The BoneXpert method for automated determination of skeletal maturity

Quantitative Imaging in Inflammatory Arthritis: Between Tradition and Innovation

Radiologic imaging is crucial for diagnosing and monitoring rheumatic inflammatory diseases. Particularly the emerging approach of precision medicine has increased the interest in quantitative imaging. Extensive research has shown that ultrasound allows a quantification of direct signs such as bone erosions and synovial thickness. Dual-energy X-ray absorptiometry and high-resolution peripheral quantitative computed tomography (CT) contribute to the quantitative assessment of secondary signs such as osteoporosis or lean mass loss. Magnetic resonance imaging (MRI), using different techniques and sequences, permits in-depth evaluations. For instance, the perfusion of the inflamed synovium can be quantified by dynamic contrast-enhanced imaging or diffusion-weighted imaging, and cartilage injury can be assessed by mapping (T1ρ, T2). Furthermore, the increased metabolic activity characterizing the inflammatory response can be reliably assessed by hybrid imaging (positron emission tomography [PET]/CT, PET/MRI). Finally, advances in intelligent systems are pushing forward quantitative imaging. Complex mathematical algorithms of lesions' segmentation and advanced pattern recognition are showing promising results.

Chronological Age Assessment in Young Individuals Using Bone Age Assessment Staging and Nonradiological Aspects: Machine Learning Multifactorial Approach

BACKGROUND

Bone age assessment (BAA) is used in numerous pediatric clinical settings as well as in legal settings when entities need an estimate of chronological age (CA) when valid documents are lacking. The latter case presents itself as critical as the law is harsher for adults and granted rights along with imputability changes drastically if the individual is a minor. Traditional BAA methods have drawbacks such as exposure of minors to radiation, they do not consider factors that might affect the bone age, and they mostly focus on a single region. Given the critical scenarios in which BAA can affect the lives of young individuals, it is important to focus on the drawbacks of the traditional methods and investigate the potential of estimating CA through BAA.

OBJECTIVE

This study aims to investigate CA estimation through BAA in young individuals aged 14-21 years with machine learning methods, addressing the drawbacks of research using magnetic resonance imaging (MRI), assessment of multiple regions of interest, and other factors that may affect the bone age.

METHODS

MRI examinations of the radius, distal tibia, proximal tibia, distal femur, and calcaneus were performed on 465 men and 473 women (aged 14-21 years). Measures of weight and height were taken from the subjects, and a questionnaire was given for additional information (self-assessed Tanner Scale, physical activity level, parents' origin, and type of residence during upbringing). Two pediatric radiologists independently assessed the MRI images to evaluate their stage of bone development (blinded to age, gender, and each other). All the gathered information was used in training machine learning models for CA estimation and minor versus adult classification (threshold of 18 years). Different machine learning methods were investigated.

RESULTS

The minor versus adult classification produced accuracies of 0.90 and 0.84 for male and female subjects, respectively, with high recalls for the classification of minors. The CA estimation for the 8 age groups (aged 14-21 years) achieved mean absolute errors of 0.95 years and 1.24 years for male and female subjects, respectively. However, for the latter, a lower error occurred only for the ages of 14 and 15 years.

CONCLUSIONS

This study investigates CA estimation through BAA using machine learning methods in 2 ways: minor versus adult classification and CA estimation in 8 age groups (aged 14-21 years), while addressing the drawbacks in the research on BAA. The first achieved good results; however, for the second case, the BAA was not precise enough for the classification.

Modernization of bone age assessment: comparing the accuracy and reliability of an artificial intelligence algorithm and shorthand bone age to Greulich and Pyle

Greulich and Pyle (GP) is one of the most common methods to determine bone age from hand radiographs. In recent years, new methods were developed to increase the efficiency in bone age analysis like the shorthand bone age (SBA) and automated artificial intelligence algorithms.

OBJECTIVE

The aim of this study is to evaluate the accuracy and reliability of these two methods and examine if the reduction in analysis time compromises their efficacy.

METHODS

Two hundred thirteen males and 213 females had their bone age determined by two separate raters using the SBA and GP methods. Three weeks later, the two raters repeated the analysis of the radiographs. The raters timed themselves using an online stopwatch. De-identified radiographs were securely uploaded to an automated algorithm developed by a group of radiologists in Toronto. The gold standard was determined to be the radiology report attached to each radiograph, written by experienced radiologists using GP.

RESULTS

Intraclass correlation between each method and the gold standard fell within the range of 0.8-0.9, highlighting significant agreement. Most of the comparisons showed a statistically significant difference between the new methods and the gold standard; however, it may not be clinically significant as it ranges between 0.25 and 0.5 years. A bone age is considered clinically abnormal if it falls outside 2 standard deviations of the chronological age; standard deviations are calculated and provided in GP atlas.

CONCLUSION

The shorthand bone age method and the automated algorithm produced values that are in agreement with the gold standard while reducing analysis time.

Bone age estimation using deep learning and hand X-ray images

Bones during growth period undergo substantial changes in shape and size. X-ray imaging has been routinely used for bone growth diagnosis purpose. Hand has been the part of choice for X-ray imaging due to its high bone parts count and relatively low radiation requirement. Traditionally, bone age estimation has been performed by referencing atlases of images of hand bone regions where aging-related metamorphoses are most conspicuous. Tanner and Whitehouse' and Greulich and Pyle's are some well known ones. The process entails manual comparison of subject's hand region images against a set of corresponding images in the atlases. It is desired to estimate bone age from hand images in an automated manner, which would facilitate more efficient estimation in terms of time and labor cost and enables quantitative and objective assessments. Deep learning method has proved to be a viable approach in a number of application domains. It is also gaining wider grounds in medical image analysis. A cascaded structure of layers can be trained to mimic the image-based cognitive and inference processes of human and other higher organisms. We employed a set of well known deep learning network architectures. In the current study, 3000 images were manually curated to mark feature points on hands. They were used as reference points in removing unnecessary image regions and to retain regions of interest (ROI) relevant to age estimation. Different ROI's were defined and used-that of rather small area mostly made up of carpal and metacarpal bones and that includes most of phalanges in addition. Irrelevant intensity variation across cropped images was minimized by applying histogram equalization. In consideration of the established gender difference in growth rates, separate gender models were built. Certain age range image data are far scarcer and exhibit rather large excursion in morphology from other age ranges-e.g. infancy and very early childhood. Many studies excluded them and addressed only elder subjects in later developmental stages. Considering infant age group's diagnosis demand is just as valid as elder groups', we included entire age ranges for our study. A number of different deep learning architectures were trained with varying region of interest definitions. Smallest mean absolute difference error was 8.890 months for a test set of 400 images. This study was preliminary, and in the future, we plan to investigate alternative approaches not taken in the present study.

Applications of Artificial Intelligence in Musculoskeletal Imaging: From the Request to the Report

Artificial intelligence (AI) will transform every step in the imaging value chain, including interpretive and noninterpretive components. Radiologists should familiarize themselves with AI developments to become leaders in their clinical implementation. This article explores the impact of AI through the entire imaging cycle of musculoskeletal radiology, from the placement of the requisition to the generation of the report, with an added Canadian perspective. Noninterpretive tasks which may be assisted by AI include the ordering of appropriate imaging tests, automatic exam protocoling, optimized scheduling, shorter magnetic resonance imaging acquisition time, computed tomography imaging with reduced artifact and radiation dose, and new methods of generation and utilization of radiology reports. Applications of AI for image interpretation consist of the determination of bone age, body composition measurements, screening for osteoporosis, identification of fractures, evaluation of segmental spine pathology, detection and temporal monitoring of osseous metastases, diagnosis of primary bone and soft tissue tumors, and grading of osteoarthritis.

Artificial Intelligence Solutions for Analysis of X-ray Images

Artificial intelligence (AI) presents a key opportunity for radiologists to improve quality of care and enhance the value of radiology in patient care and population health. The potential opportunity of AI to aid in triage and interpretation of conventional radiographs (X-ray images) is particularly significant, as radiographs are the most common imaging examinations performed in most radiology departments. Substantial progress has been made in the past few years in the development of AI algorithms for analysis of chest and musculoskeletal (MSK) radiographs, with deep learning now the dominant approach for image analysis. Large public and proprietary image data sets have been compiled and have aided the development of AI algorithms for analysis of radiographs, many of which demonstrate accuracy equivalent to radiologists for specific, focused tasks. This article describes (1) the basis for the development of AI solutions for radiograph analysis, (2) current AI solutions to aid in the triage and interpretation of chest radiographs and MSK radiographs, (3) opportunities for AI to aid in noninterpretive tasks related to radiographs, and (4) considerations for radiology practices selecting AI solutions for radiograph analysis and integrating them into existing IT systems. Although comprehensive AI solutions across modalities have yet to be developed, institutions can begin to select and integrate focused solutions which increase efficiency, increase quality and patient safety, and add value for their patients.

