Détection des maladies thoraciques à l'aide de CNN

import os
import cv2
import numpy as np
from tensorflow.keras import layers, optimizers
from tensorflow.keras.applications.resnet50 import ResNet50
from tensorflow.keras.layers import Input, Dense, Flatten, AveragePooling2D, Dropout
from tensorflow.keras.models import Model
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
import matplotlib.pyplot as plt
import seaborn as sns
 
XRay_Directory = r"Dataset"
 
os.listdir(XRay_Directory)
 
image_generator = ImageDataGenerator(rescale = 1./255, validation_split= 0.2)
 
train_generator = image_generator.flow_from_directory(batch_size = 40, directory= XRay_Directory, shuffle= True, target_size=(256,256), class_mode= 'categorical', subset="training")
 
validation_generator = image_generator.flow_from_directory(batch_size = 40, directory= XRay_Directory, shuffle= True, target_size=(256,256), class_mode = 'categorical', subset="validation")
 
train_images, train_labels = next(train_generator)
 
print(train_images.shape)
 
print(train_labels.shape)
 
label_names = {0 : 'Covid-19', 1 : 'Normal' , 2: 'Viral Pneumonia', 3 : 'Bacterial Pneumonia'}
 
L = 6
W = 6
 
fig, axes = plt.subplots(L, W, figsize = (12, 12))
axes = axes.ravel()
 
for i in np.arange(0, L * W):
    axes[i].imshow(train_images[i])
    axes[i].set_title(label_names[np.argmax(train_labels[i])])
    axes[i].axis('off')
 
plt.subplots_adjust(wspace = 0.5)
plt.show()
 
basemodel = ResNet50(weights = 'imagenet', include_top = False, input_tensor = Input(shape=(256,256,3)))
 
basemodel.summary()
 
for layer in basemodel.layers[:-10]:
  layers.trainable = False
 
#Use pretrained model
headmodel = basemodel.output
#Add average pooling layer
headmodel = AveragePooling2D(pool_size = (4,4))(headmodel)
#Add flattening layer
headmodel = Flatten(name= 'flatten')(headmodel)
# Add dense layer with 256 neurons and apply relu activation function
headmodel = Dense(256, activation = "relu")(headmodel)
#Drop 30% of the neurons to avoid Codependency
headmodel = Dropout(0.3)(headmodel)
# Add dense layer with 256 neurons and apply relu activation function
headmodel = Dense(256, activation = "relu")(headmodel)
#Drop 30% of the neurons to avoid Codependency
headmodel = Dropout(0.3)(headmodel)
# Add dense layer with 256 neurons and apply relu activation function
headmodel = Dense(256, activation = "relu")(headmodel)
#Drop 30% of the neurons to avoid Codependency
headmodel = Dropout(0.3)(headmodel)
#Add final dense layer with 4 outputs and apply softmax activation function
headmodel = Dense(4, activation = 'softmax')(headmodel)
 
model = Model(inputs = basemodel.input, outputs = headmodel)
 
model.compile(loss = 'categorical_crossentropy', optimizer=optimizers.RMSprop(lr = 1e-4, decay = 1e-6), metrics= ["accuracy"])
 
earlystopping = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=20)
 
checkpointer = ModelCheckpoint(filepath="weights.hdf5", verbose=1, save_best_only=True)
 
train_generator = image_generator.flow_from_directory(batch_size = 4, directory= XRay_Directory, shuffle= True, target_size=(256,256), class_mode= 'categorical', subset="training")
 
val_generator = image_generator.flow_from_directory(batch_size = 4, directory= XRay_Directory, shuffle= True, target_size=(256,256), class_mode= 'categorical', subset="validation")
 
history = model.fit(train_generator, steps_per_epoch= train_generator.n // 4, epochs = 25, validation_data= val_generator, validation_steps= val_generator.n // 4, callbacks=[checkpointer, earlystopping])
 
print(history.history.keys())
 
plt.plot(history.history['acc'])
plt.plot(history.history['loss'])
plt.xlabel('Epochs')
plt.ylabel('Training accuracy and loss')
 
plt.show()
 
plt.plot(history.history['val_loss'])
plt.xlabel('Epochs')
plt.ylabel('Validation loss')
 
plt.show()
 
plt.plot(history.history['val_acc'])
plt.xlabel('Epochs')
plt.ylabel('Validation accuracy')
 
plt.show()
 
test_directory =r"Test"
 
test_gen=ImageDataGenerator(rescale=1./255)
test_generator=test_gen.flow_from_directory(batch_size=40,directory=test_directory,shuffle=True,target_size=(256,256),class_mode='categorical')
 
evaluate=model.evaluate(test_generator,steps=test_generator.n //4, verbose=1)
 
print('Accuracy Test: {}'.format(evaluate[1]))
 
from sklearn.metrics import confusion_matrix, classification_report, accuracy_score
 
prediction = []
original = []
image = []
 
for i in range(len(os.listdir(test_directory))):
  for item in os.listdir(os.path.join(test_directory,str(i))):
    img= cv2.imread(os.path.join(test_directory,str(i),item))
    img = cv2.resize(img,(256,256))
    image.append(img)
    img = img / 255
    img = img.reshape(-1,256,256,3)
    predict = model.predict(img)
    predict = np.argmax(predict)
    prediction.append(predict)
    original.append(i)
 
score=accuracy_score(original,prediction)
print('Test Accuracy: {}'.format(score))
 
print(classification_report(np.asarray(original),np.asarray(prediction)))
 
cm = confusion_matrix(np.asarray(original), np.asarray(prediction))
ax = plt.subplot()
sns.heatmap(cm, annot = True, ax = ax)
 
ax.set_xlabel('Predicted')
ax.set_ylabel('Original')
ax.set_title('Confusion_matrix')
 
#Development of a 25 picture grid (5x5)
L = 5
W = 5
 
fig, axes = plt.subplots(L, W, figsize = (12, 12))
axes = axes.ravel()
 
for i in np.arange(0, L * W):
    axes[i].imshow(image[i])
    axes[i].set_title('Guess={}\nTrue={}'.format(str(label_names[prediction[i]]), str(label_names[original[i]])))
    axes[i].axis('off')
 
plt.subplots_adjust(wspace = 1.2)