Rethinking Transformer for Long Contextual Histopathology Whole Slide Image Analysis

Authors: Honglin Li, Yunlong Zhang, Pingyi Chen, Zhongyi Shui, Chenglu Zhu, Lin Yang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our method, Long-contextual MIL (Long MIL), is evaluated through extensive experiments on various WSI tasks to validate its superiority in: 1) overall performance, 2) memory usage and speed, and 3) extrapolation ability compared to previous methods.
Researcher Affiliation Academia Honglin Li1,3 Yunlong Zhang1,3 Pingyi Chen1,3 Zhongyi Shui1,3 Chenglu Zhu2,3 Lin Yang2,3 1 Zhejiang University 2 Research Center for Industries of the Future and 3 School of Engineering, Westlake University
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks clearly labeled as 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Our code will be available at https://github.com/invoker-LL/Long-MIL.
Open Datasets Yes We use four datasets to evaluate our method for both tumor subtyping and survival prediction. For data details and pre-processing, please see Appendix A.4. BReAst Carcinoma Subtyping (BRACS) [4]... The Cancer Genome Atlas Breast Cancer (TCGA-BRCA) [68, 55]... TCGA-COADREAD... TCGA-STAD...
Dataset Splits Yes For TCGA-BRCA, we perform 10-fold cross-validation with the same data split adopted in HIPT [9]. Besides, the dataset BRACS is officially split into training, validation and testing, thus the experiment is conducted 5-times with different random seeds.
Hardware Specification Yes We train our model with Py Torch on a RTX-3090 GPU
Software Dependencies No The paper mentions 'Py Torch' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes We train our model with Py Torch on a RTX-3090 GPU, with a WSI-level batchsize of 1, learning rate of 1e-4, and weight decay of 1e-2. We add positional encoding into the framework, please check our code for details.