计算机技术与发展2025，Vol.35Issue(9)：23-29,7.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0102
基于双尺度特征交互网络的CT影像分类应用
Application of Dual-scale Feature Interaction Network for CT Imaging Classification
摘要
Abstract
A CT image assessment approach for upper urinary tract urinary tract infections based on a dual-scale feature interaction network was discussed in depth.A retrospective analysis was conducted on the CT scans of upper urinary tract infections patients admitted to Fudan University Affiliated Minhang Hospital from December 2023 to April 2024,and a dataset was built for deep learning training.To improve the recognition ability,a network model DFI-Net based on dual-scale feature interaction was proposed,which integrates the advantages of visual Transformer structure on the basis of ResNet model,and introduces feature interaction mechanism to ef-fectively improve the models feature extraction and recognition ability.In addition,the model performance was evaluated by accuracy,precision,sensitivity,etc.,and compared with the classification models ResNet50,EfficientNet-B0,ConvNext-T,Swin-T,and Vision Transformer.Furthermore,an ablation experiment was designed to verify the effectiveness of the feature fusion module.We compared five network models and the accuracy rates of ResNet50,EfficientNet-B0,ConvNext-T,Vision Transformer,and Swin-T were 79.955％,78.842％,83.206％,78.485％,and 83.195％,respectively.The accuracy rate of the DFI-Net proposed is 85.582％,and the specificity and sensitivity are balanced at a good level,with 88.261％and 85.093％,respectively.The proposed model can effectively improve the recognition of urinary infection CT images.关键词
图像处理/卷积神经网络/注意力机制/特征交互/泌尿感染Key words
image processing/convolutional neural network/attention mechanism/feature interaction/urinary tract infections分类
信息技术与安全科学引用本文复制引用
孔佑进,陈锘,尹冰德,李响,何昶,陈建刚..基于双尺度特征交互网络的CT影像分类应用[J].计算机技术与发展,2025,35(9):23-29,7.基金项目
国家自然科学基金项目(82151318) （82151318）
上海市科委(22DZ2229004,22JC1403603,21Y11902500) （22DZ2229004,22JC1403603,21Y11902500）
浙江省科技厅"尖兵领雁"重点攻关研发项目(2024C03240) （2024C03240）
