基于卷积神经网络的X图像骨龄评估方法

Convolutional neural network-based method for bone age assessment in X-ray image

针对传统方法中骨骺评估区域多、评估结果对医生的依赖性强、评估准确率低等问题,在TW3-C法基础上提出一种改进的骨龄评估方法。根据中国儿童骨骼发育特点,利用卷积神经网络对评估区域进行精减和分类,将传统的13个骨骼评估区域精减至10个,并改进等级计分法。试验结果显示,在1岁误差范围内,该方法将骨龄的预测值准确率提升至男性94.42%、女性93.64%,平均绝对误差为男性0.414 3岁、女性0.428 6岁,与典型的骨龄评估方法相比,准确率得到显著提高。

To solve the problems in traditional methods such as lots of assessment areas in the epiphysis,high dependence of assessment results on doctors and low assessment accuracy,an improved bone age assessment method based on TW3-C method is proposed.According to the characteristics of Chinese children’s bone development,convolutional neural network is used to refine and classify the assessment areas,reducing the traditional 13 bone assessment areas to 10,and the grade scoring method is also improved.The experimental results show that within the error range of 1-year-old,the proposed method improves the accuracy of the predicted value of bone age to 94.42% for men and 93.64% for women.The average absolute error is 0.414 3 years old for men and 0.428 6 years old for women.Compared with that of typical bone age assessment method,the accuracy of the proposed method for bone age assessment is significantly improved.