Skip to content

Code for paper: DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification [ECCV 2024]

License

Notifications You must be signed in to change notification settings

ChongQingNoSubway/DGR-MIL

Repository files navigation

DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification (ECCV 2024) paper

Paper link (preprint): [https://arxiv.org/abs/2407.03575]

News 🔥

  • Aug 16, 2024: We will release the extracted features later and gradually improve this code repository !
  • June 17, 2024: Congratulations ! Paper has been accepted by ECCV 2024 !

Abstract. Multiple instance learning (MIL) stands as a powerful approach in weakly supervised learning, regularly employed in histological whole slide image (WSI) classification for detecting tumorous lesions. However, existing mainstream MIL methods focus on modeling correlation between instances while overlooking the inherent diversity among instances. However, few MIL methods have aimed at diversity modeling, which empirically show inferior performance but with a high computational cost. To bridge this gap, we propose a novel MIL aggregation method based on diverse global representation (DGR-MIL), by modeling diversity among instances through a set of global vectors that serve as a summary of all instances. First, we turn the instance correlation into the similarity between instance embeddings and the predefined global vectors through a cross-attention mechanism. This stems from the fact that similar instance embeddings typically would result in a higher correlation with a certain global vector. Second, we propose two mechanisms to enforce the diversity among the global vectors to be more descriptive of the entire bag: (i) positive instance alignment and (ii) a novel, efficient, and theoretically guaranteed diversification learning paradigm. Specifically, the positive instance alignment module encourages the global vectors to align with the center of positive instances (e.g., instances containing tumors in WSI). To further diversify the global representations, we propose a novel diversification learning paradigm leveraging the determinantal point process. The proposed model outperforms the state-of-the-art MIL aggregation models by a substantial margin on the CAMELYON-16 and the TCGA-lung cancer datasets.

Result. Demonstrating proposed method outperform the SOTA method across three groups extract feature based on the CAMELYON-16 and the TCGA-lung cancer datasets.

Enviroment Installation

Download Extracted Feature

C16 - resnet 50 : https://github.com/hrzhang1123/DTFD-MIL

C16 - resnet 18 : https://drive.google.com/drive/folders/1WxVhuLe1d062UbiwPy-a0AReRD-GftIz?usp=sharing

C16 - VIT: https://drive.google.com/file/d/1IjPX0MJQLuHW_hdJaRFBud7R5CBNARm8/view?usp=sharing

TCGA - resnet 50: https://github.com/hrzhang1123/DTFD-MIL

TCGA - resnet 18: https://drive.google.com/drive/folders/1NVXJKaFC39UJ_NexCNZm7520YCi8miMf?usp=sharing

TCGA - ViT: https://drive.google.com/drive/folders/1Ls-of2mQZKHCp6m3RMp32XYkVQ8WAIrv?usp=drive_link

Run the Code

Thanks for the code provided by:

Citation

If you find our work is useful in your research, please consider raising a star ⭐ and citing:

@article{zhu2024dgr,
  title={DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification},
  author={Zhu, Wenhui and Chen, Xiwen and Qiu, Peijie and Sotiras, Aristeidis and Razi, Abolfazl and Wang, Yalin},
  journal={arXiv preprint arXiv:2407.03575},
  year={2024}
}