使用Keras构建CNN神经网络在医学图像CT上的应用2

1. 数据集下载kaggle
2. 导入包

import numpy as np  # matrix tools
import matplotlib.pyplot as plt  # for basic plots
import seaborn as sns  # for nicer plots
import pandas as pd
from glob import glob
import re
from skimage.io import imread
import keras
import os
from sklearn.model_selection import train_test_split

from keras.models import Sequential
from keras.layers import Dense, Flatten
from keras.optimizers import Adam, RMSprop
from keras.layers import Conv2D, MaxPooling2D

3. 数据集路径

datasetPath = '/KaggleData/CTDataset'

4. 获取数据集的tiff文件名

def pathList(datasetPath):
    all_images_list = glob(os.path.join(datasetPath, 'tiff_images', '*.tif'))
    return all_images_list

5. 定义获取label和id函数

# 设置label
def setLabel(imagesPathList):
    check_contrast = re.compile(r'ID_([\d]+)_AGE_([\d]+)+_CONTRAST_([\d]+)_CT.tif')
    label = []
    id_list = []
    age = []
    for image in imagesPathList:
        id_list.append(check_contrast.findall(image)[0][0])
        age.append(check_contrast.findall(image)[0][1])
        label.append(check_contrast.findall(image)[0][2])

    return label, id_list

6. 创建训练集

# 创建训练集和验证集和测试集
def createSet(imagesPathList):
    # 使用切片操作imread(x)[::2, ::2],跨两步选取一个值,降低维度,提高训练速度
    jimread = lambda x: np.expand_dims(imread(x)[::2, ::2], 0)
    images = np.stack([jimread(i) for i in imagesPathList], 0)
    label, id_list = setLabel(imagesPathList)
    label_list = pd.DataFrame(label, id_list)
    X_train, X_test, y_train, y_test = train_test_split(images, label_list, test_size=0.1, random_state=0)
    n_train, depth, width, height = X_train.shape
    n_test, _, _, _ = X_test.shape

    # 归一化
    input_train = X_train.reshape((n_train, width, height, depth))
    input_train = input_train / np.max(input_train)
    # 自己加的
    input_test = X_test.reshape((n_test, width, height, depth))
    input_test = input_test / np.max(input_test)

    # 设置类别
    output_train = keras.utils.to_categorical(y_train, 2)
    output_test = keras.utils.to_categorical(y_test, 2)
    return input_train, output_train, input_test, output_test

7. 创建CNN神经网络

# 创建CNN神经网络模型
def CNNmodel(input_train, output_train, input_test, output_test):
    # 设置参数
    batch_size = 20
    epochs = 20

    # 构建网络
    model2 = Sequential()
    # input_train.shape[1:]作用单个样本的shape,第一个位置代表样本数目
    model2.add(Conv2D(50, (5, 5), activation='relu', input_shape=input_train.shape[1:]))
    # 使用32个4x4过滤器创建卷积网络
    model2.add(MaxPooling2D(pool_size=(3, 3)))  # 3x3 Maxpooling
    model2.add(Conv2D(30, (4, 4), activation='relu', input_shape=input_train.shape[1:]))
    model2.add(MaxPooling2D(pool_size=(2, 2)))  # 2x2 Maxpooling
    model2.add(Flatten())  # 膨胀已创建全连接神经网络
    model2.add(Dense(2, activation='softmax'))

    # 显示训练模型摘要
    print(model2.summary())

    # 整合模型
    model2.compile(loss='categorical_crossentropy',
                  optimizer=Adam(),
                  metrics=['accuracy'])

    # 开始训练
    history = model2.fit(input_train, output_train,
                         batch_size=batch_size,
                         epochs=epochs,
                         verbose=1,
                         validation_data=(input_test, output_test))

    # 评估结果
    score = model2.evaluate(input_test, output_test, verbose=0)
    print(score)

    # 可视化训练过程
    print(history.history.keys())
    print('Test loss:', score[0])
    print('Test accuracy:', score[1])
    plt.plot(history.history['loss'])
    plt.plot(history.history['accuracy'])
    plt.show()

9. 展示图片

# 展示图片
def showCT(imagesPathList):
    f, ax = plt.subplots(10, 10, figsize=(10, 10))
    for index, i in enumerate(imagesPathList):
        ax[int(index / 10), index % 10].matshow(imread(i), cmap='gray')
    plt.show()

10. visualize Age

# visualize Age
def visualizeAge(data):
    plt.figure(figsize=(10, 5))
    sns.distplot(data['Age'])
	# 按Contrast类别区分作图颜色
    g = sns.FacetGrid(data, hue="Contrast", size=6, legend_out=True)
    g = g.map(sns.distplot, "Age").add_legend()

    plt.show()

11. main函数

# fileInfo(datasetPath)
# 读取数据
# overview = getData(overviewPath)
#
# visualizeAge(overview)
pathlist = pathList(datasetPath)
# showCT(pathlist)
input_train, output_train, input_test, output_test = createSet(pathlist)
# 运行模型
CNNmodel(input_train, output_train, input_test, output_test)

CNN结构:
训练数据 (90,255,255,1)
卷积层(255,255,1)
池化层3x3
卷积层(255,255,1)
池化层2x2
膨胀层
Dense层