自我学习归纳|Matlab+UNet+LIDC 32行代码训练自己的肺结节数据集深度学习|matlab

版本：Matlab2021b，需要安装deeplearningToolBox
数据集是LIDC-IDRI，我已经事先切割成64*64大小的PNG了，数据集下载链接：lidc数据集增强，便于使用-互联网文档类资源-CSDN文库
话不多说直接上matlab代码

%% 数据集导入 imageDir='G:\MyLIDC\output'; trainImgLoc=fullfile(imageDir,'train','Image'); labelImgLoc = fullfile(imageDir,'train','Mask'); imds = datastore(trainImgLoc, "IncludeSubfolders",true,"Type","image",'ReadFcn',@myReader); classNames = ["background","nodule"]; pixelLabelID = [0 255]; pxds = pixelLabelDatastore(labelImgLoc,classNames,pixelLabelID, ... "IncludeSubfolders",true,'ReadFcn',@myReader); dsTrain = combine(imds,pxds); %% 网络参数配置 lgraph=myUnet(); initialLearningRate = 5e-4; maxEpochs = 1; minibatchSize = 20; l2reg = 0.0001; options = trainingOptions('adam',... 'InitialLearnRate',initialLearningRate, ... 'L2Regularization',l2reg,... 'MaxEpochs',maxEpochs,... 'MiniBatchSize',minibatchSize,... 'LearnRateSchedule','piecewise',... 'Shuffle','every-epoch',... 'GradientThresholdMethod','l2norm',... 'GradientThreshold',0.05, ... 'Plots','training-progress', ... 'VerboseFrequency',20); [unet,info] = trainNetwork(dsTrain,lgraph_1,options); modelDateTime = string(datetime('now','Format',"yyyy-MM-dd-HH-mm-ss")); save(strcat("Lidc-Unet-",modelDateTime,"-Epoch-",num2str(maxEpochs),".mat"),'unet');

训练一个epoch已经效果还ok了

自我学习归纳|Matlab+UNet+LIDC 32行代码训练自己的肺结节数据集

文章图片

再看一下模型预测图片的代码

%% 查看模型预测结果 testimg=imread('G:\MyLIDC\output\test\Image\LIDC-IDRI-0875\0875_NI001_slice004.png'); testmask=imread('G:\MyLIDC\output\test\Mask\LIDC-IDRI-0875\0875_MA001_slice004.png'); testimg=rgb2gray(testimg); testmask=rgb2gray(testmask); categoricalMask=categorical(testmask,pixelLabelID,classNames); categoricalImage=semanticseg(testimg, unet); predict=uint8(categoricalImage=='nodule')*255; img_predicmask = labeloverlay(testimg,categoricalImage,'Transparency',0.8); img_gtmask = labeloverlay(testimg,categoricalMask,'Transparency',0.8); figure subplot(2,3,1) imshow(testimg) title("原始测试图像") subplot(2,3,2) imshow(predict) title("测试图像预测输出") subplot(2,3,3) imshow(img_predicmask) title("测试图像预测覆盖") subplot(2,3,5) imshow(testmask) title("测试图像GroundTruth") subplot(2,3,6) imshow(img_gtmask) title("测试图像GroundTruth覆盖")

效果图如下