Rib Fracture Detection with Dual-Attention Enhanced U-Net

Table 2

Comparison experiment results of our proposed method, several cutting-edge rib fracture detection methods, and commonly used U-Net architecture deep neural networks are included. The best results are marked in bold. FPs/scan denote false positives per scan.

Methods	Sensitivities (FPs/scan)
	0.5	1	2	4	8	Avg.
CFSG U-Net	67.15	76.92	84.78	87.98	89.58	81.28
FracNet	60.10	70.03	79.01	82.21	85.90	75.45
VRBNet	71.31	77.56	82.21	85.25	85.25	80.32
3D U-Net	59.46	69.23	77.72	81.73	85.26	74.68
MutiResUnet	61.70	71.79	80.93	83.81	87.18	77.08
Attention U-Net	61.54	72.12	81.09	84.29	87.34	77.28
ResUNet	60.90	70.99	80.13	83.33	86.70	76.41