAA methods. We then offer a novel hypothesis: When networks predict bone age that is not aligned with a GP reference class, it is not simply statistical error (although there is that as well); they are picking up nuances in the hand X-ray that lie "outside that class." In other words, trained networks predict distributions around classes. This raises a natural question: How can we further understand the reasons for a prediction to deviate from the nominal class age? We claim that segmental aging, that is, ratings based on characteristic bone groups can be used to qualify predictions. This so-called segmental GP method has excellent properties: It can not only help identify differential maturity in the hand but also provide a systematic way to extend the use of the current GP atlas to various other populations.

Comparison of bone age assessment methods using a hand radiography in patients with active growth plate and anteromedial knee instability

Background Bone age is essential for pediatric patients with active growth zones and anteromedial instability to facilitate optimal treatment strategy and minimize postoperative complications. However, many people are unaware of various tools for determining bone age, including classical methods and modern machine learning techniques.The objective was to show and compare different methods for calculating bone age and determining surgical strategy for patients with anteromedial instability of the knee joint.Material and methods All-Inside anterior cruciate ligament reconstruction was performed for 20 patients. Wrist radiographs were performed for bone age assessment using the "point scoring system" of Tanner and Whitehouse and the "atlas matching" method of Greulich and Pyle. Machine learning programs were used in addition to standard bone age assessments.Results The findings showed an average difference of 21 months (80 %) in a group of 20 individuals with bone age ahead of the passport age and an average difference of 18 months (20 %) in patients with retarded bone age.Discussion The findings showed the difference between chronological and bone age and could be encountered in scientific articles on endocrinology and pediatrics. No scientific studies on the use of the methods could be found in the specialty “trauma and orthopaedics”.Conclusion Bone age assessment, prediction of children's target height are essential for surgical treatment of patients with open growth plates.

SVTNet: Automatic bone age assessment network based onTW3method and vision transformer

AbstractThis study aims to develop a proficient and clinically applicable algorithm that can accurately assess bone age. This algorithm is based on the principles of the Tanner‐Whitehouse 3 (TW3) integral approach, and aims to achieve efficiency, scalability, and interpretability. We developed a model for bone age prediction in children. The model was tested on a pediatric dataset from a tertiary care hospital consisting of left‐hand radiographs of children between the age of 0 and 18. Our model consists of removing the arm portion using a pre‐trained YOLO network, localizing 37 key points in the hand bone portion using a spatial configuration network, and segmenting the original image through 20 of these points to obtain 20 fixed‐size patches. Finally, each of the 20 bone images is classified by training a visual transformer (ViT) model. In this study, a hybrid network, SVTNet, was developed that incorporates visual transformers to obtain estimates of bone age in the carpal (C series) and metacarpal (RUS series) bones. The sum of the clinical TW3 scoring region scores and bone maturity scores were utilized to determine the bone age for each corresponding region. The performance of the algorithm was evaluated in terms of both training and testing by evaluating 3871 left hand X‐ray micrographs obtained from a tertiary hospital in China. The results showed that the average absolute error of bone age estimation was 0.50 years for the RUS series of bones and 0.47 years for the C series of bones. The main contribution of this study is to propose, for the first time, a ViT‐based bone age assessment method that automates the entire process of the TW3 algorithm and is clinically interpretable, with predictive accuracy comparable to that of an experienced orthopedic surgeon.

A pediatric bone age assessment method for hand bone X-ray images based on dual-path network

A pediatric bone age assessment method for hand bone X-ray images based on dual-path network

Research on an intelligent evaluation method of bone age based on multi-region combination

Bone age is one of the most important evaluation indexes for the growth and development of children and adolescents. The bone age assessment method based on deep learning generally uses the whole left wrist X-ray film or some regions of interest in the left wrist X-ray film. Based on the entire X-ray film, the intelligent evaluation process is simple, but the accuracy is low. Although intelligent evaluation based on regions of interest has high accuracy, it requires prior knowledge and the process is complex. To solve the above problems, this paper proposes a multi-region combined method for bone age assessment. A small number of regions of interest in wrist bone X-ray films are extracted, and then the whole X-ray film and these regions of interest were used to evaluate the bone age. The experiment uses the improved Inception-ResNet-V2 convolutional neural network. The results show that compared with other bone age assessment studies on the open data set published by the North American radiological Association, this method can obtain higher accuracy of bone age assessment, with an average absolute error of 7.11 months. This method improves the efficiency and accuracy of bone age assessment while simplifying the assessment process.

