FourierMIL: Fourier filtering-based multiple instance learning for whole slide image analysis

This work is published in the International Journal of Computer Vision (https://doi.org/10.1007/s11263-025-02679-x).

Introduction

This repository contains a PyTorch implementation of FourierMIL.

Saliency mapping

Visualization of nodal metastases detection. Panels (a-c) compare the ground truth to CLAM and FourierMIL saliency heatmaps between two cases, highlighting expert annotations in green. Panels (d-f) depict a challenging case with a small ROI (<1%), showcasing the model’s precision and the low rate of false positives in detecting subtle cancerous regions. The heatmap color intensities reflect each model’s attention to the detected areas

Feature generation

Create features using CLAM

Train

bash train.sh

Test

bash test.sh

Thanks to TransMIL for creating a great repo.