基于LIDC-IDRI肺结节肺癌数据集的人工智能深度学习分类良性和恶性肺癌（Python 全代码）全流程解析（二）

基于LIDC-IDRI肺结节肺癌数据集的人工智能深度学习分类良性和恶性肺癌（Python 全代码）全流程解析（二）

第一部分内容的传送门
第三部分传送门

1 环境配置和数据集预处理

1.1 环境配置

环境配置建议使用anaconda进行配置。核心的配置是keras和tensorflow的版本要匹配。
环境配置如下：
tensorboard 1.13.1
tensorflow 1.13.1
Keras 2.2.4
numpy 1.21.5
opencv-python 4.6.0.66
python 3.7

1.1 数据集预处理

数据集的预处理分为两个关键步骤。首先是图片处理，我们使用cv2库将图片转换为矩阵格式。这些矩阵随后被堆叠，并放入一个列表中，以便于深度学习模型的读取和处理。

其次是标签处理。我们从保存肺癌恶性程度信息的label.csv文件中逐行读取数据。通过切片和提取，我们获取了肺癌的恶性程度评级，这些评级在1到5之间。我们将大于3的评级归类为恶性，小于3的评级归类为良性。为了让模型更好地理解这些标签，我们用1表示良性，0表示恶性，最后将标签数据转换为one-hot编码格式。

输出与处理的函数如下：

import matplotlib.pyplot as plt
import numpy as np
import keras
import cv2
import os
from keras.preprocessing.image import img_to_array
from keras.utils import to_categorical, plot_model

defload_data(label_path,data_path):
    data_x =[]
    labels =[]
    f =open(label_path)
    label = f.readlines()for Pathimg in os.listdir(os.path.join(data_path,'x')):
        Path    = os.path.join(os.path.join(data_path,'x'),Pathimg)#print(Path)
        image = cv2.imread(Path)
        image = img_to_array(image)
        data_x.append(image)#处理label
        index_num =int(Pathimg.split('.')[0])#  print(labels)用索隐处理，
        a         = label[index_num]
        label_     =int(a[-3:-2])
        label_1 =1if label_ >3else0
        labels.append(label_1)print(labels)# print(data)#guiyihua
    data_x = np.array(data_x,dtype='float')/255.0
    labels = np.array(labels)#转化标签为张量
    labels = to_categorical(labels)#载入data——y
    data_y =[]for Pathimg in os.listdir(os.path.join(data_path,'y')):
        Path    = os.path.join(os.path.join(data_path,'y'),Pathimg)#print(Path)
        image = cv2.imread(Path)
        image = img_to_array(image)
        data_y.append(image)#guiyihua
    data_y = np.array(data_y,dtype='float')/255.0#处理Z
    data_z =[]for Pathimg in os.listdir(os.path.join(data_path,'z')):
        Path    = os.path.join(os.path.join(data_path,'z'),Pathimg)#print(Path)
        image = cv2.imread(Path)
        image = img_to_array(image)
        data_z.append(image)#guiyihua
    data_z = np.array(data_y,dtype='float')/255.0return labels,data_x,data_y,data_z

我们定义好数据预处理的函数后读取数据。图片的数据的格式如下：
其中有875个图片，每个图片的大小为50*50和3个通道

label的格式如下：

2 深度学习模型训练和评估

2.1 深度学习模型训练

我们已经在另一个文件中创建了一个深度学习模型，并且定义了一个函数来导入这个模型。在这里，我们将实例化这个函数。该函数会返回模型的结构以及模型训练的记录信息。

这个模型的输入是来自三个不同角度的图像和对应的标签。模型将被训练 5 个周期，每个周期训练 50 个图像。我们将使用 TensorBoard 查看模型的训练记录。

#######搭建模型
model,callbacks = create_model()#模型训练
H = model.fit([data_x,data_y,data_z],[labels],
              epochs=5,batch_size=50, callbacks=callbacks)

训练结果如下：
我们可以看到训练的准确率还可以为0.7451左右

训练完成后我们将模型储存起来方便对模型进行评估。

#######save model to diskprint('info:saving model.....')
model.save(os.path.join(result_path,'model_44.h5'))

2.1 深度学习模型评估

模型的评估我们首先使用了，分类模型训练的损失函数和准确率来评估模型的训练过程。之后我们还会介绍模型的评估的其他指标。包括混淆矩阵，ROC曲线，AUC值，等

#准确率损失曲线绘制
plt.style.use('ggplot')
plt.figure()
N =5
fig =  plt.plot(np.arange(0, N), H.history["loss"], label="train_loss")#plt.plot(np.arange(0, N), H.history["val_loss"], label="val_loss")
plt.plot(np.arange(0, N), H.history["acc"], label="train_acc")#plt.plot(np.arange(0, N), H.history["val_acc"], label="val_acc")
plt.title("Training Loss and Accuracy on traffic-sign classifier")
plt.xlabel("Epoch #")
plt.ylabel("Loss/Accuracy")
plt.legend(loc="lower left")#plt.show()
plt.savefig('a.jpg',dpi=800)