TW3-based ROI detection and classification using a chaotic ANN and DNN-EVGO architecture for an automated bone age assessment on hand X-ray images

ABSTRACT Bone illnesses arise at a young age itself, so bone age assessment (BAA) is primarily utilized in paediatrics to identify their growth. Several BAA-related methods are employed to determine bone maturity; however, they do not give accuracy and the rate of error increases. To overcome this issue, in this manuscript, TW3-based Region of Interest (ROI) identification and classification with Chaotic ANN (CANN), as well as Deep Neural Network optimized with Evolved Gradient Direction optimizer (EVGO) architecture, is proposed for an Automated BAA classification. In this, Hand X-Ray images are taken from the Digital Hand Atlas (DHA) dataset and the Radiological Society of North America (RSNA) database. Here, Tanner-Whitehouse (TW3) and Shallow convolution neural network (SCNN) is utilized for extracting the ROI regions such as radius, ulna, and short bones (RUS) from the wrist region of the left hand. After that, the extracted ROI feature data are given to CANN for predicting BAA. Then, the predicted data are classified using the DNN-EVGO network. The bone age evaluation is then classified depending on maturity phases. The proposed approach is activated in MATLAB. Simulation outcomes of TW3- SCNN- CANN- DNN-EVGO- BAA methods for DHA database attain higher accuracy of 45.75%, 37.64%, 24.64% when analysed to the existing models, such as Faster Region-Convolutional Neural network oriented feature learning with optimal trained Recurrent Neural Network for BAA for paediatrics (TW3-F-R-CNN-AF-SFO), Multi-objective segmentation approach for BAA under parameter tuning- U-net architecture, and TW3-based fully automated bone age assessment system using deep neural networks (TW3-F-R-CNN-VGGNet-CNN-BAA).

COMPARISON OF RADIOGRAPHIC METHODS (THE GILSANZ AND RATIBIN ATLAS AND TANNER-WHITEHOUSE 2 AND 3) OF BONE AGE ASSESSMENT WITH CHRONOLOGICAL AGE

Introduction: Bone age assessment (BAA) can play an important role in legal and medical contexts. It can help determine a person’s legal rights and responsibilities, and can also be used to estimate their chronological age (CA) when accurate birth records are not available. BAA is used in situations such as identification, employment, criminal responsibility, judicial punishment, consent for marriage, and in sports competitions. In addition, it can also be used in cases where there is a need to estimate CA, such as during immigration or in conditions where birth records are not available. Methods: The study was approved by the Institutional Ethical Committee, and written consent was obtained from both children and their parents before participating in the study. Patients were selected based on specific inclusion and exclusion criteria. A data collection form was created for each participant. X-rays of the left hand and wrist were taken for all children referred to the X-ray Department at Sahyadri Speciality Hospital in Pune. The CA of the child was determined by their date of birth, and the bone age (BA) was assessed using radiographic methods. Result: The Bland–Altman plot is a statistical tool used to compare the agreement between two methods of measurement, in this case, the CA and Tanner-Whitehouse 2 (TW3) methods for BAA. The results of the comparison showed a statistically significant agreement between the TW3 method and the GR atlas. Conclusion: The TW2 method is not reliable for BAA. However, the TW3 method is considered reliable for assessing BA in older male children (ages 105–192 months) and younger female children (ages 13–104 months). On the other hand, the GR atlas appears to be more effective for younger male children (ages 13–104 months) and older female children (ages 105–192 months). In addition, the TW3 method can be used in conjunction with the GR atlas for older female children.

Evaluation of two methods of bone age assessment in peripubertal children in Zimbabwe.