Growth and Adult Height in Girls With Turner Syndrome Following IGF-1 Titrated Growth Hormone Treatment

CONTEXT

Girls with Turner syndrome (TS) suffer linear growth failure, and TS is a registered indication for growth hormone (GH) treatment. GH is classically dosed according to body weight, and serum insulin-like growth factor-1 (IGF-1) concentrations are recommended to be kept within references according to international guidelines.

OBJECTIVE

To assess the effect of long-term GH treatment in girls with TS following GH dosing by IGF-1 titration.

DESIGN AND SETTING

A retrospective, real-world evidence, observational study consisting of data collected in a single tertiary center from 1991 to 2018.

PATIENTS

A cohort of 63 girls with TS treated with GH by IGF-1 titration with a median duration of 6.7 years (interquartile range [IQR]: 3.4-9.7 years).

MAIN OUTCOME MEASURES

Longitudinal measurements of height, IGF-1, and adult height (AH) following GH treatment were evaluated and compared between the different karyotypes (45,X, 45,X/46,XX, or miscellaneous).

RESULTS

Using GH dose titration according to IGF-1, only 6% of girls with TS had supranormal IGF-1 levels. Median dose was 33 µg/kg/day (IQR: 28-39 µg/kg/day) with no difference between the karyotype groups. AH was reached for 73% who attained a median AH of 1.25 standard deviation score (SDS) for age specific TS references (IQR: 0.64-1.50 SDS), and a median gain in height (ΔHSDS: AH SDS minus baseline height SDS of TS references) of 0.50 SDS, equal to 3.2 cm (SD 7.68) for all karyotypes.

CONCLUSION

Our real-world evidence study suggested that titration of GH dose to keep IGF-1 levels within the normal range resulted in a lower AH gain than in studies where a fixed dose was used.

Rethinking Greulich and Pyle: A Deep Learning Approach to Pediatric Bone Age Assessment Using Pediatric Trauma Hand Radiographs

Purpose

To develop a deep learning approach to bone age assessment based on a training set of developmentally normal pediatric hand radiographs and to compare this approach with automated and manual bone age assessment methods based on Greulich and Pyle (GP).

Methods

In this retrospective study, a convolutional neural network (trauma hand radiograph-trained deep learning bone age assessment method [TDL-BAAM]) was trained on 15 129 frontal view pediatric trauma hand radiographs obtained between December 14, 2009, and May 31, 2017, from Children's Hospital of New York, to predict chronological age. A total of 214 trauma hand radiographs from Hasbro Children's Hospital were used as an independent test set. The test set was rated by the TDL-BAAM model as well as a GP-based deep learning model (GPDL-BAAM) and two pediatric radiologists (radiologists 1 and 2) using the GP method. All ratings were compared with chronological age using mean absolute error (MAE), and standard concordance analyses were performed.

Results

The MAE of the TDL-BAAM model was 11.1 months, compared with 12.9 months for GPDL-BAAM (P = .0005), 14.6 months for radiologist 1 (P < .0001), and 16.0 for radiologist 2 (P < .0001). For TDL-BAAM, 95.3% of predictions were within 24 months of chronological age compared with 91.6% for GPDL-BAAM (P = .096), 86.0% for radiologist 1 (P < .0001), and 84.6% for radiologist 2 (P < .0001). Concordance was high between all methods and chronological age (intraclass coefficient > 0.93). Deep learning models demonstrated a systematic bias with a tendency to overpredict age for younger children versus radiologists who showed a consistent mean bias.

Conclusion

A deep learning model trained on pediatric trauma hand radiographs is on par with automated and manual GP-based methods for bone age assessment and provides a foundation for developing population-specific deep learning algorithms for bone age assessment in modern pediatric populations.Supplemental material is available for this article.© RSNA, 2020See also the commentary by Halabi in this issue.

Mimicking the radiologists' workflow: Estimating pediatric hand bone age with stacked deep neural networks

Pediatric endocrinologists regularly order radiographs of the left hand to estimate the degree of bone maturation in order to assess their patients for advanced or delayed growth, physical development, and to monitor consecutive therapeutic measures. The reading of such images is a labor-intensive task that requires a lot of experience and is normally performed by highly trained experts like pediatric radiologists. In this paper we build an automated system for pediatric bone age estimation that mimics and accelerates the workflow of the radiologist without breaking it. The complete system is based on two neural network based models: on the one hand a detector network, which identifies the ossification areas, on the other hand gender and region specific regression networks, which estimate the bone age from the detected areas. With a small annotated dataset an ossification area detection network can be trained, which is stable enough to work as part of a multi-stage approach. Furthermore, our system achieves competitive results on the RSNA Pediatric Bone Age Challenge test set with an average error of 4.56 months. In contrast to other approaches, especially purely encoder-based architectures, our two-stage approach provides self-explanatory results. By detecting and evaluating the individual ossification areas, thus simulating the workflow of the Tanner-Whitehouse procedure, the results are interpretable for a radiologist.

Diagnostic performance of convolutional neural network-based Tanner-Whitehouse 3 bone age assessment system

Background

Bone age can reflect the true growth and development status of a child; thus, it plays a critical role in evaluating growth and endocrine disorders. This study established and validated an optimized Tanner-Whitehouse 3 artificial intelligence (TW3-AI) bone age assessment (BAA) system based on a convolutional neural network (CNN).

Methods

A data set of 9,059 clinical radiographs of the left hand was obtained from the picture archives and communication systems (PACS) between January 2012 and December 2016. Among these, 8,005/9,059 (88%) samples were treated as the training set for model implementation, 804/9,059 (9%) samples as the validation set for parameters optimization, and the remaining 250/9,059 (3%) samples were used to verify the accuracy and reliability of the model compared to that of 4 experienced endocrinologists and 2 experienced radiologists. The overall variation of TW3-metacarpophalangeal, radius, ulna and short bones (RUS) and TW3-Carpal bone score, as well as each bone (13 RUS + 7 Carpal) between reviewers and the AI, were compared by Bland-Altman (BA) chart and Kappa test, respectively. Furthermore, the time consumption between the model and reviewers was also compared.

Results

The performance of TW3-AI model was highly consistent with the reviewers' overall estimation, and the root mean square (RMS) was 0.50 years. The accuracy of the BAA of the TW3-AI model was better than the estimate of the reviewers. Further analysis revealed that human interpretations of the male capitate, hamate, the first distal and fifth middle phalanx and female capitate, the trapezoid, and the third and fifth middle phalanx, were most inconsistent. The average image processing time was 1.5±0.2 s in the TW3-AI model, which was significantly shorter than manual interpretation.

Conclusions

The diagnostic performance of CNN-based TW3 BAA was accurate and timesaving, and possesses better stability compared to diagnostics made by experienced experts.

Assessing the Bone Age of Children in an Automatic Manner Newborn to 18 Years Range

Bone age assessment (BAA) is a radiological process to identify the growth disorders in children. Although this is a frequent task for radiologists, it is cumbersome. The objective of this study is to assess the bone age of children from newborn to 18 years old in an automatic manner through computer vision methods including histogram of oriented gradients (HOG), local binary pattern (LBP), and scale invariant feature transform (SIFT). Here, 442 left-hand radiographs are applied from the University of Southern California (USC) hand atlas. In this experiment, for the first time, HOG-LBP-dense SIFT features with background subtraction are applied to assess the bone age of the subject group. For this purpose, features are extracted from the carpal and epiphyseal regions of interest (ROIs). The SVM and 5-fold cross-validation are used for classification. The accuracy of female radiographs is 73.88% and of the male is 68.63%. The mean absolute error is 0.5 years for both genders' radiographs. The accuracy a within 1-year range is 95.32% for female and 96.51% for male radiographs. The accuracy within a 2-year range is 100% and 99.41% for female and male radiographs, respectively. The Cohen's kappa statistical test reveals that this proposed approach, Cohen's kappa coefficients are 0.71 for female and 0.66 for male radiographs, p value < 0.05, is in substantial agreement with the bone age assessed by experienced radiologists within the USC dataset. This approach is robust and easy to implement, thus, qualified for computer-aided diagnosis (CAD). The reduced processing time and number of ROIs facilitate BAA.

