基于深度学习的乳腺癌淋巴结转移诊断研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）16-0150-05

Diagnosis Research on Lymph Node Metastasis in Breast Cancer Based on Deep Learning

ZHU Jiawei， JIN Miao， YU Tong （Clinical College，Anhui MedicalUniversity，Hefei 23oo31， China）

Abstract： Breast cancer is one of the most common malignant tumors in women worldwide.Accurate detection of lymph nodemetastasisisofgreatsignifcanceforitsdiagnosisandtreatment.Inordertoimprovethedetectionacuracyofmetastatic cancercellsinpathologicalimages，aneficientNeuralNetworkArchitectureisdesignedandoptimizedbasedontheNeural ArchitectureSearchalgorithm.ThePatchCamelyondatasetisusedinthesudyandthetrainingsetisandomlytakenbacktofo threesub-trainingsets.TeNuralNtworkArchitectureistaindindependentlyintepredefinedsearchspace，nditsweightsare optimizedthroughscondarytraining.Finallytethree modelsareintegrated inparaleltoimprovetheoverallperformance.The experimentalresultsshowthattheesignednetworkissuperiortothetraditionalResNet-18，ResNet-34andVGG-16inindicatos suchas AUC，andhasasmaller modelsize.Inthe network architecture search stage，theaverage acuracyofasingle model is 73.34% ，and the AUC is 75.53% .After optimization and integration，the accuracy of the final model is increased to 90.12% ，the AUC is increased to 91.3% ，and the model size is only 30.2MB ，which has the advantages of high efficiencyand lightweight.

Keywords： breast cancer; lymph node metastasis detection; Neural Architecture Search; Ensemble Learming

0 引言

乳腺癌是全球女性中最常见的恶性肿瘤之一，其发病率和死亡率在许多国家均居于前列[1。（剩余6788字）