Bone age (BA) measurement in children is used to evaluate skeletal maturity and helps in the diagnosis of growth disorders in children. The two most used methods are Greulich and Pyle (GP), and Tanner and Whitehouse 3 (TW3), both based upon assessment of a hand-wrist radiograph. To our knowledge no study has compared and validated the two methods in sub-Saharan Africa (SSA), and only a few have determined BA despite it being a region where skeletal maturity is often impaired for example by HIV and malnutrition. This study aimed to compare BA as measured by two methods (GP and TW3) against chronological age (CA) and determine which method is most applicable in peripubertal children in Zimbabwe. We conducted a cross-sectional study of boys and girls who tested negative for HIV. Children and adolescents were recruited by stratified random sampling from six schools in Harare, Zimbabwe. Non-dominant hand-wrist radiographs were taken, and BA assessed manually using both GP and TW3. Paired sample Student t-tests were used to calculate the mean differences between BA and chronological age (CA) in boys and girls. Bland-Altman plots compared CA to BA as determined by both methods, and agreement between GP and TW3 BA. All radiographs were graded by a second radiographer and 20% of participants of each sex were randomly selected and re-graded by the first observer. Intraclass correlation coefficient assessed intra- and inter-rater reliability and coefficient of variation assessed precision. We recruited 252 children (111 [44%] girls) aged 8.0-16.5years. The boys and girls were of similar mean±SD CA (12.2±2.4 and 11.7±1.9years) and BA whether assessed by GP (11.5±2.8 and 11.5±2.1years) or TW3 (11.8±2.5 and 11.8±2.1years). In boys BA was lower than CA by 0.76years (95% CI: -0.95, -0.57) when using GP, and by 0.43years (95% CI: -0.61, -0.24) when using TW3. Among the girls there was no difference between BA and CA by either GP [-0.19years (95% CI: -0.40, 0.03)] or TW3 [0.07years (95% CI: -0.16, 0.29)]. In both boys and girls, there were no systematic differences between CA and TW3 BA across age groups whereas agreement improved between CA and GP BA as children got older. Inter-operator precision was 1.5% for TW3 and 3.7% for GP (n=252) and intra-operator precision was 1.5% for TW3 and 2.4% for GP (n=52). The TW3 BA method had better precision than GP and did not systematically differ from CA, meaning that TW3 is the preferred method of assessment of skeletal maturity in Zimbabwean children and adolescents. TW3 and GP methods do not agree for estimates of BA and therefore cannot be used interchangeably. The systematic differences in GP BA assessments over age means it is not appropriate for use in all age groups or stages of maturity in this population.

Bone age assessment based on deep neural networks with annotation-free cascaded critical bone region extraction.

Bone age assessment (BAA) from hand radiographs is crucial for diagnosing endocrinology disorders in adolescents and supplying therapeutic investigation. In practice, due to the conventional clinical assessment being a subjective estimation, the accuracy of BAA relies highly on the pediatrician's professionalism and experience. Recently, many deep learning methods have been proposed for the automatic estimation of bone age and had good results. However, these methods do not exploit sufficient discriminative information or require additional manual annotations of critical bone regions that are important biological identifiers in skeletal maturity, which may restrict the clinical application of these approaches. In this research, we propose a novel two-stage deep learning method for BAA without any manual region annotation, which consists of a cascaded critical bone region extraction network and a gender-assisted bone age estimation network. First, the cascaded critical bone region extraction network automatically and sequentially locates two discriminative bone regions via the visual heat maps. Second, in order to obtain an accurate BAA, the extracted critical bone regions are fed into the gender-assisted bone age estimation network. The results showed that the proposed method achieved a mean absolute error (MAE) of 5.45 months on the public dataset Radiological Society of North America (RSNA) and 3.34 months on our private dataset.

An Automated TW3-RUS Bone Age Assessment Method with Ordinal Regression-Based Determination of Skeletal Maturity.

The assessment of bone age is important for evaluating child development, optimizing the treatment for endocrine diseases, etc. And the well-known Tanner-Whitehouse (TW) clinical method improves the quantitative description of skeletal development based on setting up a series of distinguishable stages for each bone individually. However, the assessment is affected by rater variability, which makes the assessment result not reliable enough in clinical practice. The main goal of this work is to achieve a reliable and accurate skeletal maturity determination by proposing an automated bone age assessment method called PEARLS, which is based on the TW3-RUS system (analysis of the radius, ulna, phalanges, and metacarpal bones). The proposed method comprises the point estimation of anchor (PEA) module for accurately localizing specific bones, the ranking learning (RL) module for producing a continuous stage representation of each bone by encoding the ordinal relationship between stage labels into the learning process, and the scoring (S) module for outputting the bone age directly based on two standard transform curves. The development of each module in PEARLS is based on different datasets. Finally, corresponding results are presented to evaluate the system performance in localizing specific bones, determining the skeletal maturity stage, and assessing the bone age. The mean average precision of point estimation is 86.29%, the average stage determination precision is 97.33% overall bones, and the average bone age assessment accuracy is 96.8% within 1year for the female and male cohorts.