训练结果如下：

笑话一则开心一下喽

今天去买水果，我问一个摊主：别人家的瓜都写着不甜包退，你怎么不敢写呢，瓜不好吧。摊主：不买滚，我他么卖的是苦瓜……

深度学习，医学图像处理，机器学习精通，需要帮助的联系我（有偿哦）

完整代码如下：

import matplotlib.pyplot as plt
import numpy as np
import keras
import cv2
import os

os.chdir('F:\工作\博客\sort_lung')from models import create_model
from load_datas import load_data

label_path =r'F:\test\data\label.txt'
data_path  =r'F:\test\data\train'
result_path  =r'F:\test\result'#读取预处理数据
labels,data_x,data_y,data_z = load_data(label_path,data_path)#######搭建模型
model,callbacks = create_model()#模型训练
H = model.fit([data_x,data_y,data_z],[labels],
              epochs=5,batch_size=50, callbacks=callbacks)#save model to diskprint('info:saving model.....')
model.save(os.path.join(result_path,'model_44.h5'))#准确率损失曲线绘制
plt.style.use('ggplot')
plt.figure()
N =5
fig =  plt.plot(np.arange(0, N), H.history["loss"], label="train_loss")#plt.plot(np.arange(0, N), H.history["val_loss"], label="val_loss")
plt.plot(np.arange(0, N), H.history["acc"], label="train_acc")#plt.plot(np.arange(0, N), H.history["val_acc"], label="val_acc")
plt.title("Training Loss and Accuracy on traffic-sign classifier")
plt.xlabel("Epoch #")
plt.ylabel("Loss/Accuracy")
plt.legend(loc="lower left")#plt.show()
plt.savefig('a.jpg',dpi=800)

模型文件如下

深度学习模型讲解—待续

# -*- coding: utf-8 -*-"""
Created on Sun Apr 14 21:08:01 2024

@author: dell
"""from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
from sklearn.model_selection import train_test_split
from keras.preprocessing.image import img_to_array
from keras.utils import to_categorical, plot_model
from keras.models import Model
#from imutils import pathsimport matplotlib.pyplot as plt
import numpy as np
import keras
import cv2
import os

#定义模型from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense, Input,Concatenate

defcreate_model():
    model = Sequential()#############定义多输入
    input1  = Input(shape=(50,50,3),name ='input1')
    input2  = Input(shape=(50,50,3),name ='input2')
    input3  = Input(shape=(50,50,3),name ='input3')#############定义多输入
    x1 = Conv2D(32,(3,3),padding='same')(input1)#input is height,width,deep
    x1 = Activation('relu')(x1)
    x1 = Conv2D(32,(3,3),padding='same')(x1)
    x1 = Activation('relu')(x1)
    x1 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x1)
    x1 = Conv2D(48,(3,3),padding='same')(x1)
    x1 = Activation('relu')(x1)
    x1 = Conv2D(48,(3,3),padding='same')(x1)
    x1 = Activation('relu')(x1)
    x1 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x1)
    x1 = Conv2D(64,(3,3),padding='same')(x1)
    x1 = Activation('relu')(x1)
    x1 = Conv2D(64,(3,3),padding='same')(x1)
    x1 = Activation('relu')(x1)
    x1 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x1)# the model so far outputs 3D feature maps (height, width, features)#base_model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors
    x1 = Flatten()(x1)
    x1 = Dense(256)(x1)
    x1 = Activation('relu')(x1)
    x1 = Dropout(0.5)(x1)
    category_predict1 = Dense(100, activation='softmax', name='category_predict1')(x1)# Three loss functions#定义三个全连接层
    
    
    x2 = Conv2D(32,(3,3),padding='same')(input2)#input is height,width,deep
    x2 = Activation('relu')(x2)
    x2 = Conv2D(32,(3,3),padding='same')(x2)
    x2 = Activation('relu')(x2)
    x2 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x2)
    x2 = Conv2D(48,(3,3),padding='same')(x2)
    x2 = Activation('relu')(x2)
    x2 = Conv2D(48,(3,3),padding='same')(x2)
    x2 = Activation('relu')(x2)
    x2 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x2)
    x2 = Conv2D(64,(3,3),padding='same')(x2)
    x2 = Activation('relu')(x2)
    x2 = Conv2D(64,(3,3),padding='same')(x2)
    x2 = Activation('relu')(x2)
    x2 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x2)# the model so far outputs 3D feature maps (height, width, features)#base_model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors
    x2 = Flatten()(x2)
    x2 = Dense(256)(x2)
    x2 = Activation('relu')(x2)
    x2 = Dropout(0.5)(x2)
    category_predict2 = Dense(100, activation='relu', name='category_predict2')(x2)
    
    
    x3 = Conv2D(32,(3,3),padding='same')(input3)#input is height,width,deep
    x3 = Activation('relu')(x3)
    x3 = Conv2D(32,(3,3),padding='same')(x3)
    x3 = Activation('relu')(x3)
    x3 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x3)
    x3 = Conv2D(48,(3,3),padding='same')(x3)
    x3 = Activation('relu')(x3)
    x3 = Conv2D(48,(3,3),padding='same')(x3)
    x3 = Activation('relu')(x3)
    x3 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x3)
    x3 = Conv2D(64,(3,3),padding='same')(x3)
    x3 = Activation('relu')(x3)
    x3 = Conv2D(64,(3,3),padding='same')(x3)
    x3 = Activation('relu')(x3)
    x3 = MaxPooling2D(pool_size=(2,2),strides =(2,2))(x3)# the model so far outputs 3D feature maps (height, width, features)#base_model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors
    x3 = Flatten()(x3)
    x3 = Dense(256)(x3)
    x3 = Activation('relu')(x3)
    x3 = Dropout(0.5)(x3)
    category_predict3 = Dense(100, activation='relu', name='category_predict3')(x3)#融合全连接层
    merge = Concatenate()([category_predict1,category_predict2,category_predict3])#定义输出
    output = Dense(2,activation='sigmoid', name='output')(merge)
    
    model = Model(inputs=[input1, input2, input3], outputs=[output])
    

    callbacks =[keras.callbacks.TensorBoard(
                          log_dir='my_log_dir',)]
    
    model.compile(optimizer=Adam(lr=0.001,decay=0.01),
                  loss='binary_crossentropy',  
                  metrics=['accuracy'],)return model,callbacks