Figure 1
Proposed main architecture. This figure shows a schematic representation of the proposed main architecture. The prepared image slices are sequentially fed into the network with all its corresponding modalities. The network generates a prediction based on this current slice and then takes the next one. The first DS-Block has \(n_0\) filters for the convolution layers. The filter count doubles for every DS-Block. The number of filters in the up-sampling branch matched the number of filters in the corresponding down-sampling blocks. The predictions were stacked for the axial, coronal and sagittal view which yielded three 3D volumes with their specific class membership probabilities. Those three volumes were then combined, averaged and finally thresholded to generate one binary volume as a prediction.
