Table 2.

Summary of related works.

Reference	Method	Dataset	Limitations
[22], 2018	Traditional feature extraction	CBIS-DDSM	High computational demand and long training episode (8 h)
[23], 2019	Gradual fine-tuning with episodes of learning rate annealing schedules	CBIS-DDSM	High computational demand and long training episode (99 epochs)
[24], 2018	Traditional fine-tuning	CBIS-DDSM	High computational demand, low AUC, and overfitting
[25], 2019	Traditional fine-tuning	CBIS-DDSM	High computational demand, low AUC, and overfitting
[26], 2021	Deep adversarial domain adaptation	CBIS-DDSM	Complex algorithm with high computational demand and long training episode (400 epochs)
[27], 2022	Feature extraction plus feature selection using twin algorithms: reformed differential evaluation and reformed gray wolf algorithm.	Breast ultrasound images	Long training episodes and additional computation burden introduced by feature selection algorithms
[28], 2021	Multifractal dimension feature extraction, feature reduction using GA, and classification using ANN	DDSMMini-MIASINBreastbreast cancer digital repository	Not end-to-end trained; each algorithm introduced computational bottlenecks that aggregated to high computational demandNot compatible with SOTA CNN models