|
| Reference | | Method | Advantages | Disadvantages |
|
| Rouhi et al. [69] | 2015 | Regional growth with a cellular neural network with a specific threshold | Ability to diagnose benign and malignant tumors, high accuracy in classification | High computational complexity
(i) Lack of accurate detection of areas with tumor
(ii) Lack of comparison with previous methods |
| Kaymak et al. [70] | 2017 | Back-Propagation (BP) | A convenient way to use in neural network training, highly fast execution speed in training | Uncertainty of the exact type of approach proposed and lack of comparison at the time of classification and uncertainty of benign and malignant tumors |
| Karabatak [71] | 2015 | Naïve Bayesian | Ability to diagnose benign and malignant tumors, high accuracy in classification | High computational complexity
(i) Lack of accurate detection of areas with tumor
(ii) Lack of comparison with previous methods |
| Wang et al. [72] | 2018 | Regression-based methods | Ability to estimate and predict remaining life based on tumor size, high accuracy in detection | |
| Pereira et al. [73] | 2014 | Wavelet analysis and genetic algorithm | Ability to diagnose benign and malignant tumors, high accuracy in classification | High computational complexity
(i) Lack of accurate diagnosis of areas with tumor
(ii) Lack of comparison with previous methods |
| Cordeiro et al. [75] | 2016 | Semisupervised adaptive algorithm GrowCut | Ability to diagnose benign and malignant tumors, high accuracy in classification | High computational complexity
(i) Lack of accurate detection of areas with tumor
(ii) Lack of comparison with previous methods |
| Ahmed et al. [76] | 2020 | Mask RCNN | (i) Increased AUC for transfer learning
(ii) Use for X-ray mammographic image | (i) Low accuracy
(ii) Used low volume dataset for verification
(iii) High rate of oversampling |
| Lee et al. [77] | 2020 | Multiscale grid average pooling | (i) Utilizing global and local spatial feature
(ii) Novel attention module
(iii) Ultrasound image dataset | (i) Lower accuracy of segmentation
(ii) High computational complexity
(iii) A small volume of the dataset |
| Soulami et al. [78] | 2021 | UNet model | (i) High accuracy for breast cancer detection
(ii) High f1-score and AUC | (i) High complexity model
(ii) Overfitting in some models
(iii) Lower volume of analysis |
| Huang et al. [79] | 2021 | Fuzzy fully CNN | (i) Fuzzy membership function
(ii) Conditional random fields | (i) Low sensitivity
(ii) Low intersection over union
(iii) Low resolution and poor quality |
|
