Fig 1.
The difference between traditional pooling method and the proposed pooling method.
Fig 2.
MaxPooling of three different matrices is the same.
Fig 3.
Vector pooling block.
Fig 4.
Example of calculating max-pooling and average pooling with filter of size 2X2 and stride 2.
Fig 5.
Table 1.
Dataset description.
Fig 6.
Examples of segmentation process with U-Net before and after using the proposed pooling blocks.
(a) original images. (b) labels. (c)segmentation with U-Net. (d)segmentation with U-Net+VPB. (e) segmentation with U-Net+ AVG-MAX VPB.
Fig 7.
Illustrative chart for the evaluation metrics of the semantic segmentation process with U-net before and after using the VPB and AVG-MAX VPB presented in Table 2.
Table 2.
Semantic segmentation evaluation metrics of U-net before and after using VPB and AVG-MAX VPB.
Fig 8.
Illustrative chart for the accuracy results for three CNNs before and after using VPB and AVG-MAX VPB presented in Table 3.
Fig 9.
Illustrative chart for the sensitivity results for three CNNs before and after using VPB and AVG-MAX VPB presented in Table 3.
Fig 10.
Illustrative chart for the specificity results for three CNNs before and after using VPB and AVG-MAX VPB presented in Table 3.
Table 3.
Evaluation metrics of the classification process on three CNNs before and after using the VPB and the AVG-MAX VPB.
Table 4.
Comparison with other studies on breast cancer detection with CNNs.