Bone age determination using only the index finger: a novel approach using a convolutional neural network compared with human radiologists

BACKGROUND

Recently developed convolutional neural network (CNN) models determine bone age more accurately than radiologists.

OBJECTIVE

The purpose of this study was to determine whether a CNN and radiologists can accurately predict bone age from radiographs using only the index finger rather than the whole hand.

MATERIALS AND METHODS

We used a public anonymized dataset provided by the Radiological Society of North America (RSNA) pediatric bone age challenge. The dataset contains 12,611 hand radiographs for training and 200 radiographs for testing. The index finger was cropped from these images to create a second dataset. Separate CNN models were trained using the whole-hand radiographs and the cropped second-digit dataset using the consensus ground truth provided by the RSNA bone age challenge. Bone age determination using both models was compared with ground truth as provided by the RSNA dataset. Separately, three pediatric radiologists determined bone age from the whole-hand and index-finger radiographs, and the consensus was compared to the ground truth and CNN-model-determined bone ages.

RESULTS

The mean absolute difference between the ground truth and CNN bone age for whole-hand and index-finger was similar (4.7 months vs. 5.1 months, P=0.14), and both values were significantly smaller than that for radiologist bone age determination from the single-finger radiographs (8.0 months, P<0.0001).

CONCLUSION

CNN-model-determined bone ages from index-finger radiographs are similar to whole-hand bone age interpreted by radiologists in the dataset, as well as a model trained on the whole-hand radiograph. In addition, the index-finger model performed better than the ground truth compared to subspecialty trained pediatric radiologists also using only the index finger to determine bone age. The radiologist interpreting bone age can use the second digit as a reliable starting point in their search pattern.

Towards fully automated third molar development staging in panoramic radiographs

Staging third molar development is commonly used for age assessment in sub-adults. Current staging techniques are, at most, semi-automated and rely on manual interactions prone to operator variability. The aim of this study was to fully automate the staging process by employing the full potential of deep learning, using convolutional neural networks (CNNs) in every step of the procedure. The dataset used to train the CNNs consisted of 400 panoramic radiographs (OPGs), with 20 OPGs per developmental stage per sex, staged in consensus between three observers. The concepts of transfer learning, using pre-trained CNNs, and data augmentation were used to mitigate the issues when dealing with a limited dataset. In this work, a three-step procedure was proposed and the results were validated using fivefold cross-validation. First, a CNN localized the geometrical center of the lower left third molar, around which a square region of interest (ROI) was extracted. Second, another CNN segmented the third molar within the ROI. Third, a final CNN used both the ROI and the segmentation to classify the third molar into its developmental stage. The geometrical center of the third molar was found with an average Euclidean distance of 63 pixels. Third molars were segmented with an average Dice score of 93%. Finally, the developmental stages were classified with an accuracy of 54%, a mean absolute error of 0.69 stages, and a linear weighted Cohen's kappa coefficient of 0.79. The entire automated workflow on average took 2.72 s to compute, which is substantially faster than manual staging starting from the OPG. Taking into account the limited dataset size, this pilot study shows that the proposed fully automated approach shows promising results compared with manual staging.

Fully Automated Bone Age Assessment on Large-Scale Hand X-Ray Dataset

Bone age assessment (BAA) is an essential topic in the clinical practice of evaluating the biological maturity of children. Because the manual method is time-consuming and prone to observer variability, it is attractive to develop computer-aided and automated methods for BAA. In this paper, we present a fully automatic BAA method. To eliminate noise in a raw X-ray image, we start with using U-Net to precisely segment hand mask image from a raw X-ray image. Even though U-Net can perform the segmentation with high precision, it needs a bigger annotated dataset. To alleviate the annotation burden, we propose to use deep active learning (AL) to select unlabeled data samples with sufficient information intentionally. These samples are given to Oracle for annotation. After that, they are then used for subsequential training. In the beginning, only 300 data are manually annotated and then the improved U-Net within the AL framework can robustly segment all the 12611 images in RSNA dataset. The AL segmentation model achieved a Dice score at 0.95 in the annotated testing set. To optimize the learning process, we employ six off-the-shell deep Convolutional Neural Networks (CNNs) with pretrained weights on ImageNet. We use them to extract features of preprocessed hand images with a transfer learning technique. In the end, a variety of ensemble regression algorithms are applied to perform BAA. Besides, we choose a specific CNN to extract features and explain why we select that CNN. Experimental results show that the proposed approach achieved discrepancy between manual and predicted bone age of about 6.96 and 7.35 months for male and female cohorts, respectively, on the RSNA dataset. These accuracies are comparable to state-of-the-art performance.

High frequency of pathogenic ACAN variants including an intragenic deletion in selected individuals with short stature

CONTEXT

Defining the underlying etiology of idiopathic short stature (ISS) improves the overall management of an individual.

OBJECTIVE

To assess the frequency of pathogenic ACAN variants in selected individuals.

DESIGN

The single-center cohort study was conducted at a tertiary university children's hospital. From 51 unrelated patients with ISS, the 16 probands aged between 3 and 18 years (12 females) with advanced bone age and/or autosomal dominant inheritance pattern of short stature were selected for the study. Fifteen family members of ACAN-positive probands were included. Exome sequencing was performed in all probands, and additional copy number variation (CNV) detection was applied in selected probands with a distinct ACAN-associated phenotype.

RESULTS

Systematic phenotyping of the study cohort yielded 37.5% (6/16) ACAN-positive probands, with all novel pathogenic variants, including a 6.082 kb large intragenic deletion, detected by array comparative genomic hybridization (array CGH) and exome data analysis. All variants were co-segregated with short stature phenotype, except in one family member with the intragenic deletion who had an unexpected growth pattern within the normal range (-0.5 SDS). One patient presented with otosclerosis, a sign not previously associated with aggrecanopathy.

CONCLUSIONS

ACAN pathogenic variants presented a common cause of familial ISS. The selection criteria used in our study were suggested for a personalized approach to genetic testing of the ACAN gene in clinical practice. Our results expanded the number of pathogenic ACAN variants, including the first intragenic deletion, and suggested CNV evaluation in patients with typical clinical features of aggrecanopathy as reasonable. Intra-familial phenotypic variability in growth patterns should be considered.

Towards a Rational and Efficient Diagnostic Approach in Children Referred for Tall Stature and/or Accelerated Growth to the General Paediatrician

Tall stature and/or accelerated growth (TS/AG) in a child can be the result of a primary or secondary growth disorder, but more frequently no cause can be found (idiopathic TS). The conditions with the most important therapeutic implications are Klinefelter syndrome, Marfan syndrome and secondary growth disorders such as precocious puberty, hyperthyroidism and growth hormone excess. We propose a diagnostic flow chart offering a systematic approach to evaluate children referred for TS/AG to the general paediatrician. Based on the incidence, prevalence and clinical features of medical conditions associated with TS/AG, we identified relevant clues for primary and secondary growth disorders that may be obtained from the medical history, physical evaluation, growth analysis and additional laboratory and genetic testing. In addition to obtaining a diagnosis, a further goal is to predict adult height based on growth pattern, pubertal development and skeletal maturation. We speculate that an improved diagnostic approach in addition to expanding use of genetic testing may increase the diagnostic yield and lower the age at diagnosis of children with a pathologic cause of TS/AG.

Bone Health Index and bone turnover in pediatric patients with type 1 diabetes mellitus and poor metabolic control

BACKGROUND

There is a need for a non-invasive, affordable, and reliable method for bone health screening in pediatric patients at risk.

OBJECTIVE

To assess Bone Health Index (BHI) in pediatric patients with type 1 diabetes (T1D) and its relation to bone metabolism, age at onset, duration, control, and insulin dose.

SUBJECTS AND METHODS

Left-hand radiographs were obtained from 65 patients with T1D, mean age 11.23 ± 3.89 years, mean disease duration 5.23 ± 3.76 years and mean glycosylated hemoglobin (HbA1c)-83 mmol/mol (9.7%). Blood and 24 hours urine samples were collected for bone and mineral metabolism assessment. BoneXpert was used to determine BHI, Bone Health Index standard deviation score (BHI SDS), and bone age.

