融合DW卷积与SE注意力的ResNet50模型用于MRI预测阿尔茨海默症

打开文本图片集

中图分类号：TP391 文献标识码：A 文章编号：1006-8228（2026）02-28-07

Abstract：ThispaperproposesanimprovedResNet50deeplearmingmodel（ResNet50-DW-SEmodel）forrecognizingnuclear magneticresonanceimages（MRI）topredictAlzheimersdisease.TheimprovementstotheResNet5Omodelarereflected inits Botleneckblocks.Ononeh，depthwiseseparableconvolutions（DW）areadoptedtoreplacetradionalconvolutions，achieving modellightweighting，thatis，reducingthenumbermodelparameterstoimprovethecomputationaleficiencythemodel.On theotherhtheSqueze--Excitation（SE）atentionmoduleisintroducedtoenhancetheaccuracythemodelinimage recognition.Thepredictionprocessthemodelisasfollows：first，collctnuclearmagneticresonanceimagestoestablisha datasetconductpreprocessing;secondprerainthemodeldividethedatasetintoarainingsetndatestset;thentrain themodelwiththetrainingsettoevaluateitscomputationaleficiency；finalltestthetrainedmodelwiththetestsettaking recall，precisionF-scoreastheevaluationmetrics.ExperimentssowthatcomparedwiththeriginalResNet50theReset50- DW-SE model reduces the number parameters by 45.2% ， the average values recall，，precision F1-score all reach 0.99 which are 1.25% 2.5% 2.25% higher than those the original ResNet5O respectively.

KeyWords：Alzheimer'sDisease;DeepLearning;BotleneckBlock；DepthwiseSeparableConvolution;Atention Mechanism

0引言

阿尔茨海默症是一种严重危害人们健康的神经系统变性疾病，该疾病会导致患者记忆力减退、认知功能障碍及行为异常。（剩余10189字）