To reduce the misdiagnosis or missed diagnosis caused by subjective judgment when doctors analyze lung CT im-ages,the application effects of four convolutional neural network models,namely ResNet-18,ResNet-34,ResNet-50,and VGG16,in lung image classification tasks were compared.Ultimately,ResNet-50 was determined as the base training model,and transfer learning was used to optimize this model to further enhance its performance.The experimental results show that the improved ResNet-50 model significantly increases the accuracy of pulmonary nodule detection and classification,reaching 89.55%,which is significantly superior to 52.27%achieved by VGG16.Additionally,a cloud-based pulmonary image visualization diagnostic system was developed using Streamlit.Users can upload lung CT images through the interface,and the system will automatically detect pul-monary nodules and output classification results.This system is easy to operate and assists doctors in making quick diagnoses and patients in timely detection of their conditions.
关键词
迁移学习/ResNet-50/肺结节检测/肺部CT图像/Streamlit/云端系统
Key words
transfer learning/ResNet-50/Pulmonary nodule detection/Lung CT images/Streamlit/cloud system
万方数据