深度學習與醫學圖像處理 案例學習3——Covid19肺炎檢測（CXR圖像）

內容引用自https://www.kaggle.com/mahmoudreda55/x-ray-covid19-95

數據集中的影像模式為胸部X線，共有COVID-19數據1200例，1341例正常人影像和1345例病毒性肺炎影像。

#導入數據包

import kerastuner as kt
from tensorflow import keras
import tensorflow as tf
from kerastuner.tuners import RandomSearch
from kerastuner.engine.hyperparameters import HyperParameter as hp
from keras.layers import Dense,Dropout,Activation,Add,MaxPooling2D,Conv2D,Flatten
from keras.models import Sequential 
from keras.preprocessing.image import ImageDataGenerator
import numpy as np
import matplotlib.pyplot as plt
from keras.applications import VGG19
from keras import layers
from keras.preprocessing import image
import matplotlib.pyplot as plt
import seaborn as sns
import os
import numpy as np
import pandas as pd

#讀入數據

base_path='/kaggle/input/covid19-radiography-database/COVID-19 Radiography Database'

train_data = tf.keras.preprocessing.image_dataset_from_directory(
    base_path,
    labels="inferred",
    label_mode="int",
    class_names=None,
    color_mode="rgb",
    batch_size=32,
    image_size=(244, 244),
    shuffle=True,
    seed=123,
    validation_split=0.2,
    subset="training",
    interpolation="bilinear",
    follow_links=False,
)
val_data = tf.keras.preprocessing.image_dataset_from_directory(
    base_path,
    labels="inferred",
    label_mode="int",
    class_names=None,
    color_mode="rgb",
    batch_size=32,
    image_size=(244, 244),
    shuffle=True,
    seed=123,
    validation_split=0.2,
    subset="validation",
    interpolation="bilinear",
    follow_links=False,
)
#該方法適合整理數據，代碼比較簡單

#分類
class_names =  ["COVID-19","NORMAL","Viral Pneumonia"]
for i in range(len(class_names)):
    print(class_names[i] ," " , i)

  

#顯示一副病毒性肺炎的圖片
image_path=os.path.join(base_path,'Viral Pneumonia/Viral Pneumonia (1024).png')
new_img=image.load_img(image_path,target_size=(244,244))
print('type of new_img',type(new_img))
img = image.img_to_array(new_img)
print('shape of img',img.shape)
img = np.expand_dims(img, axis=0)
print('shape of img',img.shape)
print("Viral Pneumonia")
plt.imshow(new_img)

#顯示一例正常人的數據
image_path=os.path.join(base_path,'NORMAL/NORMAL (1013).png')
new_img = image.load_img(image_path, target_size=(244, 244))
img = image.img_to_array(new_img)
img = np.expand_dims(img, axis=0)
print("NORMAL")
plt.imshow(new_img)

#顯示1例新冠
image_path=os.path.join(base_path,'COVID/COVID (123).png')
new_img = image.load_img(image_path, target_size=(244, 244))
img = image.img_to_array(new_img)
img = np.expand_dims(img, axis=0)
print("COVID-19")
plt.imshow(new_img)

covid_num=len(os.listdir(base_path+'/COVID'))
normal_num=len(os.listdir(base_path+'/NORMAL'))
pneumonia_num=len(os.listdir(base_path+'/Viral Pneumonia'))

#查看三個數據集對比
fig, ax = plt.subplots()
ax.bar("COVID-19",covid_num,color="r",label="covid19")
ax.bar("NORMAL",normal_num,color="b",label="normal")
ax.bar("Viral Pneumonia",pneumonia_num,color="g",label="viral pneumonia")
ax.legend()

 

#CNN模型
model = tf.keras.models.Sequential([
  layers.BatchNormalization(),  #244,244,3
  layers.Conv2D(32, 3, activation='relu'),  #242,242,32
  layers.MaxPooling2D(),   #121,121,32
  layers.Conv2D(64, 3, activation='relu'),  #119,119,64
  layers.MaxPooling2D(), #59,59,64
  layers.Dropout(0.3),  #59,59,64
  layers.Conv2D(128, 3, activation='relu'),  #57,57,128
  layers.MaxPooling2D(),  #28,28,128
  layers.Dropout(0.2),    #28,28,128
  layers.Flatten(),    #100352
  layers.Dense(256, activation='relu'),  #256
  layers.Dropout(0.15),
  layers.Dense(3, activation= 'softmax') #771
])
model.summary()

#提前停止設置，5個epoch驗證損失函數無提升則終止
early = tf.keras.callbacks.EarlyStopping(monitor='val_loss',patience=5)

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy', metrics=['accuracy'])

history=model.fit(train_data,validation_data=val_data,batch_size=32,epochs = 10,callbacks=[early])

#顯示結果
sns.set()

acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(loss) + 1)

#accuracy plot
plt.plot(epochs, acc, color='green', label='Training Accuracy')
plt.plot(epochs, val_acc, color='blue', label='Validation Accuracy')
plt.title('Training and Validation Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend()

plt.figure()
#loss plot
plt.plot(epochs, loss, color='pink', label='Training Loss')
plt.plot(epochs, val_loss, color='red', label='Validation Loss')
plt.title('Training and Validation Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()

plt.show()

#使用前面的例子看結果
image_path=os.path.join(base_path,'COVID/COVID (123).png')
new_img = image.load_img(image_path, target_size=(244, 244))
img = image.img_to_array(new_img)
img = np.expand_dims(img, axis=0)
prediction = model.predict(img)
prediction = np.argmax(prediction,axis=1)
print(prediction)
print(class_names[prediction[0]])

 

 

image_path=os.path.join(base_path,'NORMAL/NORMAL (1013).png')
new_img = image.load_img(image_path, target_size=(244, 244))
img = image.img_to_array(new_img)
img = np.expand_dims(img, axis=0)
prediction = model.predict(img)
prediction = np.argmax(prediction,axis=1)
print(prediction)
print(class_names[prediction[0]])

image_path=os.path.join(base_path,'Viral Pneumonia/Viral Pneumonia (1024).png')
new_img = image.load_img(image_path, target_size=(244, 244))
img = image.img_to_array(new_img)
img = np.expand_dims(img, axis=0)
prediction = model.predict(img)
prediction = np.argmax(prediction,axis=1)
print(prediction)
print(class_names[prediction[0]])

#保存模型
model.save('model.h5')