Fig 1.
Sample breast ultrasound images of benign, malignant, and normal types.
Fig 2.
Block diagram of the proposed DAU-Net model used for segmentation of tumor in breast ultrasound images.
An input image with dimensions 128 × 128 × 1 undergoes feature extraction through the encoder, and the decoder then performs upsampling on the encoded features to predict a binary mask of size 128 × 128 × 1. The in-between connections of the encoder and the decoder are accompanied by the addition of PCBAM and SWA attention mechanisms to enhance the performance.
Fig 3.
An illustration of the PCBAM attention block.
CBAM and PAM are applied to the input feature F. The addition of the outputs of CBAM and PAM is the output of the PCBAM attention mechanism, FPCBAM.
Table 1.
Performance metrics of the segmentation models.
All values are in %. Bold values indicate superior performance. The results are in x(±y) format, where x is the mean and y is the standard deviation of the evaluation metric for the five runs of the model.
Fig 4.
Results of the ablation study indicate the improvement in model performance with each experimental modification.
GT and PM are Ground Truth and Predicted Mask, respectively. Fc is the heatmap of the bottleneck layer and it demonstrates the improvement of the model’s performance in focusing on the region of interest after the addition of the SWA in the bottleneck layer. Fa and Fb are heatmaps of the features flowing from the first and second encoder layers to the first and second decoder layers via skip connections. It can be seen that Fa and Fb get more enriched with the use of attentions such as CBAM, PAM, and PCBAM.
Table 2.
Results of the proposed DAU-Net model with 5-fold cross-validation on the BUSI dataset.
Table 3.
Results of the Mann-Whitney U test of the proposed DAU-Net model used for segmenting tumor regions in breast images of the BUSI dataset.
Table 4.
Performance metrics of the proposed model with different loss functions.
Table 5.
Performance comparison with standard segmentation models.
All values are in %. Bold values indicate superior performance.
Table 6.
Performance comparison with SOTA models.
All values are in %. Bold values indicate superior performance.
Fig 5.
Results of the proposed segmentation model on images of the BUSI dataset and the heatmaps of SWA and PCBAM layers.
PCBAM1 corresponds to the PCBAM layer just above the SWA layer, PCBAM2 corresponds to the PCBAM layer just above PCBAM1 layer, and PCBAM3 corresponds to the PCBAM layer just above PCBAM2 layer.
Fig 6.
Illustration of some of the failed cases of our model.
The encircled regions are the misclassified segmented masks. GT and PM represent the Ground Truth and Predicted Mask, respectively.
Fig 7.
Predicted mask and heatmap visualization of the proposed model on the UDIAT dataset.
GT and PM represent the Ground Truth and Predicted Mask, respectively. F a, F b, and F c are the heatmaps of the features flowing from the first and second encoder layers to the first and second decoder layers via skip connections and the bottleneck layer, respectively.
Table 7.
Performance comparison of the proposed model with past methods on UDIAT dataset.
All values are in %. Bold values indicate superior performance.