RESULTS

Mean BHI SDS was -1.15 ± 1.19 (n = 54). In 20.37% (n = 11) BHI SDS was < -2SD with mean value -2.82 ± 0. 69, P < .001. These patients had lower levels of beta cross laps (0.77  ± 0.33 ng/mL vs 1.17 ± 0.47 ng/mL), osteocalcin (47.20 ± 14.07 ng/mL vs 75.91 ± 32.08 ng/mL), serum magnesium (0.79 ± 0.05 mmol/L vs 0.83 ± 0.06 mmol/L) and phosphorus (1.48 ± 0.29 mmol/L vs 1.71 ± 0.28 mmol/L) but higher ionized calcium (1.29 ± 0.04 mmol/L vs 1.26 ± 0.05 mmol/L), P < .05, compared to patients with BHI SDS in the normal range. We found a positive correlation between BHI SDS and age at manifestation (r = 0.307, P = 0.024) and a negative one with disease duration (r = -0.284, P = .038). No correlations were found with HbA1c, insulin dose, height, weight, BMI.

CONCLUSIONS

To the best of our knowledge, this is the first study to assess bone health in pediatric patients with T1D using BHI. We found significantly decreased cortical bone density and bone turnover in 20.37%. Earlier age at onset and diabetes duration may have a negative impact on cortical bone density in patients with poor control. Longitudinal studies are needed to follow changes or to assess future interventions.

Artificial intelligence in bone age assessment: accuracy and efficiency of a novel fully automated algorithm compared to the Greulich-Pyle method

Background

Bone age (BA) assessment performed by artificial intelligence (AI) is of growing interest due to improved accuracy, precision and time efficiency in daily routine. The aim of this study was to investigate the accuracy and efficiency of a novel AI software version for automated BA assessment in comparison to the Greulich-Pyle method.

Methods

Radiographs of 514 patients were analysed in this retrospective study. Total BA was assessed independently by three blinded radiologists applying the GP method and by the AI software. Overall and gender-specific BA assessment results, as well as reading times of both approaches, were compared, while the reference BA was defined by two blinded experienced paediatric radiologists in consensus by application of the Greulich-Pyle method.

Results

Mean absolute deviation (MAD) and root mean square deviation (RSMD) were significantly lower between AI-derived BA and reference BA (MAD 0.34 years, RSMD 0.38 years) than between reader-calculated BA and reference BA (MAD 0.79 years, RSMD 0.89 years; p < 0.001). The correlation between AI-derived BA and reference BA (r = 0.99) was significantly higher than between reader-calculated BA and reference BA (r = 0.90; p < 0.001). No statistical difference was found in reader agreement and correlation analyses regarding gender (p = 0.241). Mean reading times were reduced by 87% using the AI system.

Conclusions

A novel AI software enabled highly accurate automated BA assessment. It may improve efficiency in clinical routine by reducing reading times without compromising the accuracy compared with the Greulich-Pyle method.

The distal radial physis: Exploring normal anatomy on MRI enables interpretation of stress related changes in young gymnasts

Rationale: Explore the MRI-appearance of the healthy distal radial physis and the distribution of stress-related changes in physeal thickness in young gymnasts to aid in the understanding of the pathophysiological process of stress-related physeal injury. Methods: Symptomatic gymnasts with clinically suspected overuse injury of the distal radial physis and age and gender-matched asymptomatic gymnasts and healthy non-gymnasts underwent an MRI-scan of the wrist. A cartilage-specific sequence was used to obtain three-dimensional reconstructions of the distal radial physis. Heat maps and line charts of these reconstructions visualised distribution of physeal thickness per study group and were used to explore differences between study groups. Symptomatic gymnasts displaying the most profound physeal widening (n = 10) were analysed separately. Results: Twenty-seven symptomatic - (skeletal age 12.9 ± 1.5 years), 16 asymptomatic - (skeletal age 12.8 ± 1.9 years) and 23 non-gymnasts (skeletal age 13.6 ± 1.9 years) were included for analysis. Physes of healthy non-gymnasts had a thin centre and increased in thickness towards the borders. Gymnasts demonstrated an increase in thickness of the entire physeal surface. In symptomatic gymnasts increase in physeal thickness was most prominent at the volar side when compared to asymptomatic gymnasts and non-gymnasts. Conclusion: The healthy distal radial physis is characterised by a thin centre surrounded by thicker borders. Stress applied to the wrist during gymnastics causes an overall increase in physeal thickness. Profound thickness increase is present at the volar side of the physis mainly in symptomatic gymnasts. These results can help unravel the pathophysiological mechanism of stress-related physeal injury in gymnasts and aid early injury identification.

Transition analysis applied to third molar development in a Danish population

INTRODUCTION

Age assessment based on dental development is often requested in order to assess whether an individual is older or younger than 18 years of age. There are several statistical approaches to estimate age based upon third molar development. The aim of this study was to apply the principles of transition analysis (TA) to a Danish reference material and to evaluate whether it was indicated to include a model that allows for logistic non-linearity as opposed to applying a model only allowing for logistic linearity. For this we chose to use the generalized additive model (gam) and the generalized linear model (glm), respectively.

MATERIAL AND METHOD

A cross-sectional sample comprising 1302 panoramic radiographs of Danish subjects in the chronological age range of 13-25 years was included. All present third molars had been scored according to the 10-stage method of Gleiser and Hunt. Each transition from one stage to the subsequent stage was analyzed according to the statistical approach of TA and fitted with both the generalized linear model (glm) and the generalized additive model (gam). In order to assess whether gam or glm was more parsimonious for each transition individually, the Akaikon information criterion (AIC) was applied.

RESULTS

The results emphasized the importance of applying a statistical model that sufficiently captures the spread of the age estimate. The AIC values showed that some transitions were sufficiently described by glm whereas for others the gam curves fitted significantly better.

CONCLUSION

We recommend that for an age assessment tool based on TA, both a fitting allowing for non-linearity and one allowing only for linearity should be included.

It's a thin line: development and validation of Dixon MRI-based semi-quantitative assessment of stress-related bone marrow edema in the wrists of young gymnasts and non-gymnasts

Purpose

To assess reliability and clinical utility of evaluating stress-related metaphyseal water distribution using a semi-quantitative Dixon MRI-based method for early diagnosis of physeal stress injuries in adolescent gymnasts.

Methods

Twenty-four gymnasts with clinically suspected overuse injury of the distal radial physis, 18 asymptomatic gymnasts, and 24 non-gymnast controls aged 12 ± 1.5 years prospectively underwent hand radiographs and 3T MRI of the wrist including coronal T1-weighted and T2-weighted Dixon sequences. Two raters measured metaphyseal water signal fraction in 13 radial and ulnar regions of interest (ROI). Inter- and intrarater reliability, interslice (between 3 middle radial slices), and inter-ROI (between 3 ROIs on same level) reliability were assessed using intraclass correlation coefficients (ICC). Water signal fractions and their within-person ratios in distal versus most proximal ROIs were compared between groups using one-way analysis of variance.

Results

Inter- and intrarater ICCs were 0.79–0.99 and 0.94–1.0 for T1-weighted, and 0.88–1.0 and 0.88–1.0 for T2-weighted Dixon. Interslice and inter-ROI ICCs were 0.55–0.94 and 0.95–0.97 for T1-weighted, and 0.70–0.96 and 0.96–0.97 for T2-weighted Dixon. Metaphyseal water signal fraction in symptomatic gymnasts was higher in six distal ROIs compared with asymptomatic gymnasts and in nine ROIs compared with non-gymnasts (p < 0.05). Metaphyseal water score (ratio of distal versus most proximal ROIs) was 1.61 in symptomatic gymnasts and 1.35 in asymptomatic gymnasts on T2-weighted Dixon (p < 0.05).

Conclusion

Semi-quantitative Dixon MRI-based water signal fraction assessment has good to excellent reproducibility and shows increased metaphyseal water scores in symptomatic gymnasts compared with asymptomatic gymnastic peers.

Key Points

• The proposed Dixon MRI-based semi-quantitative method for assessment of metaphyseal bone marrow water content is reliable, with off-the-shelf availability and short scan times.

• The metaphyseal water score allows comparisons between gymnasts using a within-person reference area for unaffected metaphyseal bone.

• As metaphyseal water score was increased in symptomatic gymnasts compared with asymptomatic gymnasts, this semi-quantitative method can potentially be used as an indicator of bone marrow edema in the early diagnosis of gymnastic physeal stress injury.

Electronic supplementary material

The online version of this article (10.1007/s00330-019-06446-8) contains supplementary material, which is available to authorized users.

