Hi,I have a question,If i use a pre-trained 2d architectire and change the first layer to such that it takes 3d data(i.e niftis),which has layer by layer dicoms,will the rest of the 2d model know the relation between the layers.
If we think about the working principle of a 2D convolutional filter, we know they only read and aggregate data from one and one only single 2D layer. In other words, we can’t expect those 2D filters to recognize relations across multiple 2D layers.
No.