结合高效通道注意模块的残差网络骨龄评估

Bone age assessment based on residual network combined with efficiency channel attention module

本文针对现有骨龄评估数据集数据规模小,样本分布不均匀以及现有方法评估准确度较低的问题,提出了一种新的结合高效通道注意模块的残差网络骨龄评估模型。通过结合深度残差网络和高效通道注意模块来提高卷积效率,并改进损失函数,缓解样本分布不均匀问题的影响;然后运用迁移学习的方法微调训练骨龄评估模型,提高模型训练效率;最后引入随机深度算法提高模型泛化能力。实验结果表明,该方法在RSNA数据集和DHA数据集上的平均绝对误差分别为4.69个月和5.98个月,当容忍度为12个月时,骨龄评估的准确率可以达到98.36%和94.88%,说明本文方法能够明显地提高骨龄评估的准确率,一定程度上缓解数据规模小和数据分布不均匀带来的影响。

In this paper,a new residual network bone age assessment model combined with high-efficiency channel attention module was proposed to address the problems of small data size,uneven sample distribution and low accuracy of existing methods for bone age assessment.Firstly,the deep residual network is selected as the basic convolutional neural network model,the convolution efficiency is improved by combining the high-efficiency channel attention module,and the loss function is improved to alleviate the influence of the uneven distribution of samples.Then,the model in this paper is pre-trained on the ImageNet data set to obtain the basic feature expression of the image.Finally,the open data set was fine-tuned with the random depth algorithm,and the accuracy of bone age assessment was obtained by cross validation.The results showed that the mean absolute error of this method on RSNA and DHA data sets was 4.69 months and 5.98 months,respectively.When the tolerance was 12 months,the accuracy of bone age assessment was 98.36%and 94.88%.This indicates that this paper can significantly improve the accuracy of bone age assessment,alleviate the impact of small data size and uneven data distribution to a certain extent,and restrain overfitting while improving network learning ability.