Statistical Shape Modeling of Skeletal Anatomy for Sex Discrimination: Their Training Size, Sexual Dimorphism, and Asymmetry

Purpose: Statistical shape modeling provides a powerful tool for describing and analyzing human anatomy. By linearly combining the variance of the shape of a population of a given anatomical entity, statistical shape models (SSMs) identify its main modes of variation and may approximate the total variance of that population to a selected threshold, while reducing its dimensionality. Even though SSMs have been used for over two decades, they lack in characterization of their goodness of prediction, in particular when defining whether these models are actually representative for a given population.

Methods: The current paper presents, to the authors' knowledge, the most extent lower limb anatomy shape model considering the pelvis, femur, patella, tibia, fibula, talus, and calcaneum to date. The present study includes the segmented training shapes (n = 542) obtained from 271 lower limb CT scans. The different models were evaluated in terms of accuracy, compactness, generalizability as well as specificity.

Results: The size of training samples needed in each model so that it can be considered population covering was estimated to approximate around 200 samples, based on the generalizability properties of the different models. Simultaneously differences in gender and patterns in left-right asymmetry were identified and characterized. Size was found to be the most pronounced sexual discriminator whereas intra-individual variations in asymmetry were most pronounced at the insertion site of muscles.

Conclusion: For models aimed at population covering descriptive studies, the number of training samples required should amount a sizeable 200 samples. The geometric morphometric method for sex discrimination scored excellent, however, it did not largely outperformed traditional methods based on discrete measures.

Potential Value of Conventional Ultrasound in Estimation of Bone Age in Patients from Birth to Near Adulthood

We aimed to assess the relationship between standard bone age (BA) stratification obtained using plain radiography and that obtained using ultrasound determination of ossification ratio (OR) as a novelty in patients from birth to near adulthood. The ratio of diameters of the ossification center and epiphysis was calculated to evaluate the OR of bones. The ORs of the bones assigned to different weight coefficient were then summed as the skeletal maturity score (SMS). Pearson's correlation r between SMSs derived from ORs and BAs was 0.97 in girls (p < 0.001) and 0.97 in boys (p < 0.001), respectively. There are significant positive correlations between SMSs measured by conventional ultrasound imaging and BAs obtained by radiography of the hand and wrist in patients from birth to near adulthood. The scoring system may potentially provide a quantitative modality to estimate BA by conventional ultrasound.

Imaging in Short Stature and Bone Age Estimation

Short stature in children is a diagnostic challenge to the physician. Bone age assessment can be done using various methods. The causes of short stature are variable; often leading to a series of investigations. The endocrine conditions have typical imaging features. This chapter provides a short overview of the methods of bone age estimation, and imaging findings and algorithmic approach towards a child with short stature.

Effect of Lower Third Molar Segmentations on Automated Tooth Development Staging using a Convolutional Neural Network

Staging third molar development is commonly used for age estimation in subadults. Automated developmental stage allocation to the mandibular left third molar in panoramic radiographs has been examined in a pilot study. This method used an AlexNet Deep Convolutional Neural Network (CNN) approach to stage lower left third molars, which had been selected by manually drawn bounding boxes around them. This method (bounding box AlexNet = BA) still contained parts of surrounding structures which may have affected the automated stage allocation performance. We hypothesize that segmenting only the third molar could further improve the automated stage allocation performance. Therefore, the current study aimed to determine and validate the effect of lower third molar segmentations on automated tooth development staging. Retrospectively, 400 panoramic radiographs were collected, processed and segmented in three ways: bounding box (BB), rough (RS), and full (FS) tooth segmentation. A DenseNet201 CNN was used for automated stage allocation. Automated staging results were compared with reference stages - allocated by human observers - overall and per stage. FS rendered the best results with a stage allocation accuracy of 0.61, a mean absolute difference of 0.53 stages and a Cohen's linear κ of 0.84. Misallocated stages were mostly neighboring stages, and DenseNet201 rendered better results than AlexNet by increasing the percentage of correctly allocated stages by 3% (BA compared to BB). FS increased the percentage of correctly allocated stages by 7% compared to BB. In conclusion, full tooth segmentation and a DenseNet CNN optimize automated dental stage allocation for age estimation.

Applicability of two commonly used bone age assessment methods to twenty-first century UK children

Objectives

To assess the effect of secular change on skeletal maturation and thus on the applicability of the Greulich and Pyle (G&P) and Tanner and Whitehouse (TW3) methods.

Methods

BoneXpert was used to assess bone age from 392 hand trauma radiographs (206 males, 257 left). The paired sample t test was performed to assess the difference between mean bone age (BA) and mean chronological age (CA). ANOVA was used to assess the differences between groups based on socioeconomic status (taken from the Index of Multiple Deprivation).

Results

CA ranged from 2 to 15 years for females and 2.5 to 15 years for males. Numbers of children living in low, average and high socioeconomic areas were 216 (55%), 74 (19%) and 102 (26%) respectively. We found no statistically significant difference between BA and CA when using G&P. However, using TW3, CA was underestimated in females beyond the age of 3 years, with significant differences between BA and CA (− 0.43 years, SD 1.05, p = < 0.001) but not in males (0.01 years, SD 0.97, p = 0.76). Of the difference in females, 17.8% was accounted for by socioeconomic status.

Conclusion

No significant difference exists between BoneXpert-derived BA and CA when using the G&P atlas in our study population. There was a statistically significant underestimation of BoneXpert-derived BA compared with CA in females when using TW3, particularly in those from low and average socioeconomic backgrounds. Secular change has not led to significant advancement in skeletal maturation within our study population.

Key Points

• The Greulich and Pyle method can be applied to the present-day United Kingdom (UK) population.

• The Tanner and Whitehouse (TW3) method consistently underestimates the age of twenty-first century UK females by an average of 5 months.

• Secular change has not advanced skeletal maturity of present-day UK children compared with those of the mid-twentieth century.

Electronic supplementary material

The online version of this article (10.1007/s00330-019-06300-x) contains supplementary material, which is available to authorized users.

Bone age determination in eutrophic, overweight and obese Brazilian children and adolescents: a comparison between computerized BoneXpert and Greulich-Pyle methods

BACKGROUND

Bone age determination is usually employed to evaluate growth disorders and their treatment. The Greulich-Pyle method is the simplest and most frequently used type of evaluation, but it presents huge interobserver variability. The BoneXpert is a computer-automated method developed to avoid significant bone age variability among distinct observers.

OBJECTIVE

To compare the BoneXpert and Greulich-Pyle methods of bone age determination in eutrophic children and adolescents, as well as in overweight and obese pediatric patients.

MATERIALS AND METHODS

The sample comprised 515 participants, 253 boys (159 eutrophic, 53 overweight and 41 obese) and 262 girls (146 eutrophic, 76 overweight and 40 obese). Left hand and wrist radiographs were acquired for bone age determination using both methods.

RESULTS

There was a positive correlation between chronological age and Greulich-Pyle, chronological age and BoneXpert, and Greulich-Pyle and BoneXpert. There was a significant increase (P<0.05) in bone age in both the Greulich-Pyle and BoneXpert methods in obese boys when compared to eutrophic or overweight boys of the same age. In girls, there was an increase in bone age in both obese and overweight individuals when compared to eutrophic girls (P<0.05). The Greulich-Pyle bone age was advanced in comparison to that of BoneXpert in all groups, except in obese boys, in which bone age was similarly advanced in both methods.

CONCLUSION

The BoneXpert computer-automated bone age determination method showed a significant positive correlation with chronological age and Greulich-Pyle. Furthermore, the impact of being overweight or obese on bone age could be identified by both methods.

Automated age estimation from MRI volumes of the hand

Highly relevant for both clinical and legal medicine applications, the established radiological methods for estimating unknown age in children and adolescents are based on visual examination of bone ossification in X-ray images of the hand. Our group has initiated the development of fully automatic age estimation methods from 3D MRI scans of the hand, in order to simultaneously overcome the problems of the radiological methods including (1) exposure to ionizing radiation, (2) necessity to define new, MRI specific staging systems, and (3) subjective influence of the examiner. The present work provides a theoretical background for understanding the nonlinear regression problem of biological age estimation and chronological age approximation. Based on this theoretical background, we comprehensively evaluate machine learning methods (random forests, deep convolutional neural networks) with different simplifications of the image information used as an input for learning. Trained on a large dataset of 328 MR images, we compare the performance of the different input strategies and demonstrate unprecedented results. For estimating biological age, we obtain a mean absolute error of 0.37 ± 0.51 years for the age range of the subjects  ≤  18 years, i.e. where bone ossification has not yet saturated. Finally, we validate our findings by adapting our best performing method to 2D images and applying it to a publicly available dataset of X-ray images, showing that we are in line with the state-of-the-art automatic methods for this task.

