|
| References and year | Dataset | No. of samples | Technique | Key points | Limitation |
|
| Wu et al. [18] | Optimam mammography image Database | 26456 | Traditional + contextual GAN | The model performance on malignancy classification was improved | Classification of normal and malignant only |
| Desai et al. [19] | DDSM | 287 | DCGAN | The authors show that GAN is a workable choice for training such models with a data shortage | The model is evaluated with only two batch sizes |
| Alyafi et al. [17] | OPTIMAM mammography image Database (OMI-DB) | 80000 | DCGAN | The work can be extended to other similar tasks | Their work is limited to small patches of mammograms |
| Shen et al. [20] | DDSM and local dataset collected from Nanfang Hospital, China | 11218 | GAN | The model is a viable option for generating labelled breast images | Due to the inherent complexity and variability of breast tissue structures, this type of GAN might not be able to produce various realistic images |
| Swiderski et al. [21] | DDSM | 11218 | AGAN | The novelty of the model in data augmentation as compared to other deep learning models | Modified GAN is based on autoencoder architecture, which may misclassify key features for BC diagnosis |
| Lee et al. [22] | The local dataset was collected from the University of Pittsburgh Medical center, USA | 1366 | CGAN | The system can identify patients for further screening in the early detection of MO-related cancer | They did not consider benchmark datasets |
| Park et al. [23] | Local dataset from Asan medical center, Korea | 105,948 | StyleGAN2 | Their model has comparable fidelity to real mammograms | The system was only limited to normal mammographic images |
|
