Fig 1.
Examples of breast ultrasound images and associated masks in the BUSI dataset.
Table 1.
The number of images in each of the three kinds of breast cases.
Fig 2.
A sample image with multiple masks, and the single mask obtained after merging them.
Fig 3.
Overall pipeline of the proposed breast ultrasound image segmentation model.
Fig 4.
Sample original images from the BUSI dataset and their corresponding edge-enhanced images after undergoing the Roberts edge operator.
The top row of the dataset shows the original images, while the bottom row displays the corresponding images produced by applying the Roberts edge operator.
Fig 5.
The architecture of the proposed DBU-Net model used for tumor segmentation in breast ultrasound images.
Table 2.
Hyperparameter configurations to train the proposed model.
Table 3.
Segmentation performance comparison between GSU-Net and the proposed DBU-Net on the BUSI dataset.
Fig 6.
Performance comparison between the GSU-Net and the DBU-Net over the mean values of 5 folds.
Fig 7.
Learning curves (Loss vs Epochs, Recall vs Epochs, Precision vs Epochs, and Accuracy vs Epochs) of the proposed model on Fold 1 of experiments.
Table 4.
Performance comparison of the DBU-Net model with Prewitt and Sobel filters on the BUSI dataset.
All values are in %.
Fig 8.
Comparative study of IoU and Dice score concerning different DBU-net models based on the mean values of 5 folds.
Fig 9.
Analysis of predicted segmentation masks produced by different filter based DBU-Nets.
Table 5.
Wilcoxon signed rank test results of the proposed DBU-Net model.
Fig 10.
Some successful predictions by the proposed model.
Fig 11.
Some unsuccessful predictions by the proposed model.
Table 6.
Results of the proposed DBU-Net model with dice loss function using the BUSI dataset.
Fig 12.
Performance comparison of the proposed method using dice loss and the hybrid loss.
Table 7.
Comparative performance analysis of DBU-Net model with SOTA approaches for BUSI dataset, highlighting superior results in bold.
Fig 13.
Segmentation performance of the proposed model on the UDIAT dataset.
Table 8.
Performance comparison of the proposed DBU-Net model with previous models using the UDIAT dataset.