Lets say that I have a image showing cell tissue and I hope to segment the image to outline the healthy and unhealthy areas. This image is called the current image. I also have images taken an hour before, and two hours before. The model should look at all three images so it can understand how the cell tissue is changing from one period to another, and thus make better predictions on the current image. In other words, the data should be of shape (L, W, C, T) where l, w, c, t are the length, width, color, and time. This is very similar to feeding a video into a neural network. And the model should output a mask only for the current image. So are there image segmentation networks that can take advantage of time information? This is similar to 3D convolutional network for action recognition. Or should I just concatenate the images along the color dimension so it has a shape of (L, W, C*T).