基于U-Net的视网膜血管分割（Pytorch完整版）

基于 U-Net 的视网膜血管分割是一种应用深度学习的方法，特别是 U-Net 结构，用于从眼底图像中分割出视网膜血管。U-Net 是一种全卷积神经网络（FCN），通常用于图像分割任务。以下是基于 U-Net 的视网膜血管分割的内容：
框架结构：

代码结构：

U-Net分割代码：

unet_model.py

import torch.nn.functional as F
from .unet_parts import *
class UNet(nn.Module):
    def __init__(self, n_channels, n_classes, bilinear=True):
        super(UNet, self).__init__()
        self.n_channels = n_channels
        self.n_classes = n_classes
        self.bilinear = bilinear

        self.inc = DoubleConv(n_channels, 64)
        self.down1 = Down(64, 128)
        self.down2 = Down(128, 256)
        self.down3 = Down(256, 512)
        self.down4 = Down(512, 512)
        self.up1 = Up(1024, 256, bilinear)
        self.up2 = Up(512, 128, bilinear)
        self.up3 = Up(256, 64, bilinear)
        self.up4 = Up(128, 64, bilinear)
        self.outc = OutConv(64, n_classes)



    def forward(self, x):
        x1 = self.inc(x)

        # 在编码器下采样过程加空间注意力
        # x2 = self.down1(self.sp1(x1))
        # x3 = self.down2(self.sp2(x2))
        # x4 = self.down3(self.sp3(x3))
        # x5 = self.down4(self.sp4(x4))

        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)

        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)

        logits = self.outc(x)
        return logits

if __name__ == '__main__':
    net = UNet(n_channels=3, n_classes=1)
    print(net)

unet_parts.py

import torch
import torch.nn as nn
import torch.nn.functional as F


class DoubleConv(nn.Module):
    """(convolution => [BN] => ReLU) * 2"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.double_conv = nn.Sequential(
            nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True)
        )

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.constant_(m.bias, 0)


    def forward(self, x):
        return self.double_conv(x)


class Down(nn.Module):
    """Downscaling with maxpool then double conv"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.maxpool_conv = nn.Sequential(
            nn.MaxPool2d(2),
            DoubleConv(in_channels, out_channels)
        )

    def forward(self, x):
        return self.maxpool_conv(x)


class Up(nn.Module):
    """Upscaling then double conv"""

    def __init__(self, in_channels, out_channels, bilinear=True):
        super().__init__()

        # if bilinear, use the normal convolutions to reduce the number of channels
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
        else:
            self.up = nn.ConvTranspose2d(in_channels // 2, in_channels // 2, kernel_size=2, stride=2)

        self.conv = DoubleConv(in_channels, out_channels)

    def forward(self, x1, x2):
        x1 = self.up(x1)
        # input is CHW
        diffY = torch.tensor([x2.size()[2] - x1.size()[2]])
        diffX = torch.tensor([x2.size()[3] - x1.size()[3]])

        x1 = F.pad(x1, [diffX // 2, diffX - diffX // 2,
                        diffY // 2, diffY - diffY // 2])

        x = torch.cat([x2, x1], dim=1)
        return self.conv(x)


class OutConv(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(OutConv, self).__init__()
        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.constant_(m.bias, 0)

    def forward(self, x):
        return self.conv(x)

trainval.py

from model.unet_model import UNet
from utils.dataset import FundusSeg_Loader

from torch import optim
import torch.nn as nn
import torch
import sys
import matplotlib.pyplot as plt
from tqdm import tqdm
import time

train_data_path = "DRIVE/drive_train/"
valid_data_path = "DRIVE/drive_test/"
# hyperparameter-settings
N_epochs = 500
Init_lr = 0.00001

def train_net(net, device, epochs=N_epochs, batch_size=1, lr=Init_lr):
    # 加载训练集
    train_dataset = FundusSeg_Loader(train_data_path, 1)
    valid_dataset = FundusSeg_Loader(valid_data_path, 0)
    train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True)
    valid_loader = torch.utils.data.DataLoader(dataset=valid_dataset, batch_size=batch_size, shuffle=False)
    print('Traing images: %s' % len(train_loader))
    print('Valid  images: %s' % len(valid_loader))

