Current Status and Future Perspectives of Artificial Intelligence in Magnetic Resonance Breast Imaging

Table 3

Detection of breast lesions in breast MRI using DL.

DL technique	Evaluation results	Dataset	References
Model agnostic saliency	TPR = 80; FPs/image = 8	117 subjects; DCE-MRI and T1W images	[86]
U-net	Acc = 94.2	67 MR images T1W, T2W, DWI, and DCE-MRI	[87]
Patch-based analysis with ResNet50 backbone	AUC = 0.817	335 MR images of 17 different histological subtypes	[65]
Deep Q-network	Sn = 80; FPs/image = 3.2	117 DCE-MR and T1-weighted images	[63]
Unsupervised saliency analysis and CNN	Acc = 86 ± 2; AUC = 0.94 ± 0.01	193 DCE-MR images	[88]
Two-level U-net and dual-stream CNN	CPM = 64.29	Training: 201 DCE-MR images; Testing: 160 DCE-MR images	[64]