clc clears all the text from the Command Window, resulting in a clear screen.
clear removes all variables from the current workspace, releasing them from system memory.
close all closes all figures.
clc;
clear all;
close all;
fullfile(filepart1,...,filepartN) builds a full file specification from the specified folder and file
names.
imageFolder = fullfile("../br/br/Testing/");
imageDatastore creates a datastore from the collection of image data specified by location.
imds = imageDatastore(imageFolder, 'LabelSource', 'foldernames', 'IncludeSubfolders',tr
countEachLabel counts the number of times each unique label occurs in the datastore.
tb1 = countEachLabel(imds)
tb1 = 4×2 table
Label Count
1 glioma 300
2 meningioma 306
3 notumor 405
4 pituitary 300
minSetCount = min(tb1{:,2});
maxNumImages = 300;
minSetCount = min(maxNumImages,minSetCount);
splitEachLabel randomly assigns the specified proportion of files from each label to the new datastores.
imds = splitEachLabel(imds,minSetCount,'randomized');
tb2 = countEachLabel(imds)
tb2 = 4×2 table
Label Count
1 glioma 300
2 meningioma 300
3 notumor 300
1
Label Count
4 pituitary 300
Spliting the imds datastore into imdstrain datastore with 70% of imds datastore and imdtest datastore with
30% of imds datastore.
[imdstrain,imdstest] = splitEachLabel(imds,0.7,'randomized');
imageSize=[256 256 1];
An augmented image datastore transforms training dataset with optional preprocessing such as resizing,
grayconversion.
datastore = augmentedImageDatastore(imageSize,imdstrain,"ColorPreprocessing","rgb2gray"
Convolutional Neural Network:
An image input layer inputs 2-D images to a network.
A 2-D convolutional layer applies sliding convolutional filters to 2-D input.
A batch normalization layer normalizes a mini-batch of data across all observations for each channel
independently.
A ReLU layer performs a threshold operation to each element of the input, where any value less than zero is set
to zero.
A 2-D max pooling layer performs downsampling by dividing the input into rectangular pooling regions, then
computing the maximum of each region.
A fully connected layer multiplies the input by a weight matrix and then adds a bias vector.
A softmax layer applies a softmax function to the input.
A classification layer computes the cross-entropy loss for classification and weighted classification tasks with
mutually exclusive classes.
layers = [ ...
imageInputLayer(imageSize,'Name','input')
convolution2dLayer(3,8,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)
convolution2dLayer(3,16,'Padding','same')
batchNormalizationLayer
2
reluLayer
maxPooling2dLayer(2,'Stride',2)
convolution2dLayer(3,32,'Padding','same')
batchNormalizationLayer
reluLayer
maxPooling2dLayer(2,'Stride',2)
convolution2dLayer(3,64,'Padding','same')
batchNormalizationLayer
reluLayer
fullyConnectedLayer(32)
reluLayer
fullyConnectedLayer(16)
reluLayer
fullyConnectedLayer(8)
fullyConnectedLayer(4)
softmaxLayer
classificationLayer ];
A layer graph specifies the architecture of a deep learning network
lgraph = layerGraph(layers);
figure
plot(lgraph)
3
trainingOptions returns training options with additional options specified by one or more name-value
arguments.
options = trainingOptions('sgdm', ...
'MaxEpochs',100,...
'InitialLearnRate',1e-4, ...
'Verbose',true, ...
'Plots','training-progress');
A trainNetwork trains CNN with train datastore and training options.
net = trainNetwork(datastore,layers,options);
Training on single CPU.
Initializing input data normalization.
|========================================================================================|
| Epoch | Iteration | Time Elapsed | Mini-batch | Mini-batch | Base Learning |
| | | (hh:mm:ss) | Accuracy | Loss | Rate |
|========================================================================================|
| 1 | 1 | 00:00:15 | 28.12% | 1.5097 | 1.0000e-04 |
| 9 | 50 | 00:02:10 | 85.16% | 0.5273 | 1.0000e-04 |
| 17 | 100 | 00:04:05 | 89.84% | 0.3735 | 1.0000e-04 |
| 25 | 150 | 00:06:04 | 96.09% | 0.2077 | 1.0000e-04 |
| 34 | 200 | 00:08:01 | 99.22% | 0.1276 | 1.0000e-04 |
| 42 | 250 | 00:09:52 | 97.66% | 0.1500 | 1.0000e-04 |
| 50 | 300 | 00:11:43 | 98.44% | 0.0965 | 1.0000e-04 |
| 59 | 350 | 00:13:36 | 100.00% | 0.0650 | 1.0000e-04 |
| 67 | 400 | 00:15:29 | 99.22% | 0.0830 | 1.0000e-04 |
| 75 | 450 | 00:17:22 | 99.22% | 0.0546 | 1.0000e-04 |
| 84 | 500 | 00:19:20 | 100.00% | 0.0401 | 1.0000e-04 |
| 92 | 550 | 00:21:09 | 100.00% | 0.0548 | 1.0000e-04 |
| 100 | 600 | 00:22:59 | 100.00% | 0.0356 | 1.0000e-04 |
|========================================================================================|
Training finished: Max epochs completed.
Save the net model as net.mat for future use.
save('net',"net")
An augmented image datastore transforms test dataset with optional preprocessing such as resizing,
grayconversion.
imdsTest_rsz = augmentedImageDatastore(imageSize,imdstest,'ColorPreprocessing','rgb2gra
imdsTest_rsz =
augmentedImageDatastore with properties:
NumObservations: 360
Files: {360×1 cell}
AlternateFileSystemRoots: {}
MiniBatchSize: 128
DataAugmentation: 'none'
ColorPreprocessing: 'rgb2gray'
OutputSize: [256 256]
4
OutputSizeMode: 'resize'
DispatchInBackground: 0
load the model which was stored as net.mat
load net
Classify test dataawt using trained deep learning neural network
YPred = classify(net,imdsTest_rsz);
confusionchart(trueLabels,predictedLabels) creates a confusion matrix chart from true labels
trueLabels and predicted labels predictedLabels and returns a ConfusionMatrixChart object.
cm = confusionchart(imdstest.Labels,YPred);
cm.Title = 'Brain Tumor Classification Using CNN';
Accuracy:
accuracy = sum(imdstest.Labels == YPred,'all')/numel(YPred)
accuracy = 0.7944
5
6
