In the pretrained network, each image had to be normalized by an average image first, and as we know, the nodule images were grayscale images, so they were normalized by one channel (red, green, or blue) only. Experiments were performed with red, green, and blue channels separately, taking one channel at a time and ignoring the other two (i.e., removing the weights and connections of the other two from the CNN). We also used the grayscale image three times to simulate an image with three color channels and did normalization using the appropriate color channel image. 解释

在预训练网络中，每个图像都必须先由平均图像进行归一化处理。由于结节图像是灰度图像，因此只能使用一个通道（红色、绿色或蓝色）进行归一化。我们进行了不同的实验，分别将红色、绿色和蓝色通道分别用于训练，一次只使用一个通道，忽略其他两个通道（即从CNN中删除其他两个通道的权重和连接）；我们还使用灰度图像三次，模拟具有三个颜色通道的图像，并使用相应的颜色通道图像进行归一化处理。

相关问题

The RGB images have three color channels (24-bit image), but the grayscale image had only a single grayscale image (8-bit image). In our previous experiment, we normalized the images of the pre- trained network by each color channel separately. This approach lost the information provided by the other two channels. Here, we used the same grayscale LDCT image for each channel to make it somewhat analogous to an RGB image for the CNN. Doing so engages all the weights and exploits all the learned knowledge when extracting features from the pretrained net- work. Since the images experimented with were smaller than the required input size of the pretrained CNN (224 × 224), we used bicubic interpolation for resizing. The dimension of a deep feature vector extracted for each image was 4096. The features were the output of the last fully connected layer (the full 2 layer as shown in Table 2) before the output layer in an ImageNet pretrained CNN. The architectures and para 解释

这段话主要是关于一个实验中的图像处理和深度学习网络的使用。该实验中使用了一个预训练的卷积神经网络（CNN）来提取图像的特征，以用于后续的任务。由于原始图像是RGB图像，而预训练网络是对彩色图像进行训练的，因此将灰度图像复制三份，以便处理成类似于RGB图像的形式，以便充分利用预训练网络的知识。因为图像的大小比预训练网络的输入大小要小，所以使用了双三次插值的方法进行图像的缩放。最后，提取了每个图像的深度特征向量，其维度为4096，这些特征是在ImageNet预训练CNN的输出层之前的最后一个全连接层（即表2中的完整2层）的输出结果。

Transfer learning25,26 is a method where previously learned knowledge is applied to another task and the task domains may be different. In our case, the domain is very differ- ent. ImageNet consists of natural camera images and does not include any type of lung nodule or cancer image. Our image set consists of only lung nodules in CT images. We experimented with three different pretrained CNN’s [vgg (visual geometry group)-m/vgg-f/vgg-s]27 in this study using a MA TLAB toolbox named MA TCONVNET.28 The f, m, and s after vgg- stand for fast, medium, and slow and refer to training time (so partly the number of weights). We obtained deep features from the outputs of the last fully connected layer after applying the activation function using a rectified linear unit (post-ReLU), which changes all values <0 to be 0. The LDCT images were grayscale (no color component and we changed the voxel intensities of LDCT images to 0-255 or 8 bit), but the pretrained network was trained on RGB images, so we normalized the images by the average red, green, and blue channel images, and exper- imented by using each channel separately. 解释

Transfer learning是一种方法，其中先前学习的知识被应用于另一个任务，任务域可能不同。在我们的情况下，领域非常不同。ImageNet由自然相机图像组成，不包括任何类型的肺结节或癌症图像。我们的图像集仅包含CT图像中的肺结节。在本研究中，我们使用名为MATCONVNET的MATLAB工具箱尝试了三种不同的预训练CNN（vgg（visual geometry group）-m / vgg-f / vgg-s）进行实验。vgg之后的f，m和s代表快速，中等和慢，并且指的是训练时间（因此部分权重的数量）。我们从应用激活函数后的最后一个完全连接层的输出中获得深层特征，使用修正线性单元（post-ReLU）对所有值<0进行更改以使其为0。LDCT图像是灰度的（没有彩色分量），我们将LDCT图像的体素强度更改为0-255或8位，但预训练网络是在RGB图像上训练的，因此我们通过平均红色，绿色和蓝色通道图像对图像进行归一化，并尝试分别使用每个通道。