Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification
Abstract Digital medical imaging, particularly pathology images, is essential for cancer diagnosis but faces challenges in direct model training due to its super-resolution nature. Although weakly supervised learning has reduced the need for manual annotations, many multiple instance learning (MIL)...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-03-01
|
| Series: | Complex & Intelligent Systems |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s40747-025-01779-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850042321609949184 |
|---|---|
| author | Bin Yang Lei Ding Jianqiang Li Yong Li Guangzhi Qu Jingyi Wang Qiang Wang Bo Liu |
| author_facet | Bin Yang Lei Ding Jianqiang Li Yong Li Guangzhi Qu Jingyi Wang Qiang Wang Bo Liu |
| author_sort | Bin Yang |
| collection | DOAJ |
| description | Abstract Digital medical imaging, particularly pathology images, is essential for cancer diagnosis but faces challenges in direct model training due to its super-resolution nature. Although weakly supervised learning has reduced the need for manual annotations, many multiple instance learning (MIL) methods struggle to effectively capture crucial spatial relationships in histopathological images. Existing methods incorporating positional information often overlook nuanced spatial correlations or use positional encoding strategies that do not fully capture the unique spatial dynamics of pathology images. To address this issue, we propose a new framework named TMIL (Transformer-based Multiple Instance Learning Network with 2D positional encoding), which leverages multiple instance learning for weakly supervised classification of histopathological images. TMIL incorporates a 2D positional encoding module, based on the Transformer, to model positional information and explore correlations between instances. Furthermore, TMIL divides histopathological images into pseudo-bags and trains patch-level feature vectors with deep metric learning to enhance classification performance. Finally, the proposed approach is evaluated on a public colorectal adenoma dataset. The experimental results show that TMIL outperforms existing MIL methods, achieving an AUC of 97.28% and an ACC of 95.19%. These findings suggest that TMIL’s integration of deep metric learning and positional encoding offers a promising approach for improving the efficiency and accuracy of pathology image analysis in cancer diagnosis. |
| format | Article |
| id | doaj-art-5ae0bf642121442697e9a6313aedf30a |
| institution | DOAJ |
| issn | 2199-4536 2198-6053 |
| language | English |
| publishDate | 2025-03-01 |
| publisher | Springer |
| record_format | Article |
| series | Complex & Intelligent Systems |
| spelling | doaj-art-5ae0bf642121442697e9a6313aedf30a2025-08-20T02:55:36ZengSpringerComplex & Intelligent Systems2199-45362198-60532025-03-0111511710.1007/s40747-025-01779-yTransformer-based multiple instance learning network with 2D positional encoding for histopathology image classificationBin Yang0Lei Ding1Jianqiang Li2Yong Li3Guangzhi Qu4Jingyi Wang5Qiang Wang6Bo Liu7Center for Strategic Assessment and Consulting, Academy of Military ScienceFaculty of Information Technology, Beijing University of TechnologyFaculty of Information Technology, Beijing University of TechnologyFaculty of Information Technology, Beijing University of TechnologyComputer Science and Engineering Department, Oakland UniversityFaculty of Information Technology, Beijing University of TechnologyFaculty of Information Technology, Beijing University of TechnologySchool of Mathematical and Computational Sciences, Massey UniversityAbstract Digital medical imaging, particularly pathology images, is essential for cancer diagnosis but faces challenges in direct model training due to its super-resolution nature. Although weakly supervised learning has reduced the need for manual annotations, many multiple instance learning (MIL) methods struggle to effectively capture crucial spatial relationships in histopathological images. Existing methods incorporating positional information often overlook nuanced spatial correlations or use positional encoding strategies that do not fully capture the unique spatial dynamics of pathology images. To address this issue, we propose a new framework named TMIL (Transformer-based Multiple Instance Learning Network with 2D positional encoding), which leverages multiple instance learning for weakly supervised classification of histopathological images. TMIL incorporates a 2D positional encoding module, based on the Transformer, to model positional information and explore correlations between instances. Furthermore, TMIL divides histopathological images into pseudo-bags and trains patch-level feature vectors with deep metric learning to enhance classification performance. Finally, the proposed approach is evaluated on a public colorectal adenoma dataset. The experimental results show that TMIL outperforms existing MIL methods, achieving an AUC of 97.28% and an ACC of 95.19%. These findings suggest that TMIL’s integration of deep metric learning and positional encoding offers a promising approach for improving the efficiency and accuracy of pathology image analysis in cancer diagnosis.https://doi.org/10.1007/s40747-025-01779-yWeakly supervised trainingImage classificationMultiple instance learning |
| spellingShingle | Bin Yang Lei Ding Jianqiang Li Yong Li Guangzhi Qu Jingyi Wang Qiang Wang Bo Liu Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification Complex & Intelligent Systems Weakly supervised training Image classification Multiple instance learning |
| title | Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification |
| title_full | Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification |
| title_fullStr | Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification |
| title_full_unstemmed | Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification |
| title_short | Transformer-based multiple instance learning network with 2D positional encoding for histopathology image classification |
| title_sort | transformer based multiple instance learning network with 2d positional encoding for histopathology image classification |
| topic | Weakly supervised training Image classification Multiple instance learning |
| url | https://doi.org/10.1007/s40747-025-01779-y |
| work_keys_str_mv | AT binyang transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT leiding transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT jianqiangli transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT yongli transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT guangzhiqu transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT jingyiwang transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT qiangwang transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification AT boliu transformerbasedmultipleinstancelearningnetworkwith2dpositionalencodingforhistopathologyimageclassification |