Table 1.
Comparison of state of the art approaches.
Fig 1.
Modified ResNet-50 model: (a) Architecture, (b) block diagram.
Table 2.
ResNet-50 and modified ResNet-50 comparison.
Fig 2.
Work flow of proposed methodology.
Table 3.
Hyperparameters, their values and settings.
Table 4.
Performance measures and their equations.
Fig 3.
Breast cancer images of dataset (a) adenosis (b) ductal carcinoma.
Table 5.
Classification results at 40 magnification, before feature optimization on BreakHis dataset.
Boldface values depict significant results.
Table 6.
Classification results at 40 magnification, after feature optimization on BreakHis dataset.
Boldface values depict significant results.
Table 7.
Classification results at 100 magnification, before feature optimization on BreakHis dataset.
Table 8.
Classification results at 100 magnification, after feature optimization on BreakHis dataset.
Table 9.
Classification results at 200 magnification, before feature optimization on BreakHis dataset.
Table 10.
Classification results at 200 magnification, after feature optimization on BreakHis dataset.
Table 11.
Classification results at 400 magnification, before feature optimization on BreakHis dataset.
Table 12.
Classification results at 400 magnification, after feature optimization on BreakHis dataset.
Fig 4.
Training graphs of BreakHis dataset with respective magnification levels.
Fig 5.
Confusion matrix of classifiers at 40 magnification (a) before optimization (b) after optimization.
Fig 6.
Confusion matrix of classifiers at 100 magnification (a) before optimization (b) after optimization.
Fig 7.
Confusion Matrix of classifiers at 200 magnification (a) before optimization (b) after optimization.
Fig 8.
Confusion matrix of classifiers at 400 magnification (a) before optimization (b) after optimization.
Fig 9.
Accuracy graphs of classifiers at 40 and 100
.
Fig 10.
Accuracy graphs of classifiers at 200 and 400
.
Fig 11.
Error graphs of classifiers.
Fig 12.
Precision and recall graphs of classifiers at 40 and 100
.
Fig 13.
Precision and recall graphs of classifiers at 200 and 400
.
Fig 14.
Kappa, MCC and F1_Score graphs of classifiers at 40 and 100
.
Fig 15.
Kappa, MCC and F1_Score graphs of classifiers at 200 and 400
.
Fig 16.
Training time graphs of classifiers.
Fig 17.
Percent improvement in error, recall, specificity, precision and accuracy.
Table 13.
Comparison of results with state-of-the-art approaches.
Fig 18.
Performance comparison of ResNet-LASSO, ResNet-FOA and hybrid ResNet-FOA-LASSO.
Table 14.
Comparison of ResNet-based feature selection methods in terms of training epochs, time, final performance (MSE), and gradient.
Lower MSE and gradient indicate better accuracy and convergence.
Table 15.
Intra-class, inter-class dissimilarity, class margin, and classification accuracy for breast cancer classification.