Bone age assessment with various machine learning techniques: A systematic literature review and meta-analysis

BACKGROUND

The assessment of bone age and skeletal maturity and its comparison to chronological age is an important task in the medical environment for the diagnosis of pediatric endocrinology, orthodontics and orthopedic disorders, and legal environment in what concerns if an individual is a minor or not when there is a lack of documents. Being a time-consuming activity that can be prone to inter- and intra-rater variability, the use of methods which can automate it, like Machine Learning techniques, is of value.

OBJECTIVE

The goal of this paper is to present the state of the art evidence, trends and gaps in the research related to bone age assessment studies that make use of Machine Learning techniques.

METHOD

A systematic literature review was carried out, starting with the writing of the protocol, followed by searches on three databases: Pubmed, Scopus and Web of Science to identify the relevant evidence related to bone age assessment using Machine Learning techniques. One round of backward snowballing was performed to find additional studies. A quality assessment was performed on the selected studies to check for bias and low quality studies, which were removed. Data was extracted from the included studies to build summary tables. Lastly, a meta-analysis was performed on the performances of the selected studies.

RESULTS

26 studies constituted the final set of included studies. Most of them proposed automatic systems for bone age assessment and investigated methods for bone age assessment based on hand and wrist radiographs. The samples used in the studies were mostly comprehensive or bordered the age of 18, and the data origin was in most of cases from United States and West Europe. Few studies explored ethnic differences.

CONCLUSIONS

There is a clear focus of the research on bone age assessment methods based on radiographs whilst other types of medical imaging without radiation exposure (e.g. magnetic resonance imaging) are not much explored in the literature. Also, socioeconomic and other aspects that could influence in bone age were not addressed in the literature. Finally, studies that make use of more than one region of interest for bone age assessment are scarce.

Damage of the distal radial physis in young gymnasts: can three-dimensional assessment of physeal volume on MRI serve as a biomarker?

Objective

To explore the use of quantitative volume assessment to identify the presence and extent of stress-related changes of the distal radial physis in gymnasts with suspected physeal injury, asymptomatic gymnasts, and non-gymnasts.

Methods

Symptomatic gymnasts with clinically suspected distal radial physeal injury, asymptomatic gymnasts, and non-gymnasts (n = 69) were included and matched on skeletal age and sex. Volume measurements were performed on coronal water selective cartilage MRI images by creating three-dimensional physeal reconstructions semi-automatically using active-contour segmentation based on image-intensity thresholding. Inter- and intra-rater reliability of the measurements were assessed using intra-class correlation coefficients (ICC) for absolute agreement.

Results

Twenty-seven symptomatic-, 18 asymptomatic-, and 24 non-gymnasts were included with a median age of 13.9 years (interquartile range (IQR) 13.0–15.0 years). Median physeal volume was significantly increased (p < 0.05) in symptomatic- (971 mm³, IQR 787–1237 mm³) and asymptomatic gymnasts (951 mm³, IQR 871–1004 mm³) compared with non-gymnasts (646 mm³, IQR 538–795 mm³). Inter-rater (ICC 0.96, 95% CI 0.92–0.98) and intra-rater (ICC 0.93, 95% CI 0.85–0.97) reliability of volume measurements were excellent. Of the 10 participants with the highest physeal volumes, nine were symptomatic gymnasts.

Conclusion

Increased volume of the distal radial physis can reliably be assessed and is a sign of physeal stress that can be present in both symptomatic- and asymptomatic gymnasts, but gymnasts with suspected physeal injury showed larger volume increases. Future studies should explore if volume assessment can be used to (early) identify athletes with or at risk for physeal stress injuries of the wrist.

Key Points

• The volume of the distal radial physis can be reliably assessed by creating three-dimensional physeal reconstructions.

• Stress-related volume increase of the distal radial physis is present in symptomatic and asymptomatic gymnasts.

• Gymnasts with clinically suspected physeal injury showed larger volume increases compared with asymptomatic gymnasts and may therefore be a valuable addition in the (early) diagnostic workup of physeal stress injuries.

Incorporated region detection and classification using deep convolutional networks for bone age assessment

Bone age assessment plays an important role in the endocrinology and genetic investigation of patients. In this paper, we proposed a deep learning-based approach for bone age assessment by integration of the Tanner-Whitehouse (TW3) methods and deep convolution networks based on extracted regions of interest (ROI)-detection and classification using Faster-RCNN and Inception-v4 networks, respectively. The proposed method allows exploration of expert knowledge from TW3 and features engineering from deep convolution networks to enhance the accuracy of bone age assessment. The experimental results showed a mean absolute error of about 0.59 years between expert radiologists and the proposed method, which is the best performance among state-of-the-art methods.

A multi-scale data fusion framework for bone age assessment with convolutional neural networks

Bone age assessment (BAA) has various clinical applications such as diagnosis of endocrine disorders and prediction of final adult height for adolescents. Recent studies indicate that deep learning techniques have great potential in developing automated BAA methods with significant advantages over the conventional methods based on handcrafted features. In this paper, we propose a multi-scale data fusion framework for bone age assessment with X-ray images based on non-subsampled contourlet transform (NSCT) and convolutional neural networks (CNNs). Unlike the existing CNN-based BAA methods that adopt the original spatial domain image as network input directly, we pre-extract a rich set of features for the input image by performing NSCT to obtain its multi-scale and multi-direction representations. This feature pre-extraction strategy could be beneficial to network training as the number of annotated examples in the problem of BAA is typically quite limited. The obtained NSCT coefficient maps at each scale are fed into a convolutional network individually and the information from different scales are then merged to achieve the final prediction. Specifically, two CNN models with different data fusion strategies are presented for BAA: a regression model with feature-level fusion and a classification model with decision-level fusion. Experiments on the public BAA dataset Digital Hand Atlas demonstrate that the proposed method can obtain promising results and outperform many state-of-the-art BAA methods. In particular, the proposed approaches exhibit obvious advantages over the corresponding spatial domain approaches (generally with an improvement of more than 0.1 years on the mean absolute error), showing great potential in the future study of this field.

Bone age as a correction factor for the analysis of trabecular bone score (TBS) in children

Trabecular bone score (TBS) is a tool to improve evaluation of DXA scans, barely used in children. We proposed to evaluate TBS with bone age (BA) compared to chronological age (CA). In girls, TBS value using BA is constant until age 8, and in boys until age 10, and then starts to increase steadily. This data may help widen TBS use in pediatric populations.

INTRODUCTION

Trabecular bone score (TBS) is a software-based tool for the analysis of DXA images to assess bone microarchitecture in the lumbar region. It is used widely in adults to improve evaluation of fracture risk, yet it has been rarely studied in children and no normal curves have been developed for pediatrics. The purpose of this study was to evaluate bone (skeletal) age compared to chronological age to determine which is better in the pediatric population since both bone age (BA) and trabecular density are equally susceptible to change in response to similar factors.

METHODS

Total body, lumbar region, and non-dominant hand scans were obtained with an iDXA device in all participants. DXA scans of lumbar region for TBS analysis and AP images of non-dominant hand-for-BA were obtained for 565 children (269 female) aged 4to 19.

RESULTS

Simple correlation was calculated and r² values for TBS and chronological age were obtained by linear regression, with low correlations (0.36 for boys and 0.38 for girls), and then we created Loess curves to show the change for consecutive ages. In girls, the curve forms a U shape with a nadir point at approximately age 10. We then replaced chronological age with BA, and significant change was seen in the girls' curve, where a turning point is seen at age 8. In boys, a similar trend shows a turning point at age 10. Finally, BA-corrected TBS curves were constructed using LMS, obtaining curves with percentiles.

CONCLUSIONS

The use of BA in the analysis and interpretation of TBS may help widen its use in pediatric populations by enabling the appearance of normative data, but more information is needed to confirm this finding.

Artificial intelligence-assisted interpretation of bone age radiographs improves accuracy and decreases variability

OBJECTIVE

Radiographic bone age assessment (BAA) is used in the evaluation of pediatric endocrine and metabolic disorders. We previously developed an automated artificial intelligence (AI) deep learning algorithm to perform BAA using convolutional neural networks. We compared the BAA performance of a cohort of pediatric radiologists with and without AI assistance.

