Aim:

Determination of bone age is an important method to skeletal maturity and growth potential. This paper proposes a determination bone age via X-ray image recognition to obtain a better identification effect by comparison with state-of-the-art techniques.

Methods

: The proposed approach comprises two steps: the feature extraction and classification method. The feature extraction utilizes depth neural network to study the features of X-ray image, and the Local Binary Patterns (LBP) features and Glutamate cysteine ligase modifier subunit (GCLM) features in the image are extracted. Then, the classification method base on support vector machine is used to classify the features.

Results:

The experimental results show that the average absolute error of bone age assessment model based on multi-dimensional data feature fusion is 0.455, which is superior to the traditional method and support vector machine method. Because the model is based on feature extraction of deep neural network, it shows that the feature extraction method based on deep neural network can extract feature information better than traditional image analysis method.

Conclusion:

Compared with the traditional feature extraction method, the feature extraction based on deep convolution neural network has better performance in the bone age regression model. Combining population and gender information, the accuracy of bone age prediction based on image can be further improved.

中文翻译：

---

基于深度学习方法的X射线图像自动特征提取以确定骨龄

目标：

确定骨龄是骨骼成熟和生长潜力的重要方法。本文提出了一种通过X射线图像识别来确定骨龄的方法，以通过与最新技术进行比较来获得更好的识别效果。

方法

：建议的方法包括两个步骤：特征提取和分类方法。特征提取利用深度神经网络研究X射线图像的特征，并提取图像中的本地二元模式（LBP）特征和谷氨酸半胱氨酸连接酶修饰子亚基（GCLM）特征。然后，采用基于支持向量机的分类方法对特征进行分类。

结果：

实验结果表明，基于多维数据特征融合的骨龄评估模型的平均绝对误差为0.455，优于传统方法和支持向量机方法。由于该模型基于深度神经网络的特征提取，因此表明基于深度神经网络的特征提取方法比传统的图像分析方法能够更好地提取特征信息。

结论：

与传统特征提取方法相比，基于深度卷积神经网络的特征提取在骨龄回归模型中具有更好的性能。结合人口和性别信息，可以进一步提高基于图像的骨龄预测的准确性。