基于深度学习的颈动脉粥样硬化斑块成分识别

Carotid Atherosclerosis Plaque Recognition Algorithm based on Deep Learning

为实现颈动脉粥样硬化斑块成分的自动分割,提出一种基于深度学习的分割方法。针对颈动脉粥样硬化斑块成分特征复杂,医生手动分割提取费时且有误差等问题,使用基于Deeplab V3+网络的算法对颈动脉粥样硬化斑块成分进行自动分割来识别斑块成分。首先对已标识斑块成分的超声图像数据文件进行数据预处理、数据扩充以及感兴趣区域提取等操作,建立颈动脉粥样硬化斑块数据集,将数据集放到Deeplab V3+网络中训练和测试。针对Deeplab V3+网络复杂且参数量大的问题,在实验中结合MobileNet网络优点对原始网络进行优化。对比实验结果表明,优化后的模型在减少参数降低计算量的同时可以对颈动脉粥样硬化斑块成分进行有效分割与识别。

In order to achieve the automatic segmentation of carotid atherosclerotic plaque components,a deep learning based segmentation method was proposed.Firstly,the carotid atherosclerotic plaque data set was established through data preprocessing,data expansion and extraction of regions of interest,and the data set was trained and tested in Deeplab V3+network.To solve the problem of Deeplab V3+network with complex and large number of parameters,the original network was optimized by combining the advantages of MobileNet network in the experiment.According to the comparative experimental results,the optimized model can effectively segment and identify the components of carotid atherosclerosis plaque while reducing the amount of parameters and computation.