Bone age recognition based on mask R-CNN using xception regression model.

Background and Objective: Bone age detection plays an important role in medical care, sports, judicial expertise and other fields. Traditional bone age identification and detection is according to manual interpretation of X-ray images of hand bone by doctors. This method is subjective and requires experience, and has certain errors. Computer-aided detection can effectually enhance the validity of medical diagnosis, especially with the fast development of machine learning and neural network, the method of bone age recognition using machine learning has gradually become the focus of research, which has the advantages of simple data pretreatment, good robustness and high recognition accuracy. Methods: In this paper, the hand bone segmentation network based on Mask R-CNN was proposed to segment the hand bone area, and the segmented hand bone region was directly input into the regression network for bone age evaluation. The regression network is using an enhancd network Xception of InceptionV3. After the output of Xception, the convolutional block attention module is connected to refine the feature mapping from channel and space to obtain more effective features. Results: According to the experimental results, the hand bone segmentation network model based on Mask R-CNN can segment the hand bone region and eliminate the interference of redundant background information. The average Dice coefficient on the verification set is 0.976. The mean absolute error of predicting bone age on our data set was only 4.97 months, which exceeded the accuracy of most other bone age assessment methods. Conclusion: Experiments show that the accuracy of bone age assessment can be enhancd by using the Mask R-CNN-based hand bone segmentation network and the Xception bone age regression network to form a model, which can be well applied to actual clinical bone age assessment.

Human Bone Age Estimation of Carpal Bone X-Ray Using Residual Network with Batch Normalization Classification

Bone age is an index used by pediatric radiology and endocrinology departments worldwide to define skeletal maturity for medical and non-medical purposes. In general, the clinical method for bone age assessment (BAA) is based on examining the visual ossification of individual bones in the left hand and then comparing it with a standard radiographic atlas of the hand. However, this method is highly dependent on the experience and conditions of the forensic expert. This paper proposes a new approach to age estimation of human bone based on the carpal bones in the hand and using a residual network architecture. The classification layer was modified with batch normalization to optimize the training process. Before carrying out the training process, we performed an image augmentation technique to make the dataset more varied. The following augmentation techniques were used: resizing; random affine transformation; horizontal flipping; adjusting brightness, contrast, saturation, and hue; and image inversion. The output is the classification of bone age in the range of 1 to 19 years. The results obtained when using a VGG16 model were an MAE value of 5.19 and an R2 value of 0.56 while using the newly developed ResNeXt50(32x4d) model produced an MAE value of 4.75 and an R2 value of 0.63. The research results indicate that the proposed modification of the residual training model improved classification compared to using the VGG16 model, as indicated by an MAE value of 4.75 and an R2 value of 0.63.

Combined assisted bone age assessment and adult height prediction methods in Chinese girls with early puberty: analysis of three artificial intelligence systems

BackgroundThe applicability and accuracy of artificial intelligence (AI)-assisted bone age assessment and adult height prediction methods in girls with early puberty are unknown.ObjectiveTo analyze the performance of AI-assisted bone age assessment methods by comparing the corresponding methods for predicted adult height with actual adult height.Materials and methodsThis retrospective review included 726 girls with early puberty, 87 of whom had reached adult height at last follow-up. Bone age was evaluated using the Greulich–Pyle (GP), Tanner–Whitehouse (TW3–RUS) and China 05 RUS–CHN (RUS-CHN) methods. Predicted adult height was calculated using the China 05 (CH05), TW3 and Bayley–Pinneau (BP) methods.ResultsWe analyzed 1,663 left-hand radiographs, including 155 from girls who had reached adult height. In the 6–8- and 9–11-years age groups, bone age differences were smaller than those in the 12–14-years group; however, the differences between predicted adult height and actual adult height were larger than those in the 12–14-years group. TW3 overestimated adult height by 0.4±2.8 cm, while CH05 and BP significantly underestimated adult height by 2.9±3.6 cm and 1.3±3.8 cm, respectively. TW3 yielded the highest proportion of predicted adult height within ±5 cm of actual adult height (92.9%), with the highest correlation between predicted and actual adult heights.ConclusionThe differences in measured bone ages increased with increasing bone age. However, the corresponding method for predicting adult height was more accurate when the bone age was older. TW3 might be more suitable than CH05 and BP for predicting adult height in girls with early puberty. Methods for predicting adult height should be optimized for populations of the same ethnicity and disease.

