基于深度神经网络的颈动脉斑块超声图像诊断技术研究

目的 构建颈动脉斑块超声图像数据集并探讨深度学习技术对颈动脉斑块自动分类诊断的应用价值。方法 首先采集254例患者和354例正常人的颈部动脉超声图像,每例采集2幅图像,构建共包含1216幅图像的颈动脉超声图像数据集;然后,基于已构建的颈动脉超声图像数据集对传统的HOG+SVM方法和14种不同结构的深度神经网络模型进行训练;最后,通过三个量化指标(分类精确率、召回率和F1值)确定现有的颈动脉斑块超声图像分类性能最好的深度神经网络模型。 结果 通过综合比较15种不同的颈动脉斑块超声图像分类方法,得出性能最好的模型是深度残差网络模型ResNet50,其精确率、召回率和F1值分别为97.36%、97.32%和97.34%。 结论 本文通过数据集构建、模型选择、模型训练和测试验证了深度学习技术在颈动脉斑块超声图像自动诊断应用中的有效性,其中深度残差网络模型ResNet50对颈动脉超声图像能进行高准确度自动分类。

Research on Deep neural network-based automatic diagnosis for carotid plaque using ultrasound images

Purpose To verify the effectiveness of deep learning in ultrasonic diagnosis, we construct the ultrasound image dataset of the carotid and test 15 different classification models for carotid plaque diagnosis. Methods Firstly, 254 patients and 354 normal persons" carotid ultrasound images were collected to construct a carotid ultrasound image dataset. Two images were collected for each case, so the constructed dataset contains 1216 images in total. Then, traditional method HOG+SVM and 14 different deep neural network models were trained based on our proposed dataset. Finally, the best network model is determined based on three evaluation indexes: precision, recall and F1 score. Results By comparing 15 different ultrasonic image classification methods for carotid plaque diagnosis, it is concluded that the depth residual network model ResNet50 has the best performance, and its accuracy, recall and F1 values are 97.36%, 97.32% and 97.34%, respectively. Conclusion ResNet50 can classify and recognize carotid plaque images with the highest accuracy, which verifies the effectiveness of deep learning in the application of automatic diagnosis by using ultrasound images.