    # 定义RMSprop算法
    optimizer = optim.RMSprop(net.parameters(), lr=lr, weight_decay=1e-8, momentum=0.9)
    # 定义Loss算法
    # BCEWithLogitsLoss会对predict进行sigmoid处理
    # criterion 常被用来定义损失函数，方便调换损失函数
    criterion = nn.BCEWithLogitsLoss()
    # 训练epochs次
    # 求最小值，所以初始化为正无穷
    best_loss = float('inf')
    train_loss_list = []
    val_loss_list = []
    for epoch in range(epochs):
        # 训练模式
        net.train()
        train_loss = 0
        print(f'Epoch {epoch + 1}/{epochs}')
        # SGD
        # train_loss_list = []
        # val_loss_list = []
        with tqdm(total=train_loader.__len__()) as pbar:
            for i, (image, label, filename) in enumerate(train_loader):
                optimizer.zero_grad()
                # 将数据拷贝到device中
                image = image.to(device=device, dtype=torch.float32)
                label = label.to(device=device, dtype=torch.float32)
                # 使用网络参数，输出预测结果
                pred = net(image)
                # print(pred)
                # 计算loss
                loss = criterion(pred, label)
                # print(loss)
                train_loss = train_loss + loss.item()

                loss.backward()
                optimizer.step()
                pbar.set_postfix(loss=float(loss.cpu()), epoch=epoch)
                pbar.update(1)

        train_loss_list.append(train_loss / i)
        print('Loss/train', train_loss / i)

        # Validation
        net.eval()
        val_loss = 0
        for i, (image, label, filename) in tqdm(enumerate(valid_loader), total=len(valid_loader)):
            image = image.to(device=device, dtype=torch.float32)
            label = label.to(device=device, dtype=torch.float32)
            pred = net(image)
            loss = criterion(pred, label)
            val_loss = val_loss + loss.item()
            # net.state_dict()就是用来保存模型参数的
            if val_loss < best_loss:
                best_loss = val_loss
                torch.save(net.state_dict(), 'best_model.pth')
                print('saving model............................................')

        val_loss_list.append(val_loss / i)
        print('Loss/valid', val_loss / i)
        sys.stdout.flush()
    return val_loss_list, train_loss_list


if __name__ == "__main__":
    # 选择设备cuda
    device = torch.device('cuda')
    # 加载网络，图片单通道1，分类为1。
    net = UNet(n_channels=3, n_classes=1)
    # 将网络拷贝到deivce中
    net.to(device=device)
    # 开始训练
    val_loss_list, train_loss_list = train_net(net, device)
    # 保存loss值到txt文件
    fileObject1 = open('train_loss.txt', 'w')
    for train_loss in train_loss_list:
        fileObject1.write(str(train_loss))
        fileObject1.write('\n')
    fileObject1.close()
    fileObject2 = open('val_loss.txt', 'w')
    for val_loss in val_loss_list:
        fileObject2.write(str(val_loss))
        fileObject2.write('\n')
    fileObject2.close()
    # 我这里迭代了5次，所以x的取值范围为(0，5)，然后再将每次相对应的5损失率附在x上
    x = range(0, N_epochs)
    y1 = val_loss_list
    y2 = train_loss_list
    # 两行一列第一个
    plt.subplot(1, 1, 1)
    plt.plot(x, y1, 'r.-', label=u'val_loss')
    plt.plot(x, y2, 'g.-', label =u'train_loss')
    plt.title('loss')
    plt.xlabel('epochs')
    plt.ylabel('loss')
    plt.savefig("accuracy_loss.jpg")
    plt.show()

predict.py

import numpy as np
import torch
import cv2
from model.unet_model import UNet

from utils.dataset import FundusSeg_Loader
import copy
from sklearn.metrics import roc_auc_score

model_path='./best_model.pth'
test_data_path = "DRIVE/drive_test/"
save_path='./results/'

if __name__ == "__main__":
    test_dataset = FundusSeg_Loader(test_data_path,0)
    test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=1, shuffle=False)
    print('Testing images: %s' %len(test_loader))
    # 选择设备CUDA
    device = torch.device('cuda')
    # 加载网络，图片单通道，分类为1。
    net = UNet(n_channels=3, n_classes=1)
    # 将网络拷贝到deivce中
    net.to(device=device)
    # 加载模型参数
    print(f'Loading model {model_path}')
    net.load_state_dict(torch.load(model_path, map_location=device))

