Pre- and Post-processing Strategies for domain-specific Slice-wise Segmentation of Tomographic 3D datasets Utilizing U-Net Deep Learning Models