Table 1 A summary of related studies, their approaches, and application as compared with what is obtained in this study.

Chen et al.⁵⁴

InfoGAN: based on unsupervised learning which maximizes mutual information in a small subset of latent variables. Applied to writing styles using MNIST

RoiMammoGAN: used semi-supervised learning. Applied to breast images from mammograms using MIAS

Zhang et al.⁵⁵

StackGAN: stacked two GANs on each other. Applied to generating photorealistic images using CUB and Oxford-102 datasets

RoiMammoGAN: one GAN model sufficiently and accurately achieved our aim. Applied to breast images from mammograms using MIAS

Wang and Gupta⁵⁶

S2-GAN: composes of a style and structure GANS. Applied for generating structure and style in 2D images

RoiMammoGAN: learns the pattern and structure of abnormalities in medical images. Applied to breast images from mammograms using MIAS

Nguyen et al.⁵⁷

PPGNs: used probabilistic interpretation and performance gradient to generate realistic and high-resolution images

RoiMammoGAN: combined the Adam gradient algorithm and performance increment to generate images. Applied to breast images from mammograms using MIAS

Neff⁵⁸

WGAN: improved loss function performance using Wasserstein

RoiMammoGAN: a combination of RELU and LeakyRELU were used for computing loss function

Arjovsky et al.⁵⁹

WGAN: aimed to stabilize learning pattern and reducing mode collapse in GAN

RoiMammoGAN: architectural composition showed that this model overcome mode collapse

Gulrajani et al.⁶⁰

Used a penalization mechanism for norm gradient to overcome clipping weights. Used CIFAR-10 and LSUN bedrooms

RoiMammoGAN: the challenge of clipping weights was eliminated in our model. Applied to breast images from mammograms using MIAS

Mao et al.⁶¹

LSGAN: least squares loss function was used to curtail the vanishing gradients problem

RoiMammoGAN: kernel sizes of \(D\) and \(G\) were intelligently selected through investigative experimentation to overcome vansing gradient problem

Ilya et al.⁶²

AdaGAN: addition of component through iterative procedure to avoid training problem

RoiMammoGAN: to eliminate complexity of learning features during training, staged-class-based learning was applied

Odena et al.⁶³

CGAN: uses label conditioning to generate high resolution images

RoiMammoGAN: adopted the label conditioning strategy in addition to label flipping

Pana et al.⁶⁴

SalGAN: designed as a data-driven metric-based saliency prediction method and trained with an adversarial loss function

RoiMammoGAN: the concept of saliency map was not considered in the study

Wang et al.⁶⁵

SRGAN: high resolution focused GAN model

RoiMammoGAN: also a high-resolution focused GAN model

Wu et al.⁶⁶

ciGAN: used for contextual in-filling for synthesizing lesions. Applied to mammogram patches

RoiMammoGAN: uses the class label to condition the learning and training process. Applied to mammogram ROIs

Guibas et al.⁶⁷

two-stage pipeline and pair-based GAN for medical image synthesis

RoiMammoGAN: one-stage and single-based GAN for breast cancer mammography image synthesis

Kazuhiro et al.¹⁴

DCGAN: based on deep convolutional. Applied to magnetic resonance (MR) images

RoiMammoGAN: based on deep convolutional-transpose network. Applied to breast images from mammograms using MIAS

Brock et al.⁶⁹

BigGAN: class-conditioning and orthogonal regularization was used in the generator to achieve fidelity and variety

RoiMammoGAN: class-label guided approach was used

Chen et al.⁶⁹

Combined conditional and unconditional GANs with adversarial training and self-supervision

RoiMammoGAN: based on conditional GAN with adversarial training and semi-supervision

Shaham et al.⁷⁰

SinGAN: unconditional GAN capable of learning from a single natural image without an accompanying label

RoiMammoGAN: learns from batch of images using conditional GAN approach

Wu et al.⁷¹

GP-GAN: leverage the strengths of the classical gradient-based for GAN

RoiMammoGAN: Adam gradient-based approach was used

Yi et al.¹⁰

A GAN can help explore and discover the underlying structure of medical images

RoiMammoGAN: can detect the structure of abnormalities in a digital mammogram (medical images)

Wu et al.⁷²

U-net-based GAN was designed to generate lesions on mammograms

RoiMammoGAN: was also designed to generate lesions on mammograms

Oyelade and Ezugwu⁷³

ArchGAN: capable of synthesizing mammograms with only architectural distortion

RoiMammoGAN: an advanced model of the ArchGAN