面向免疫固定电泳分类的输入增强双尺度网络模型OA

To address the clinical challenges of immunofixation electrophoresis images,such as background fluctuations,noise interference,and faint bands leading to missed detections and misdiagnoses,we pro-pose a DGDB-IE2FNet model.The model features a dynamically gated dual-branch residual block and an IFE input enhancement dual-scale branch.The former adaptively fuses complementary convolutional branches,while the latter suppresses background noise and enhances directional information of bands.We formulate the classification of positive/negative results for ELP,G,A,M,K,and L lanes as a multi-label prediction task and conduct comparative experiments with several mainstream models using 20 000 real-world cases.Results show that the DGDB-IE2FNet achieves an accuracy of 99.63%,negative predictive value of 99.74%,sensitivity of 98.07%,and F1 score of 98.56%on the test set,outperforming other compared models in overall prediction performance.