3D UNet 激活函数和数字类答案

【问题标题】：3D UNet activation function and number classes3D UNet 激活函数和数字类
【发布时间】：2022-08-04 03:36:38
【问题描述】：

我是深度学习的初学者，正在尝试开发一个 U-Net 模型，用于在 nifti 图像上进行血管分割（血管（白色像素）、背景（黑色像素））。我对定义类的数量和 sigmoid/softmax 激活函数感到困惑。我应该为此设置 n_classes = 2 和 softmax 激活函数，还是设置 n_classes = 1 和 sigmoid 激活函数？下面是 DataGenerator 和 UNet 模型的代码。

n_classes = 2
class DataGenerator(tf.keras.utils.Sequence):

    def __init__(self, img_paths, mask_paths, batch_size, n_classes):
        self.x, self.y = img_paths, mask_paths
        self.batch_size = batch_size
        self.n_classes = n_classes

    def __len__(self):
        return math.ceil(len(self.x) / self.batch_size)

    def read_nifti(self, filepath):
        volume = nib.load(filepath).get_fdata()
        volume = np.array(volume)
        return volume


    def __getitem__(self, idx):

        batch_x = self.x[idx * self.batch_size:(idx + 1) * self.batch_size]
        batch_y = self.y[idx * self.batch_size:(idx + 1) * self.batch_size]

        image = [self.read_nifti(image_file) for image_file in batch_x]
        image = np.array(image, dtype=np.float32)
        image = tf.expand_dims(image, axis=-1)
        
        label = [self.read_nifti(mask_file) for mask_file in batch_y]
        label = np.array(label, dtype=np.float32)
        label = tf.keras.utils.to_categorical(label, num_classes=self.n_classes)
        
        return image, label

\'\'\'---------------------build CNN model -------------------\'\'\'
def unet3d_model1(nx= 224, ny=224, nz=64):

    inputs = Input((nx, ny, nz, 1))
    conv1 = Conv3D(32, (3, 3, 3), activation=\'relu\', padding=\'same\')(inputs)
    conv1 = Conv3D(32, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv1)
    pool1 = MaxPool3D(pool_size=(2, 2, 2))(conv1)

    conv2 = Conv3D(64, (3, 3, 3), activation=\'relu\', padding=\'same\')(pool1)
    conv2 = Conv3D(64, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv2)
    pool2 = MaxPool3D(pool_size=(2, 2, 2))(conv2)

    conv3 = Conv3D(128, (3, 3, 3), activation=\'relu\', padding=\'same\')(pool2)
    conv3 = Conv3D(128, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv3)
    pool3 = MaxPool3D(pool_size=(2, 2, 2))(conv3)

    conv4 = Conv3D(256, (3, 3, 3), activation=\'relu\', padding=\'same\')(pool3)
    conv4 = Conv3D(256, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv4)

    up5 = UpSampling3D(size=(2, 2, 2))(conv4)
    merge5 = concatenate([up5, conv3])
    conv5 = Conv3D(128, (3, 3, 3), activation=\'relu\', padding=\'same\')(merge5)
    conv5 = Conv3D(128, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv5)

    up6 = UpSampling3D(size=(2, 2, 2))(conv5)
    merge6 = concatenate([up6, conv2])
    conv6 = Conv3D(64, (3, 3, 3), activation=\'relu\', padding=\'same\')(merge6)
    conv6 = Conv3D(64, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv6)

    up7 = UpSampling3D(size=(2, 2, 2))(conv6)
    merge7 = concatenate([up7, conv1])
    conv7 = Conv3D(32, (3, 3, 3), activation=\'relu\', padding=\'same\')(merge7)
    conv7 = Conv3D(32, (3, 3, 3), activation=\'relu\', padding=\'same\')(conv7)

    conv8 = Conv3D(n_classes, (1, 1, 1), activation=\'softmax\')(conv7)

    model = Model(inputs=inputs, outputs=conv8)

    return model

标签： python keras deep-learning image-segmentation

【解决方案1】：

正如您所说，像素可以属于血管像素或背景像素。那么，使用带有 sigmoid 激活输出层的单个节点有什么意义。这样，您就有可能获得相关像素的“血管性”。那么显然，您需要应用阈值来提取血管图像。

希望有帮助。

【讨论】：