An ultrasound image segmentation method for thyroid nodules based on dual-path attention mechanism-enhanced UNet++

Table 3 Segmentation results of different models

Model	F1_score	IoU	Accuracy	Precision	Recall	HD95
SGUNet [38]	0.7323	0.6298	0.9540	0.8522	0.7120	60.06
UNet [34]	0.7486	0.6507	0.9583	0.8184	0.7668	62.27
CMUNeXt [35]	0.7812	0.6908	0.9643	0.8194	0.8192	50.47
TransAttUNet [39]	0.7992	0.7054	0.9645	0.8374	0.8271	39.56
TRFE+ [4]	0.8003	0.7109	0.9663	0.8536	0.8172	40.44
DeepLabV3+ [36]	0.8058	0.7159	0.9681	0.8525	0.8208	36.50
TransUNet [37]	0.8067	0.7188	0.9667	0.8574	0.8161	37.73
DPAM-UNet++(ours)	0.8310	0.7451	0.9718	0.8443	0.8702	35.31

ISSN: 1471-2342