Bone age is one of biological indicators of maturity used in clinical practice and it is a very important parameter of a child's assessment, especially in paediatric endocrinology. The most widely used method of bone age assessment is by performing a hand and wrist radiograph and its analysis with Greulich-Pyle or Tanner-Whitehouse atlases, although it has been about 60 years since they were published. Due to the progress in the area of Computer-Aided Diagnosis and application of artificial intelligence in medicine, lately, numerous programs for automatic bone age assessment have been created. Most of them have been verified in clinical studies in comparison to traditional methods showing good precision while eliminating inter- and intra-rater variability and reducing significantly the time of assessment. Also, there are available methods of assessment of bone age without X-ray exposure, like via ultrasound devices or MRI.

Traditional and New Methods of Bone Age Assessment-An Overview

Pediatric bone age assessment (BAA) is a common clinical practice to investigate endocrinology, genetic and growth disorders of children. The morphological characters of different specific bone parts, such as wrist and phalanx, have important reference significance in BAA. Previous deep learning approaches can be divided into two branches, (1) the single-stage structure ignores the attention on specific bone parts, thus it can be trained end-to-end but suffers from low accuracy, (2) the multi-stage structure extracts the bone parts with human prior, thus it exhibits high accuracy but suffers from model generalization and resource consumption problem. To enable an end-to-end training method extracting discriminative bone parts automatically without human prior, in this paper, we propose a novel single-stage Attention-Recognition Convolutional Neural Network (AR-CNN). The AR-CNN consists of one attention agent for discriminative bone parts proposing and one recognition agent for feature learning and age assessment. The attention agent can discover and extract bone parts automatically, meanwhile the recognition agent can learn the features from the proposing bone parts and assess the bone age. Furthermore, the assessment result will be fed back to attention agent for the optimization of bone parts extracting. Therefore, the two agents can reinforce each other mutually and the overall network can be trained end-to-end without human prior. To the best of our knowledge, this is the first end-to-end structure to extract bone parts for BAA without segmentation, detection and human prior. Experimental results show that our approach achieves state-of-the-art accuracy on the public RSNA datasets with mean absolute error(MAE) of 4.38 months.

Extract Bone Parts Without Human Prior: End-to-end Convolutional Neural Network for Pediatric Bone Age Assessment

Assessment of skeletal maturity is important for a clinician to make a decision of the most appropriate treatment on various skeletal disorders. This task is very challenging when using machine learning method due to the limited data and large anatomical variations among different subjects. In this article, we propose an ensemble-based deep learning pipeline to automatically assess the distal radius and ulna (DRU) maturity from left-hand radiographs. At the same time, we adapted the concept of densely connected mechanism in the proposed network architecture to reuse features and prevent gradient disappearance. Therefore, the model acquires two convincing advantages: first, our model preserves the maximum information flow and has a much faster convergence rate. Second, our model avoids overfitting even if training with limited data. The experimental dataset contains 1189 left-hand <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$X$ </tex-math></inline-formula> -ray scans of children and teenagers. The proposed method achieves 85.27% and 91.68% for radius and ulna classification, respectively. Extensive experiments prove that our model performs better than using other network structures. 

An Ensemble-Based Densely-Connected Deep Learning System for Assessment of Skeletal Maturity

Pediatricians and pediatric endocrinologists utilize Bone Age Assessment (BAA) for in-vestigations pertaining to genetic disorders, hormonal complications and abnormalities in the skeletal system maturity of children. Conventional methods dating back to 1950 were often tedious and suscep-tible to inter-observer variability, and preceding attempts to improve these traditional techniques have inadequately addressed the human expert inter-observer variability so as to significantly refine bone age evaluations. In this paper, an automated and efficient approach with regression convolutional neu-ral network is proposed. This approach automatically exploits the carpal bones as the region of interest (ROI) and performs boundary extraction of carpal bones, then based on the regression convolutional neural network it evaluates the skeletal age from the left hand wrist radiograph of young children. Experiments show that the proposed method achieves an average discrepancy of 2.75 months between clinical and automatic bone age evaluations, and achieves 90.15% accuracy within 6 months from the ground truth for male. Further experimental results with test radiographs assigned an accuracy within 1 year achieved 99.43% accuracy.

Skeletal bone age assessments for young children based on regression convolutional neural networks.

Pediatric bone age assessment (BAA) is a common clinical practice to investigate endocrinology, genetic and growth disorders of children. Different specific bone parts are extracted as anatomical Regions of Interest (RoIs) during this task, since their morphological characters have important biological identification in skeletal maturity. Following this clinical prior knowledge, recently developed deep learning methods address BAA with an RoI-based attention mechanism, which segments or detects the discriminative RoIs for meticulous analysis. Great strides have been made, however, these methods strictly require large and precise RoIs annotations, which limits the real-world clinical value. To overcome the severe requirements on RoIs annotations, in this paper, we propose a novel self-supervised learning mechanism to effectively discover the informative RoIs without the need of extra knowledge and precise annotation—only image-level weak annotation is all we take. Our model, termed PEAR-Net for Part Extracting and Age Recognition Network, consists of one Part Extracting (PE) agent for discriminative RoIs discovering and one Age Recognition (AR) agent for age assessment. Without precise supervision, the PE agent is designed to discover and extract RoIs fully automatically. Then the proposed RoIs are fed into AR agent for feature learning and age recognition. Furthermore, we utilize the self-consistency of RoIs to optimize PE agent to understand the part relation and select the most useful RoIs. With this self-supervised design, the PE agent and AR agent can reinforce each other mutually. To the best of our knowledge, this is the first end-to-end bone age assessment method which can discover RoIs automatically with only image-level annotation. We conduct extensive experiments on the public RSNA 2017 dataset and achieve state-of-the-art performance with MAE 3.99 months. Project is available at http://imcc.ustc.edu.cn/project/ssambaa/ .

Self-Supervised Attention Mechanism for Pediatric Bone Age Assessment With Efficient Weak Annotation

Bone Age Assessment (BAA) is a pediatric examination performed to determine the difference between children’s skeletal bone age and chronological age, the inconsistency between the two will often indicate either hormonal problems or abnormalities in the skeletal system maturity. Previous works to upgrade the tedious traditional techniques had failed to address the human expert inter-observer variability in order to significantly refine BAA evaluations. This paper proposes a deep learning method that detects and segments carpal bones as the region of interests within the left hand and wrist radiographs, and then feed the image data into a deep convolutional neural network. Tests are then made to determine whether it is more efficient to use full hand radiographs or segmented regions of interest, and also made comparisons with some CNN models. Evaluations show that the proposed method can dramatically increase the accuracy.

Sharon Chokuwa

Papers

Skeletal bone age assessments for young children based on regression convolutional neural networks.

Skeletal Bone Age Assessment Based on Deep Convolutional Neural Networks