    # 测试模式
    net.eval()
    tp = 0
    tn = 0
    fp = 0
    fn = 0
    pred_list = []
    label_list = []
    for image, label, filename in test_loader:
        image = image.to(device=device, dtype=torch.float32)
        pred = net(image)
        # Normalize to [0, 1]
        pred = torch.sigmoid(pred)
        pred = np.array(pred.data.cpu()[0])[0]
        pred_list.append(pred)
        # ConfusionMAtrix
        pred_bin = copy.deepcopy(pred)
        label = np.array(label.data.cpu()[0])[0]
        label_list.append(label)
        pred_bin[pred_bin >= 0.5] = 1
        pred_bin[pred_bin < 0.5] = 0
        tp += ((pred_bin == 1) & (label == 1)).sum()
        tn += ((pred_bin == 0) & (label == 0)).sum()
        fn += ((pred_bin == 0) & (label == 1)).sum()
        fp += ((pred_bin == 1) & (label == 0)).sum()
        # 保存图片
        pred = pred * 255
        save_filename = save_path + filename[0] + '.png'
        cv2.imwrite(save_filename, pred)
        print(f'{save_filename} done!')
    # Evaluaiton Indicators
    precision = tp / (tp + fp)   # 预测为真并且正确/预测正确样本总和
    sen = tp / (tp + fn)    # 预测为真并且正确/正样本总和
    spe = tn / (tn + fp)
    acc = (tp + tn) / (tp + tn + fp + fn)
    f1score = 2 * precision * sen / (precision + sen)
    # auc computing
    pred_auc = np.stack(pred_list, axis=0)
    label_auc = np.stack(label_list, axis=0)
    auc = roc_auc_score(label_auc.reshape(-1), pred_auc.reshape(-1))
    print(f'Precision: {precision} Sen: {sen} Spe:{spe} F1-score: {f1score} Acc: {acc} AUC: {auc}')

dataset.py

import torch
import cv2
import os
import glob
from torch.utils.data import Dataset

# import random
# from PIL import Image
# import numpy as np


class FundusSeg_Loader(Dataset):
    def __init__(self, data_path, is_train):
        # 初始化函数，读取所有data_path下的图片
        self.data_path = data_path
        self.imgs_path = glob.glob(os.path.join(data_path, 'image/*.tif'))
        self.labels_path = glob.glob(os.path.join(data_path, 'label/*.tif'))
        self.is_train = is_train
        print(self.imgs_path)
        print(self.labels_path)
    def __getitem__(self, index):
        # 根据index读取图片
        image_path = self.imgs_path[index]
        if self.is_train == 1:
            label_path = image_path.replace('image', 'label')
            label_path = label_path.replace('training', 'manual1')
        else:
            label_path = image_path.replace('image', 'label')
            label_path = label_path.replace('test.tif', 'manual1.tif')
            
        
        # 读取训练图片和标签图片
        image = cv2.imread(image_path)
        label = cv2.imread(label_path)

        # image = np.array(image)
        # label = np.array(label)
        # label = cv2.imread(label_path)
        # image = cv2.resize(image, (600,400))
        # label = cv2.resize(label, (600,400))
        # 转为单通道的图片
        # image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        label = cv2.cvtColor(label, cv2.COLOR_BGR2GRAY)
        # label = Image.fromarray(label)
        # label = label.convert("1")
        # reshape()函数可以改变数组的形状，并且原始数据不发生变化。
        image = image.transpose(2, 0, 1)
        # image = image.reshape(1, label.shape[0], label.shape[1])
        label = label.reshape(1, label.shape[0], label.shape[1])
        # 处理标签，将像素值为255的改为1
        if label.max() > 1:
            label[label > 1] = 1

        return image, label, image_path[len(image_path)-12:len(image_path)-4]

    def __len__(self):
        # 返回训练集大小
        return len(self.imgs_path)

visual.py

import numpy as np
import matplotlib.pyplot as plt
import pylab as pl

from mpl_toolkits.axes_grid1.inset_locator import inset_axes
data1_loss =np.loadtxt("E:\\code\\UNet_lr00001\\train_loss.txt",dtype=str )
data2_loss = np.loadtxt("E:\\code\\UNet_lr00001\\val_loss.txt",dtype=str)
x = range(0,10)
y = data1_loss[:, 0]
x1 = range(0,10)
y1 = data2_loss[:, 0]
fig = plt.figure(figsize = (7,5))    #figsize是图片的大小`
ax1 = fig.add_subplot(1, 1, 1) # ax1是子图的名字`
pl.plot(x,y,'g-',label=u'Dense_Unet(block layer=5)')
# ''g''代表"green",表示画出的曲线是绿色，"-"代表画的曲线是实线，可自行选择，label代表的是图例的名称，一般要在名称前面加一个u，如果名称是中文，会显示不出来，目前还不知道怎么解决。
p2 = pl.plot(x, y,'r-', label = u'train_loss')
pl.legend()
#显示图例
p3 = pl.plot(x1,y1, 'b-', label = u'val_loss')
pl.legend()
pl.xlabel(u'epoch')
pl.ylabel(u'loss')
plt.title('Compare loss for different models in training')

这种基于 U-Net 的方法已在医学图像分割领域取得了一些成功，特别是在视网膜图像处理中。通过深度学习的方法，这种技术能够更准确地提取视网膜血管，为眼科医生提供辅助诊断和治疗的信息。

如有疑问，请评论。