基于弱监督细粒度深度网络的骨龄自动评估方法
首发时间:2019-09-02
摘要:人类的骨骼发育在不同基于弱监督细粒度深度网络的骨龄自动评估方法阶段呈现出不同的特点,青少年骨龄评估能较准确地反映个体的生长发育水平和成熟程度,它不仅可以确定儿童的生物学年龄,而且还可以通过骨龄及早了解儿童的生长发育潜力以及性成熟的趋势。本文利用一种新的基于细粒度图像识别的深度卷积神经网络进行骨龄评估,该网络在手骨图像识别的过程中可以自动定位目标的复杂信息区域并提取其局部特征,将提取到的局部特征与全局特征进行结合,输出最终的骨龄评估结果。该方法不需要借助任何外部标注信息,即可实现端到端的骨龄评估,大大提高了骨龄评估的速度和准确度。通过实验表明,在本研究所使用的数据集上实现了最高82.5%的识别准确率以及0.35岁的平均绝对误差。
关键词: 模式识别:自动骨龄评估:深度学习:卷积神经网络:细粒度图像
For information in English, please click here
Automatic evaluation method of bone age based on weakly supervised fine-grained network
Abstract:Human bone development shows different characteristics at different stages. Adolescent bone age assessment can accurately reflect the level of individual growth and development and maturity, it can not only determine the biological age of children, It is also possible to understand the growth and development potential of children and the trend of sexual maturity early through bone age. In this paper, a new depth convolution neural network based on fine-grained image recognition is used to evaluate bone age. In the process of hand bone image recognition, the network can automatically locate the complex information region of the target and extract its local features. The extracted local features are combined with the global features, and the final bone age evaluation results are output. This method does not need any external labeling information to realize the end-to-end bone age evaluation, which greatly improves the speed and accuracy of bone age evaluation. The experimental results show that the maximum recognition accuracy of 82.5% and the Mean Absolute Error of 0.35 years old are achieved on the data set used in this study.
Keywords: Pattern recognition Automatic bone age assessment Deep learning Convolutional neural network Fine-grained image
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于弱监督细粒度深度网络的骨龄自动评估方法
评论
全部评论0/1000