This repository covers the official implementation of our Dual-Query multiple instance learning approach for histopathological image analysis, a BMVC 2023 (Oral) paper. A novel method for histopathological slide assessment, extending on the perceiver architecture and leveraging a dynamic meta-embedding strategy.

Dual-Query Multiple Instance Learning for Dynamic Meta-Embedding based Tumor Classification
Simon Holdenried-Krafft¹, Peter Somers³, Ivonne A. Montes-Majarro², Diana Silimon², Cristina Tarín³, Falko Fend², Hendrik P. A. Lensch^1
1University of Tübingen, ²University Hospital of Tübingen, ³University of Stuttgart

A conda environment is used for dependency management

conda create -n dqmil python=3.7
conda activate dqmil
pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install -r requirements.txt

The configuration is based on hydra. The settings can be found in the "configs" folder.

To train the Dual-Query Perceiver run:

python ./src/train.py 

The implementation relies on lmdbs. The corresponding code to create the datasets will be released in the upcoming weeks.

If you find this code useful, please consider citing:

@inproceedings{Holdenried-Krafft_2023_BMVC,
author    = {Simon Holdenried-Krafft and Peter Somers and Ivonne Montes-Mojarro and Diana Silimon and Cristina Tarín and Falko Fend and Hendrik P. A. Lensch},
title     = {Dual-Query Multiple Instance Learning for Dynamic Meta-Embedding based Tumor Classification},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {https://papers.bmvc2023.org/0575.pdf}
}