Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model
BackgroundAccurate segmentation of thyroid nodules in ultrasound imaging remains a significant challenge in medical diagnostics, primarily due to edge blurring and substantial variability in nodule size. These challenges directly affect the precision of thyroid disorder diagnoses, which are crucial...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2025-02-01
|
| Series: | Frontiers in Oncology |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fonc.2025.1456563/full |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832087435874402304 |
|---|---|
| author | Changan Yang Muhammad Awais Ashraf Mudassar Riaz Pascal Umwanzavugaye Kavimbi Chipusu Kavimbi Chipusu Hongyuan Huang Yueqin Xu |
| author_facet | Changan Yang Muhammad Awais Ashraf Mudassar Riaz Pascal Umwanzavugaye Kavimbi Chipusu Kavimbi Chipusu Hongyuan Huang Yueqin Xu |
| author_sort | Changan Yang |
| collection | DOAJ |
| description | BackgroundAccurate segmentation of thyroid nodules in ultrasound imaging remains a significant challenge in medical diagnostics, primarily due to edge blurring and substantial variability in nodule size. These challenges directly affect the precision of thyroid disorder diagnoses, which are crucial for metabolic and hormonal regulation.MethodsThis study proposes a novel segmentation approach utilizing a Swin U-Net architecture enhanced with a self-attention mechanism. The model integrates residual and multiscale convolutional structures in the encoder path, with long skip connections feeding into an attention module to improve edge preservation and feature extraction. The decoder path employs these refined features to achieve precise segmentation. Comparative evaluations were conducted against traditional models, including U-Net and DeepLabv3+.ResultsThe Swin U-Net model demonstrated superior performance, achieving an average Dice Similarity Coefficient (DSC) of 0.78, surpassing baseline models such as U-Net and DeepLabv3+. The incorporation of residual and multiscale convolutional structures, along with the use of long skip connections, effectively addressed issues of edge blurring and nodule size variability. These advancements resulted in significant improvements in segmentation accuracy, highlighting the model’s potential for addressing the inherent challenges of thyroid ultrasound imaging.ConclusionThe enhanced Swin U-Net architecture exhibits notable improvements in the robustness and accuracy of thyroid nodule segmentation, offering considerable potential for clinical applications in thyroid disorder diagnosis. While the study acknowledges dataset size limitations, the findings demonstrate the effectiveness of the proposed approach. This method represents a significant step toward more reliable and precise diagnostics in thyroid disease management, with potential implications for enhanced patient outcomes in clinical practice. |
| format | Article |
| id | doaj-art-e5b64b81d80f46989f528e3d75d788ac |
| institution | Kabale University |
| issn | 2234-943X |
| language | English |
| publishDate | 2025-02-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Oncology |
| spelling | doaj-art-e5b64b81d80f46989f528e3d75d788ac2025-02-06T05:21:45ZengFrontiers Media S.A.Frontiers in Oncology2234-943X2025-02-011510.3389/fonc.2025.14565631456563Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net modelChangan Yang0Muhammad Awais Ashraf1Mudassar Riaz2Pascal Umwanzavugaye3Kavimbi Chipusu4Kavimbi Chipusu5Hongyuan Huang6Yueqin Xu7Department of Thyroid and Breast Surgery, Jinjiang Municipal Hospital (Shanghai Sixth People’s Hospital Fujian), Quanzhou, Fujian, ChinaDepartment of Mechanical Engineering, Division of Biomedical Engineering, University of Saskatchewan, Saskatoon, SK, CanadaDepartment of Computer Science, Central South University, Changsha, Hunan, ChinaDepartment of Computer Science, Central South University, Changsha, Hunan, ChinaDepartment of Mechanical Engineering, Division of Biomedical Engineering, University of Saskatchewan, Saskatoon, SK, CanadaDepartment of Computer Science, Central South University, Changsha, Hunan, ChinaDepartment of Thyroid and Breast Surgery, Jinjiang Municipal Hospital (Shanghai Sixth People’s Hospital Fujian), Quanzhou, Fujian, ChinaDepartment of Thyroid and Breast Surgery, Jinjiang Municipal Hospital (Shanghai Sixth People’s Hospital Fujian), Quanzhou, Fujian, ChinaBackgroundAccurate segmentation of thyroid nodules in ultrasound imaging remains a significant challenge in medical diagnostics, primarily due to edge blurring and substantial variability in nodule size. These challenges directly affect the precision of thyroid disorder diagnoses, which are crucial for metabolic and hormonal regulation.MethodsThis study proposes a novel segmentation approach utilizing a Swin U-Net architecture enhanced with a self-attention mechanism. The model integrates residual and multiscale convolutional structures in the encoder path, with long skip connections feeding into an attention module to improve edge preservation and feature extraction. The decoder path employs these refined features to achieve precise segmentation. Comparative evaluations were conducted against traditional models, including U-Net and DeepLabv3+.ResultsThe Swin U-Net model demonstrated superior performance, achieving an average Dice Similarity Coefficient (DSC) of 0.78, surpassing baseline models such as U-Net and DeepLabv3+. The incorporation of residual and multiscale convolutional structures, along with the use of long skip connections, effectively addressed issues of edge blurring and nodule size variability. These advancements resulted in significant improvements in segmentation accuracy, highlighting the model’s potential for addressing the inherent challenges of thyroid ultrasound imaging.ConclusionThe enhanced Swin U-Net architecture exhibits notable improvements in the robustness and accuracy of thyroid nodule segmentation, offering considerable potential for clinical applications in thyroid disorder diagnosis. While the study acknowledges dataset size limitations, the findings demonstrate the effectiveness of the proposed approach. This method represents a significant step toward more reliable and precise diagnostics in thyroid disease management, with potential implications for enhanced patient outcomes in clinical practice.https://www.frontiersin.org/articles/10.3389/fonc.2025.1456563/fullSwin U-Netimage segmentationdeep learningimage datasetthyroidultrasound images |
| spellingShingle | Changan Yang Muhammad Awais Ashraf Mudassar Riaz Pascal Umwanzavugaye Kavimbi Chipusu Kavimbi Chipusu Hongyuan Huang Yueqin Xu Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model Frontiers in Oncology Swin U-Net image segmentation deep learning image dataset thyroid ultrasound images |
| title | Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model |
| title_full | Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model |
| title_fullStr | Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model |
| title_full_unstemmed | Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model |
| title_short | Improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self-attention mechanism-based Swin U-Net model |
| title_sort | improving diagnostic precision in thyroid nodule segmentation from ultrasound images with a self attention mechanism based swin u net model |
| topic | Swin U-Net image segmentation deep learning image dataset thyroid ultrasound images |
| url | https://www.frontiersin.org/articles/10.3389/fonc.2025.1456563/full |
| work_keys_str_mv | AT changanyang improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT muhammadawaisashraf improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT mudassarriaz improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT pascalumwanzavugaye improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT kavimbichipusu improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT kavimbichipusu improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT hongyuanhuang improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel AT yueqinxu improvingdiagnosticprecisioninthyroidnodulesegmentationfromultrasoundimageswithaselfattentionmechanismbasedswinunetmodel |