Applying Convolutional Neural Network in Automatic Assessment of Bone Age Using Multi-Stage and Cross-Category Strategy

Bone age is a common indicator of children’s growth. However, traditional bone age assessment methods usually take a long time and are jeopardized by human error. To address the aforementioned problem, we propose an automatic bone age assessment system based on the convolutional neural network (CNN) framework. Generally, bone age assessment is utilized amongst 0–18-year-old children. In order to reduce its variation in terms of regression model building, our system consists of two steps. First, we build a maturity stage classifier to identify the maturity stage, and then build regression models for each maturity stage. In this way, assessing bone age through the use of several independent regression models will reduce the variation and make the assessment of bone age more accurate. Some bone sections are particularly useful for distinguishing certain maturity stages, but may not be effective for other stages, and thus we first perform a rough classification to generally distinguish the maturity stage, and then undertake fine classification. Because the skeleton is constantly growing during bone development, it is not easy to obtain a clear decision boundary between the various stages of maturation. Therefore, we propose a cross-stage class strategy for this problem. In addition, because fewer children undergo X-rays in the early and late stages, this causes an imbalance in the data. Under the cross-stage class strategy, this problem can also be alleviated. In our proposed framework, we utilize an MSCS-CNN (Multi-Step and Cross-Stage CNN). We experiment on our dataset, and the accuracy of the MSCS-CNN in identifying both female and male maturity stages is above 0.96. After determining maturity stage during bone age assessment, we obtain a 0.532 and 0.56 MAE (mean absolute error) for females and males, respectively.

A comparison of bone age assessments using automated and manual methods in children of Indian ethnicity.

Bone age is useful for pediatric endocrinologists in evaluating various disorders related to growth and puberty. Traditional methods of bone age assessment, namely Greulich and Pyle (GP) and Tanner-Whitehouse (TW), have intra- and interobserver variations. Use of computer-automated methods like BoneXpert might overcome these subjective variations. The aim of our study was to assess the validity of BoneXpert in comparison to manual GP and TW methods for assessing bone age in children of Asian Indian ethnicity. We extracted from a previous study the deidentified left hand radiographs of 920 healthy children aged 2-19 years. We compared bone age as determined by four well-trained manual raters using GP and TW methods with the BoneXpert ratings. We computed accuracy using root mean square error (RMSE) to assess how close the bone age estimated by BoneXpert was to the reference rating. The standard deviations (SDs) of rating among the four manual raters were 0.52 years, 0.52 years and 0.47 years for GP, TW2 and TW3 methods, respectively. The RMSEs between the automated bone age estimates and the true ratings were 0.39 years, 0.41 years and 0.36 years, respectively, for the same methods. The RMSE values were significantly lower in girls than in boys (0.53, 0.5 and 0.47 vs. 0.39, 0.47 and 0.4) by all the methods; however, no such difference was noted in classification by body mass index. The best agreement between BoneXpert and manual rating was obtained by using 50% weight on carpals (GP50). The carpal bone age was retarded in Indian children, more so in boys. BoneXpert was accurate and performed well in estimating bone age by both GP and TW methods in healthy Asian Indian children; the error was larger in boys. The GP50 establishes "backward compatibility" with manual rating.

A comparative study of three bone age assessment methods on Chinese preschool-aged children.

BackgroundBone age assessment (BAA) is an essential tool utilized in outpatient pediatric clinics. Three major BAA methods, Greulich–Pyle (GP), Tanner–Whitehouse 3 (TW3), and China 05 RUS–CHN (RUS–CHN), were applied to comprehensively compare bone age (BA) and chronological age (CA) in a Chinese sample of preschool children. This study was designed to determine the most reliable method.MethodsThe BAA sample consisted of 207 females and 183 males aged 3–6 years from the Zhejiang Province in China. The radiographs were estimated according to the GP, TW3, and RUS–CHN methods by two pediatric radiologists. The data was analyzed statistically using boxplots, the Wilcoxon rank test, and Student’s t-test to explore the difference (D) between BA and CA.ResultsAccording to the distributions of D, the boxplots showed that the median D of the TW3 method was close to zero for both male and female subjects. The TW3 and RUS–CHN methods overestimated the age of both genders. The TW3 method had the highest correct classification rate for males but a similar rate for females. The GP method did not show any significant difference between the BA and CA when applied to 3-year-old males and 4-year-old females while the TW3 method showed similar results when applied to 6-year-old females. The RUS–CHN method showed the least consistent results among the three methods.ConclusionThe TW3 method was superior to the GP and RUS–CHN methods but not reliable on its own. It should be noted that a precise age diagnosis for preschool children cannot be easily made if only one of the methods is utilized. Therefore, it is advantageous to combine multiple methods when assessing bone age.

