融合中华-05注意力与叠加纹理构建残差网络智能模型评估骨龄

目的 融合中华-05注意力机制与多层纹理叠加建立残差网络模型,观察其评估骨龄的价值。方法 通过组合寻优引入多层叠加纹理增强处理层,以更少信息量更好地表征手骨X线片全局特征,减少杂质信息干扰并释放算力。设计中华-05空间注意力机制,引入我国人群手骨发育标准,使模型智能化聚焦ROI,并自动定位、学习图像关键信息。建立50层深度残差网络,集成融合叠加增强层与注意力机制,观察其评估骨龄的价值。结果 构建的混合式改进的深度残差网络模型ZH05-DL-ResNet50评估骨龄准确率达98.14%,平均绝对误差为0.312岁。结论 成功建立了我国人群中华-05注意力与叠加纹理残差网络智能模型,用于评估骨龄准确率高。

Residual network intelligent bone age assessment model established base on China-05 attention and overlay texture

Objective To establish a residual network model combining China-05 attention mechanism and multi-layer texture overlay, and to explore its value for assessing bone age. Methods Through combinational optimization to better characterize the global features of hand bone X-ray images with less information, a multi-layer overlay texture enhancement processing layer was introduced to reduce the interference of impurity information and release computing power. The China-05 spatial attention mechanism was designed, and the standard of human hand bone development in China was introduced to make the model intelligently focusing on ROI and automatically locating and learning the key information of images. The 50-layer deep residual network was built to integrate the overlay enhancement layer and the attention mechanism, and its value for assessing bone age was evaluated. Results The hybrid improved deep residual network model ZH05-DL-ResNet50 was successfully established, with the accuracy of 98.14% for assessing bone age, and the mean absolute error of 0.312 year. Conclusion A residual network intelligent bone age assessment model based on China-05 attention and overlay texture was successfully established, which was helpful to improving the accuracy of assessing bone age in Chinese population.