Table 1.

Summary of previous transfer learning (TL) approaches for breast cancer diagnosis using ultrasound. OASBUD, open access series of breast ultrasound data; US, ultrasound; UDIAT, UDIAT Diagnostic Centre of the Parc Tauli Corporation ultrasound image data; dataset A, ultrasound images obtained with BK Medical Panther 2002 and BK Medical Hawk 2102; dataset B, UDIAT Diagnostic Centre of the Parc Tauli Corporation ultrasound image data.

Study	TL Approach Used	Pre-Training Model Used	Application	Image Dataset	Pre-Processing	Pre-Training Dataset
Byra et al. [26]	Fine-tuning	VGG19 & InceptionV3	Classification	OASBUD	Compression and augmentation	ImageNet
Byra et al. [24]	Fine-tuning	VGG19	Classification	882 US images of their own and public images UDIAT and OASBUD	Matching layer	ImageNet
Hijab et al. [27]	Fine-tuning	VGG16	Classification	1300 US Images	Augmentation	ImageNet
Yap et al. [25]	Fine-tuning	AlexNet	Detection	Dataset A and B	Splitting in to patches	ImageNet
Yap et al. [28]	Fine-tuning	AlexNet	Detection	Dataset A and B	Ground-truth labeling	ImageNet
Huynh et al. [23]	Feature extractor	AlexNet	Classification	Breast mammogram dataset with 2393 regions of interest (ROIs)	Compression and augmentation	ImageNet
Hadad et al. [29]	Fine-tuning	VGG128	Detection and classification	MRI data	Augmentation	Medical Image (Mammography image)