基于深度学习的眼科超声图像智能辅助诊断方法研究

Research on intelligent assisted diagnosis method of ophthalmic ultrasound images based on deep learning

目的:提出一种基于深度学习的眼科超声图像智能辅助诊断方法,以为眼科疾病的智能化临床诊断提供辅助分析。方法:首先,建立眼科超声图像数据集,在数据集上完成Res Net-50、Dense Net-121和Mobile Net 3种网络模型的训练,并通过准确率、宏平均精确度、宏平均敏感度、宏平均F_(1)值以及AUC值评估模型的分类性能,通过模型文件大小和检测测试集图像的平均用时评估模型的实时性能,选出最合适的眼科超声图像智能辅助诊断模型。然后,通过梯度加权类激活映射方法实现图像异常组织区域的热力图可解释性分析。最后,基于Py Qt工具包完成眼科超声图像智能辅助诊断软件开发。结果:与Res Net-50和Dense Net-121模型相比,Mobile Net模型整体性能较优,准确率、宏平均精确度、宏平均敏感度、宏平均F_(1)值、AUC值分别为0.9485、0.9266、0.9305、0.9281和0.9958,模型文件大小和检测平均用时分别为56.69 Mi B和0.0890 s。设计的眼科智能超声辅助诊断软件能够实现眼科超声图像的智能识别和热力图可解释性分析。结论:基于深度学习的眼科超声图像智能辅助诊断方法可以准确地识别眼科超声图像,能够满足眼科疾病临床诊断需求。

Objective To propose a deep learning-based intelligent assisted diagnosis method for ophthalmic ultrasound images to facilitate intelligent diagnosis of ophthalmic diseases.Methods An ophthalmic ultrasound image dataset was established,which was used for the training of three network models of Res Net-50,Dense Net-121 and Mobile Net.The three models underwent classification performance evaluation in terms of accuracy,macro-average precision,macro-average sensitivity,macro-average F_(1)score and AUC value,and real-time performance evaluation from the aspects of model file size and the average time consumed for detecting the images in the test set,so as to determine the optimal intelligent assisted diagnostic model for ophthalmic ultrasound images.Then the gradient weighted class activation mapping method was used to achieve interpretable analysis of the heat map of the abnormally organized regions of the image.Finally,the ophthalmic ultrasound image intelligent assisted diagnosis software development was completed based on Py Qt toolkit.Results Mobile Net model gained advantages over Res Net-50 and Dense Net-121 models,which had the accuracy,macro-average precision,macro-average sensitivity,macro-average F_1score,AUC value,model file size and the average time for detection being 0.948 5,0.926 6,0.930 5,0.928 1,0.995 8,56.69 Mi B and 0.089 0 s,respectively.The ophthalmic ultrasound image intelligent assisted diagnosis software realized intelligent recognition and heat map interpretable analysis of ophthalmic ultrasound images.Conclusion The deep learning-based ophthalmic ultrasound image intelligent assisted diagnosis method can accurately identify ophthalmic ultrasound images,and can meet clinical diagnosis requirements of ophthalmic diseases.