Table 6 Component ablation study of the proposed architecture on the tibia µCT validation dataset. Configuration descriptions: Configuration 1—DBAHNet with no bottleneck and using standard convolution in both the encoder and decoder; Configuration 2—DBAHNet using the Channel-wise Attention-Based Convolution Module (CACM) in both the encoder and decoder, still without a bottleneck; Configuration 3—DBAHNet with the CACM in the encoder and the Spatial-Wise Attention-Based Convolution Module (SACM) in the decoder, still without a bottleneck; Configuration 4—The full DBAHNet with all the components. Abbreviations: GFLOPS (Giga Floating Point Operations Per Second).

From: A robust deep learning approach for segmenting cortical and trabecular bone from 3D high resolution µCT scans of mouse bone