MATERIALS AND METHODS

Six board-certified, subspecialty trained pediatric radiologists interpreted 280 age- and gender-matched bone age radiographs ranging from 5 to 18 years. Three of those radiologists then performed BAA with AI assistance. Bone age accuracy and root mean squared error (RMSE) were used as measures of accuracy. Intraclass correlation coefficient evaluated inter-rater variation.

RESULTS

AI BAA accuracy was 68.2% overall and 98.6% within 1 year, and the mean six-reader cohort accuracy was 63.6 and 97.4% within 1 year. AI RMSE was 0.601 years, while mean single-reader RMSE was 0.661 years. Pooled RMSE decreased from 0.661 to 0.508 years, all individually decreasing with AI assistance. ICC without AI was 0.9914 and with AI was 0.9951.

CONCLUSIONS

AI improves radiologist's bone age assessment by increasing accuracy and decreasing variability and RMSE. The utilization of AI by radiologists improves performance compared to AI alone, a radiologist alone, or a pooled cohort of experts. This suggests that AI may optimally be utilized as an adjunct to radiologist interpretation of imaging studies to improve performance.

Letrozole versus testosterone for promotion of endogenous puberty in boys with constitutional delay of growth and puberty: a randomised controlled phase 3 trial

BACKGROUND

The treatment of constitutional delay of growth and puberty (CDGP) is an underinvestigated area in adolescent medicine. We tested the hypothesis that peroral aromatase inhibition with letrozole is more efficacious than intramuscular injection of low-dose testosterone in inducing puberty in boys with CDGP.

METHODS

We did a randomised, controlled, open-label trial at four paediatric centres in Finland. Boys aged at least 14 years with CDGP who wanted medical intervention and exhibited the first signs of puberty were randomly assigned in blocks of ten to receive either six intramuscular injections of low-dose testosterone (about 1 mg/kg bodyweight) every 4 weeks for 6 months or peroral letrozole 2·5 mg once daily for 6 months. All boys were followed up for 6 months after the end of treatment. The primary outcomes were changes in testicular volume and hormonal markers of puberty at 6 months after treatment initiation, which were assessed in all participants who received the assigned treatment. All patients were included in the safety analysis. This study is registered with ClinicalTrials.gov, number NCT01797718.

FINDINGS

Between Aug 1, 2013, and Jan 30, 2017, 30 boys were randomly assigned to receive testosterone (n=15) or letrozole (n=15). One boy in the testosterone group was excluded from the primary analyses because of a protocol deviation. During treatment, boys in the letrozole group had higher serum concentrations of luteinising hormone, follicle-stimulating hormone, testosterone, and inhibin B than did boys in the testosterone group. Testicular growth from baseline to 6 months was greater in the letrozole group than in the testosterone group (7·2 mL [95% CI 5·2-9·3] vs 2·2 mL [1·4-2·9]; between-group difference per month 0·9 mL [95% CI 0·6-1·2], p<0·0001). Most adverse events were mild. One boy in the testosterone group had aggressive behaviour for 1 week after each injection, and one boy in the letrozole group had increased irritability at 6 months.

INTERPRETATION

Letrozole might be a feasible alternative treatment to low-dose testosterone for boys with CDGP who opt for medical intervention. However, the risks and benefits of manipulating the reproductive axis during early puberty should be weighed carefully.

FUNDING

Helsinki University Hospital, Academy of Finland, and Finnish Foundation for Pediatric Research.

Ossification area localization in pediatric hand radiographs using deep neural networks for object detection

BACKGROUND

Detection of ossification areas of hand bones in X-ray images is an important task, e.g. as a preprocessing step in automated bone age estimation. Deep neural networks have emerged recently as de facto standard detection methods, but their drawback is the need of large annotated datasets. Finetuning pre-trained networks is a viable alternative, but it is not clear a priori if training with small annotated datasets will be successful, as it depends on the problem at hand. In this paper, we show that pre-trained networks can be utilized to produce an effective detector of ossification areas in pediatric X-ray images of hands.

METHODS AND FINDINGS

A publicly available Faster R-CNN network, pre-trained on the COCO dataset, was utilized and finetuned with 240 manually annotated radiographs from the RSNA Pediatric Bone Age Challenge, which comprises over 14.000 pediatric radiographs. The validation is done on another 89 radiographs from the dataset and the performance is measured by Intersection-over-Union (IoU). To understand the effect of the data size on the pre-trained network, subsampling was applied to the training data and the training was repeated. Additionally, the network was trained from scratch without any pre-trained weights. Finally, to understand whether the trained model could be useful, we compared the inference of the network to an annotation of an expert radiologist. The finetuned network was able to achieve an average precision (mAP@0.5IoU) of 92.92 ± 1.93. Apart from the wrist region, all ossification areas were able to benefit from more data. In contrast, the network trained from scratch was not able to produce any correct results. When compared to the annotations of the expert radiologist, the network was able to localize the regions quite well, as the F1-Score was on average 91.85 ± 1.06.

CONCLUSIONS

By finetuning a pre-trained deep neural network, with 240 annotated radiographs, we were able to successfully detect ossification areas in prediatric hand radiographs.

A Deep Automated Skeletal Bone Age Assessment Model with Heterogeneous Features Learning

Skeletal bone age assessment is a widely used standard procedure in both disease detection and growth prediction for children in endocrinology. Conventional manual assessment methods mainly rely on personal experience in observing X-ray images of left hand and wrist to calculate bone age, which show some intrinsic limitations from low efficiency to unstable accuracy. To address these problems, some automated methods based on image processing or machine learning have been proposed, while their performances are not satisfying enough yet in assessment accuracy. Motivated by the remarkable success of deep learning (DL) techniques in the fields of image classification and speech recognition, we develop a deep automated skeletal bone age assessment model based on convolutional neural networks (CNNs) and support vector regression (SVR) using multiple kernel learning (MKL) algorithm to process heterogeneous features in this paper. This deep framework has been constructed, not only exploring the X-ray images of hand and twist but also some other heterogeneous information like race and gender. The experiment results prove its better performance with higher bone age assessment accuracy on two different data sets compared with the state of the art, indicating that the fused heterogeneous features provide a better description of the degree of bones' maturation.

Automatic Age Estimation and Majority Age Classification From Multi-Factorial MRI Data

Age estimation from radiologic data is an important topic both in clinical medicine as well as in forensic applications, where it is used to assess unknown chronological age or to discriminate minors from adults. In this paper, we propose an automatic multi-factorial age estimation method based on MRI data of hand, clavicle, and teeth to extend the maximal age range from up to 19 years, as commonly used for age assessment based on hand bones, to up to 25 years, when combined with clavicle bones and wisdom teeth. Fusing age-relevant information from all three anatomical sites, our method utilizes a deep convolutional neural network that is trained on a dataset of 322 subjects in the age range between 13 and 25 years, to achieve a mean absolute prediction error in regressing chronological age of 1.01±0.74 years. Furthermore, when used for majority age classification, we show that a classifier derived from thresholding our regression-based predictor is better suited than a classifier directly trained with a classification loss, especially when taking into account that those cases of minors being wrongly classified as adults need to be minimized. In conclusion, we overcome the limitations of the multi-factorial methods currently used in forensic practice, i.e., dependence on ionizing radiation, subjectivity in quantifying age-relevant information, and lack of an established approach to fuse this information from individual anatomical sites.

Evaluation of a Computer-Aided Diagnosis System for Automated Bone Age Assessment in Comparison to the Greulich-Pyle Atlas Method: A Multireader Study

OBJECTIVE

The aim of this study was to investigate a novel version of a computer-aided diagnosis (CAD) system developed for automated bone age (BA) assessment in comparison to the Greulich and Pyle method, regarding its accuracy and the influence of carpal bones on BA assessment.

METHODS

Total BA, BA of the left distal radius, and BA of carpal bones in 305 patients were determined independently by 3 blinded radiologists and assessed by the CAD system. Pearson product-moment correlation, Bland-Altman plot, root-mean-square deviation, and further agreement analyses were computed.

RESULTS

Mean total BA and BA of the distal radius showed high correlation between both approaches (r = 0.985 and r = 0.963). There was significantly higher correlation between values of total BA and BA of the distal radius (r = 0.969) compared with values of total BA and BA of carpal bones (r = 0.923). The assessment of carpal bones showed significantly lower interreader agreement compared with measurements of the distal radius (κ = 0.79 vs κ = 0.98).