Attention-based multiple-instance learning for Pediatric bone age assessment with efficient and interpretable

Pediatric bone age assessment (BAA) is a common clinical technique for evaluating children's endocrine, genetic, and growth disorders. However, the deep learning BAA method based on global images neglects fine-grained concerns, and regions of interest (ROIs) need additional annotation and complex processing. To overcome these shortcomings, we proposed an interpretable deep-learning architecture based on multiple-instance learning to address BAA efficiently without additional annotations. We cropped the entire image into small patches and got patch features by feature extraction network. Then, an attention backbone ranked feature vectors of the entire image and aggregates its information according to its relative importance. Finally, each image's features and gender were aggregated to predict bone age. The proposed method can identify ROIs by attention-based multi-instance aggregation without additional labels and produce interpretable heatmaps. Moreover, by cropping the complete image into patches and reducing the dimensionality, the proposed model can notice the fine-grained information of the image and improve the model training speed. We validated the proposed method in the Radiological Society of North America 2017 dataset. The results showed that the proposed model achieved an advanced performance of MAE 4.17 months. Furthermore, the visualization results indicated that the proposed model was highly interpretable, which can localize the ROIs without spatial labeling. In conclusion, a novel method for high performance and interpretable bone age prediction without additional manual annotations has been developed, which can be used to effectively assess the pediatric's bone age.

Fine-grained precise-bone age assessment by integrating prior knowledge and recursive feature pyramid network

Bone age assessment (BAA) evaluates individual skeletal maturity by comparing the characteristics of skeletal development to the standard in a specific population. The X-ray image examination for bone age is tedious and subjective, and it requires high professional skills. Therefore, AI techniques are desired to innovate and improve BAA methods. Most of the BAA method use the whole X-ray image in an end-to-end model directly. Such whole-image-based approaches fail to characterize local changes and provide limited aid for diagnosis and understanding disease progress. To address these issues, we collected and curated a dataset of 2129 cases for the study of BAA with fine-grained skeletal maturity level labels of the 13 ROIs in hand bone based on the expert knowledge from TW method. We designed a four-stage automatic BAA model based on recursive feature pyramid network. Firstly, the palm region was segmented using U-Net, followed by the extraction of multi-target ROIs of hand bone using a recursive feature pyramid network. Given the extracted ROIs, we employed a transfer learning model with attention mechanism to predict the skeletal maturity level of each ROI. Finally, the bone age is assessed based on the percentile curve of bone maturity. The proposed BAA model can automate the BAA. In addition, it provides the detection result of the 13 ROIs and their ROI-level skeletal maturity. The MAE can reach 0.61 years on the dataset with the labeling precision of one year. All the data and annotations used in this paper are released publicly.

SMANet: multi-region ensemble of convolutional neural network model for skeletal maturity assessment.

Bone age assessment (BAA) is a crucial research topic in pediatric radiology. Interest in the development of automated methods for BAA is increasing. The current BAA algorithms based on deep learning have displayed the following deficiencies: (I) most methods involve end-to-end prediction, lacking integration with clinically interpretable methods; (II) BAA methods exhibit racial and geographical differences. A novel, automatic skeletal maturity assessment (SMA) method with clinically interpretable methods was proposed based on a multi-region ensemble of convolutional neural networks (CNNs). This method predicted skeletal maturity scores and thus assessed bone age by utilizing left-hand radiographs and key regional patches of clinical concern. Experiments included 4,861 left-hand radiographs from the database of Beijing Jishuitan Hospital and revealed that the mean absolute error (MAE) was 31.4±0.19 points (skeletal maturity scores) and 0.45±0.13 years (bone age) for the carpal bones-series and 29.9±0.21 points and 0.43±0.17 years, respectively, for the radius, ulna, and short (RUS) bones series based on the Tanner-Whitehouse 3 (TW3) method. The proposed automatic SMA method, which was without racial and geographical influence, is a novel, automatic method for assessing childhood bone development by utilizing skeletal maturity. Furthermore, it provides a comparable performance to endocrinologists, with greater stability and efficiency.