Table 1.
Summary of related work.
| Author | Year | Type | Network | Result | Advantages | Disadvantages |
|
| ||||||
| Dong et al. [22] | 2022 | Breast cancer diagnosis and classification | Random forest and regression tree | The application of machine learning techniques like CART and random forests coupled with geographical methodologies provides a viable alternative for future inequalities studies | (i) Low complexity (ii) High accuracy |
(i) Possible overfitting (ii) Used classic feature extraction (iii) Lowest robustness |
|
| ||||||
| Guha et al. [19] | 2022 | Breast cancer risk factors | SEER-Medicare analysis | The incidence of AF in women after a breast cancer diagnosis is much higher. AF is strongly linked to a higher stage of breast cancer upon diagnosis. Women newly diagnosed with breast cancer who develop AF suffer an increased risk of cardiovascular death but not breast-cancer-related death | (i) Ability of risk assessment (ii) Technical assessment |
(i) Needs feature extraction (ii) Unable to diagnose the patient |
|
| ||||||
| Chamieh et al. [63] | 2022 | Breast cancer diagnosis using fine-needle aspiration cytology | Begg and Greenes method | Irrespective of the recommended technique, the FNAC test's specificity was always greater than its sensitivity. For all approaches, the probability ratios were positive. Both positive and negative yields were high, demonstrating the test's exact discriminating qualities. | (i) Technical assessment method (ii) Low complexity |
(i) Unable to diagnose illness type (ii) Limited dataset |
|
| ||||||
| Thangarajan et al. [64] | 2022 | Breast cancer biomarkers validated in plasma | BC gene expression profiling | Methylation status of SOSTDC1, DACT2, and WIF1 can distinguish BC from benign and control conditions with 100% sensitivity and 91% specificity. Therefore, SOSTDC1, DACT2, and WIF1 may be used as a supplemental diagnostic tool to distinguish noninvasive and invasive breast cancer from benign breast conditions and healthy individuals | (i) Using biomarkers instead of mathematical features | (i) Lower sensitivity (ii) Unable to diagnose illness type |
|
| ||||||
| Chakravarthy et al. [65] | 2022 | Diagnosis of breast cancer | Multideep CNN | By fuzzing deep features for both datasets (97.93 percent for MIAS and 96.646 percent for INbreast), we achieved the highest classification accuracy among state-of-the-art frameworks. When the PCA was applied to combined deep features, classification performance did not improve, but execution time was shorter, resulting in a lower computing cost | (i) Low complexity (ii) High accuracy (iii) Technical assessment method |
(i) Possible overfitting (ii) Needs feature extraction (iii) Lower sensitivity (iv) Lowest robustness |
| Wang et al. [66] | 2022 | Metastasis of breast cancer axillary lymph nodes forecasting | CNN | The T2WI sequence outperformed the other three sequences in the testing set, with accuracy and AUC of 0.933/0.989. In comparison with T1WI, which has accuracy and AUC of 0.691/0.806, the increase is substantial | (i) Ability of risk assessment (ii) Technical assessment |
(i) Unable to diagnose the patient (ii) Used classic feature extraction (iii) Limited dataset (iv) Lower sensitivity |
|
| ||||||
| Melekoodappattu et al. [67] | 2022 | Breast cancer detection | CNN and texture feature-based approach | Using our ensemble method, we measured 97.8% specificity and 98.6% accuracy for MIAS and 98.3% and 97.9% for DDSM. Experimental data indicate that the combination strategy enhances measurement metrics independently for each phase | (i) Low complexity (ii) High accuracy |
(i) Unable to diagnose illness type |
|
| ||||||
| Gonçalves et al. [68] | 2022 | Breast cancer detection | CNNs | VGG-16 produced F1-scores greater than 0.90 for all three networks, an increase from 0.66 to 0.92. Furthermore, compared to earlier research, we were able to improve the F1-score of ResNet-50 from 0.83 to 0.90 | (i) Comparative study (ii) Used high-rank methods |
(i) Unable to diagnose illness type (ii) High complexity |