CONCLUSION

A novel version of a CAD system enables highly accurate automated BA assessment. The assessment of carpal bones revealed lower precision and interreader agreement. Therefore, methods determining BA without analyzing carpal bones may be more precise and accurate.

MABAL: a Novel Deep-Learning Architecture for Machine-Assisted Bone Age Labeling

Bone age assessment (BAA) is a commonly performed diagnostic study in pediatric radiology to assess skeletal maturity. The most commonly utilized method for assessment of BAA is the Greulich and Pyle method (Pediatr Radiol 46.9:1269-1274, 2016; Arch Dis Child 81.2:172-173, 1999) atlas. The evaluation of BAA can be a tedious and time-consuming process for the radiologist. As such, several computer-assisted detection/diagnosis (CAD) methods have been proposed for automation of BAA. Classical CAD tools have traditionally relied on hard-coded algorithmic features for BAA which suffer from a variety of drawbacks. Recently, the advent and proliferation of convolutional neural networks (CNNs) has shown promise in a variety of medical imaging applications. There have been at least two published applications of using deep learning for evaluation of bone age (Med Image Anal 36:41-51, 2017; JDI 1-5, 2017). However, current implementations are limited by a combination of both architecture design and relatively small datasets. The purpose of this study is to demonstrate the benefits of a customized neural network algorithm carefully calibrated to the evaluation of bone age utilizing a relatively large institutional dataset. In doing so, this study will aim to show that advanced architectures can be successfully trained from scratch in the medical imaging domain and can generate results that outperform any existing proposed algorithm. The training data consisted of 10,289 images of different skeletal age examinations, 8909 from the hospital Picture Archiving and Communication System at our institution and 1383 from the public Digital Hand Atlas Database. The data was separated into four cohorts, one each for male and female children above the age of 8, and one each for male and female children below the age of 10. The testing set consisted of 20 radiographs of each 1-year-age cohort from 0 to 1 years to 14-15+ years, half male and half female. The testing set included left-hand radiographs done for bone age assessment, trauma evaluation without significant findings, and skeletal surveys. A 14 hidden layer-customized neural network was designed for this study. The network included several state of the art techniques including residual-style connections, inception layers, and spatial transformer layers. Data augmentation was applied to the network inputs to prevent overfitting. A linear regression output was utilized. Mean square error was used as the network loss function and mean absolute error (MAE) was utilized as the primary performance metric. MAE accuracies on the validation and test sets for young females were 0.654 and 0.561 respectively. For older females, validation and test accuracies were 0.662 and 0.497 respectively. For young males, validation and test accuracies were 0.649 and 0.585 respectively. Finally, for older males, validation and test set accuracies were 0.581 and 0.501 respectively. The female cohorts were trained for 900 epochs each and the male cohorts were trained for 600 epochs. An eightfold cross-validation set was employed for hyperparameter tuning. Test error was obtained after training on a full data set with the selected hyperparameters. Using our proposed customized neural network architecture on our large available data, we achieved an aggregate validation and test set mean absolute errors of 0.637 and 0.536 respectively. To date, this is the best published performance on utilizing deep learning for bone age assessment. Our results support our initial hypothesis that customized, purpose-built neural networks provide improved performance over networks derived from pre-trained imaging data sets. We build on that initial work by showing that the addition of state-of-the-art techniques such as residual connections and inception architecture further improves prediction accuracy. This is important because the current assumption for use of residual and/or inception architectures is that a large pre-trained network is required for successful implementation given the relatively small datasets in medical imaging. Instead we show that a small, customized architecture incorporating advanced CNN strategies can indeed be trained from scratch, yielding significant improvements in algorithm accuracy. It should be noted that for all four cohorts, testing error outperformed validation error. One reason for this is that our ground truth for our test set was obtained by averaging two pediatric radiologist reads compared to our training data for which only a single read was used. This suggests that despite relatively noisy training data, the algorithm could successfully model the variation between observers and generate estimates that are close to the expected ground truth.

Quantification of osteoblastic activity in epiphyseal growth plates by quantitative bone SPECT/CT

OBJECTIVE

Quantifying the function of the epiphyseal plate is worthwhile for the management of children with growth disorders. The aim of this retrospective study was to quantify the osteoblastic activity at the epiphyseal plate using the quantitative bone SPECT/CT.

MATERIALS AND METHODS

We enrolled patients under the age of 20 years who received Tc-99m hydroxymethylene diphosphonate bone scintigraphy acquired by a quantitative SPECT/CT scanner. The images were reconstructed by ordered subset conjugate-gradient minimizer, and the uptake on the distal margin of the femur was quantified by peak standardized uptake value (SUVpeak). A public database of standard body height was used to calculate growth velocities (cm/year).

RESULTS

Fifteen patients (6.9-19.7 years, 9 female, 6 male) were enrolled and a total of 25 legs were analyzed. SUVpeak in the epiphyseal plate was 18.9 ± 2.4 (average ± standard deviation) in the subjects under 15 years and decreased gradually by aging. The SUVpeak correlated significantly with the age- and sex-matched growth velocity obtained from the database (R² = 0.83, p < 0.0001).

CONCLUSION

The SUV measured by quantitative bone SPECT/CT was increased at the epiphyseal plates of children under the age of 15 years in comparison with the older group, corresponding to higher osteoblastic activity. Moreover, this study suggested a correlation between growth velocity and the SUV. Although this is a small retrospective pilot study, the objective and quantitative values measured by the quantitative bone SPECT/CT has the potential to improve the management of children with growth disorder.

An automated technique to stage lower third molar development on panoramic radiographs for age estimation: a pilot study

BACKGROUND

Automated methods to evaluate growth of hand and wrist bones on radiographs and magnetic resonance imaging have been developed. They can be applied to estimate age in children and subadults. Automated methods require the software to (1) recognise the region of interest in the image(s), (2) evaluate the degree of development and (3) correlate this to the age of the subject based on a reference population. For age estimation based on third molars an automated method for step (1) has been presented for 3D magnetic resonance imaging and is currently being optimised (Unterpirker et al. 2015).

AIM

To develop an automated method for step (2) based on lower third molars on panoramic radiographs.

MATERIALS AND METHODS

A modified Demirjian staging technique including ten developmental stages was developed. Twenty panoramic radiographs per stage per gender were retrospectively selected for FDI element 38. Two observers decided in consensus about the stages. When necessary, a third observer acted as a referee to establish the reference stage for the considered third molar. This set of radiographs was used as training data for machine learning algorithms for automated staging. First, image contrast settings were optimised to evaluate the third molar of interest and a rectangular bounding box was placed around it in a standardised way using Adobe Photoshop CC 2017 software. This bounding box indicated the region of interest for the next step. Second, several machine learning algorithms available in MATLAB R2017a software were applied for automated stage recognition. Third, the classification performance was evaluated in a 5-fold cross-validation scenario, using different validation metrics (accuracy, Rank-N recognition rate, mean absolute difference, linear kappa coefficient).

RESULTS

Transfer Learning as a type of Deep Learning Convolutional Neural Network approach outperformed all other tested approaches. Mean accuracy equalled 0.51, mean absolute difference was 0.6 stages and mean linearly weighted kappa was 0.82.

CONCLUSION

The overall performance of the presented automated pilot technique to stage lower third molar development on panoramic radiographs was similar to staging by human observers. It will be further optimised in future research, since it represents a necessary step to achieve a fully automated dental age estimation method, which to date is not available.

Bone Age: A Handy Tool for Pediatric Providers

Pediatricians have relied on methods for determining skeletal maturation for >75 years. Bone age continues to be a valuable tool in assessing children's health. New technology for bone age determination includes computer-automated readings and assessments obtained from alternative imaging modalities. In addition, new nonclinical bone age applications are evolving, particularly pertaining to immigration and children's rights to asylum. Given the significant implications when bone ages are used in high-stake decisions, it is necessary to recognize recently described limitations in predicting accurate age in various ethnicities and diseases. Current methods of assessing skeletal maturation are derived from primarily white populations. In modern studies, researchers have explored the accuracy of bone age across various ethnicities in the United States. Researchers suggest there is evidence that indicates the bone ages obtained from current methods are less generalizable to children of other ethnicities, particularly children with African and certain Asian backgrounds. Many of the contemporary methods of bone age determination may be calibrated to individual populations and hold promise to perform better in a wider range of ethnicities, but more data are needed.