Laser & Optoelectronics Progress, Volume. 62, Issue 2, 0217001(2025)
Hierarchical Transformer with Multi-Scale Parallel Aggregation for Breast Tumor Segmentation
Figures & Tables(14)
Fig. 8. Visualization of the segmentation results for different methods on three datasets. (a) Original images; (b) mask; (c) U-Net; (d) Attention U-Net; (e) UNet++; (f) TransUNet; (g) TransFuse; (h) proposed
Fig. 9. Visualization of segmentation results for cases with large targets and low contrast in BUSI using different methods. (a) Original images; (b)mask; (c) U-Net;(d) Attention U-Net;(e) UNet++;(f) TransUNet;(g) TransFuse; (h) proposed
Fig. 10. Visualization of the segmentation results for cases with large targets and low contrast in the UDIAT using different methods. (a) Original images; (b)mask; (c) U-Net;(d) Attention U-Net;(e) UNet++;(f) TransUNet;(g) TransFuse; (h) proposed
Fig. 11. Intuitive comparison of Dice and IoU metrics with different segmentation methods on three datasets. (a) BUSI dataset; (b) UDIAT dataset; (c) BUS dataset
Fig. 12. Visualization of segmentation results of ablation studies on UDIAT dataset. (a) Original images; (b)mask; (c) U-Net; (d) MiT; (e) MiT+ RFB; (f) MiT+SA; (g) MiT+RFB+SA; (h) RBF+SA+Aggregation Block
Table 1. Segmentation results of different segmentation methods on three datasets
View tableTable 1. Segmentation results of different segmentation methods on three datasets
Dataset Metric U-Net Attention U-Net UNet++ TransUNet TransFuse Proposed BUSI Dice 0.7209 0.7453 0.7737 0.8032 0.8193 0.8514 IoU 0.6284 0.6547 0.6863 0.7181 0.7435 0.7754 Recall 0.7507 0.8314 0.8358 0.8502 0.8361 0.8680 Precision 0.7250 0.7291 0.7694 0.7977 0.8246 0.8505 Specificity 0.9812 0.9745 0.9744 0.9675 0.9675 0.9802 Accuracy 0.9512 0.9504 0.9574 0.9658 0.9658 0.9710 UDIAT Dice 0.7433 0.7913 0.7836 0.8310 0.8260 0.8418 IoU 0.6336 0.7029 0.6723 0.7411 0.7313 0.7484 Recall 0.7775 0.7953 0.8439 0.8349 0.8668 0.8981 Precision 0.7554 0.8156 0.7724 0.8634 0.8214 0.8295 Specificity 0.9915 0.9944 0.9916 0.9940 0.9918 0.9910 Accuracy 0.9785 0.9856 0.9810 0.9846 0.9865 0.9865 BUS Dice 0.9271 0.9315 0.9292 0.9371 0.9324 0.9517 IoU 0.8676 0.8745 0.8705 0.8840 0.8752 0.9081 Recall 0.9503 0.9382 0.9579 0.9664 0.9405 0.9591 Precision 0.9117 0.9312 0.9080 0.9143 0.8988 0.9514 Specificity 0.9317 0.9478 0.9267 0.9326 0.9162 0.9604 Accuracy 0.9380 0.9428 0.9392 0.9461 0.9405 0.9584
Table 2. Results of ablation experiments on BUSI and UDIAT datasets
View tableTable 2. Results of ablation experiments on BUSI and UDIAT datasets
Dataset Method Metric MiT RBF SA Aggregation Block Dice IoU Recall Precision Specificity Accuracy BUSI 0.7209 0.6284 0.7507 0.7250 0.9812 0.9512 √ 0.7866 0.6963 0.8079 0.8311 0.9833 0.9538 √ √ 0.8314 0.7512 0.8360 0.8523 0.9830 0.9660 √ √ 0.8238 0.7456 0.8456 0.8317 0.9775 0.9639 √ √ √ 0.8365 0.7533 0.8638 0.8421 0.9783 0.9663 √ √ √ √ 0.8514 0.7754 0.8680 0.8505 0.9802 0.9710 UDIAT 0.7433 0.6336 0.7775 0.7554 0.9915 0.9785 √ 0.7898 0.6925 0.8256 0.8209 0.9931 0.9833 √ √ 0.8205 0.7240 0.8439 0.8369 0.9927 0.9853 √ √ 0.8111 0.7145 0.8540 0.7824 0.9906 0.9850 √ √ √ 0.8291 0.7315 0.8896 0.8264 0.9922 0.9855 √ √ √ √ 0.8418 0.7484 0.8981 0.8295 0.9910 0.9865
Tools
Get Citation
Copy Citation Text
Ping Xia, Yudie Wang, Bangjun Lei, Cheng Peng, Guangyi Zhang, Tinglong Tang. Hierarchical Transformer with Multi-Scale Parallel Aggregation for Breast Tumor Segmentation[J]. Laser & Optoelectronics Progress, 2025, 62(2): 0217001
Paper Information
Category: Medical Optics and Biotechnology
Received: Mar. 6, 2024
Accepted: Apr. 25, 2024
Published Online: Jan. 9, 2025
The Author Email:
CSTR:32186.14.LOP240836
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20