Fig 1.
The DenseNet-121 architecture.
Fig 2.
Overview of the proposed IDSNet architecture.
Fig 3.
The dense block.
Fig 4.
The transition layer.
Fig 5.
The SENet architecture.
Fig 6.
The classification sub-network.
Fig 7.
Image samples from BreakHis dataset.
(a)~(d) benign tumor and (e)~(f) malignant tumor with the magnification factor of 40×, 100×, 200×, 400×.
Table 1.
Image distribution by magnification factor and class.
Table 2.
Performance of BC classification using VGG16, Resnet50 and DenseNet-121.
The best performance is highlighted by boldface.
Table 3.
Performance of BC classification by the different network combinations: Case1) the complete IDSNet based on DenseNet interleaved with SENet and classification sub-network; Case2) the DenseNet combined with classification sub-network only; Case3) DenseNet combined with SENet module only.
The best performance is highlighted by boldface.
Table 4.
Performance of BC classification for the proposed IDSNet in comparison with other literature methods.
The best performance is highlighted by boldface.
Table 5.
Parameter and model size for the different CNN models used to test the BreakHis dataset.