Automatic segmentation model of intercondylar fossa based on deep learning: a novel and effective assessment method for the notch volume

Table 2 Results for the external test dataset with Different Networks

Network	Dice similarity coefficient	Automatic Segmentation Volume (cm³)	Manual Segmentation Volume (cm³)	Relative Error
U-Net	0.914 ± 0.04	7.206 ± 1.497	7.421 ± 1.476	0.064 ± 0.018
Seg-Net	0.906 ± 0.10	7.184 ± 1.498	7.421 ± 1.476	0.071 ± 0.016
Res-UNet	0.916 ± 0.04	7.251 ± 1.492	7.421 ± 1.476	0.054 ± 0.019
Dense-UNet	0.901 ± 0.08	7.169 ± 1.490	7.421 ± 1.476	0.075 ± 0.016
Mobile-UNet	0.906 ± 0.07	7.178 ± 1.516	7.421 ± 1.476	0.078 ± 0.013

The relative error is expressed as the following: \(E=\frac{R-P}{R}\) R is the real volume and P is the predicted volume

ISSN: 1471-2474