Weakly Supervised Multiple Instance Learning Histopathological Tumor Segmentation

Marvin Lerousseau, Maria Vakalopoulou, Marion Classe, Julien Adam, Enzo Battistella, Alexandre Carré, Théo Estienne, Théophraste Henry, Eric Deutsch, Nikos Paragios

Introduction

Training deep learning models for histopathological image analysis typically requires extensive manual annotation. In digital pathology, Whole Slide Images (WSIs) are gigapixel-scale files, making pixel-level manual tracing of tumor boundaries—a prerequisite for standard fully supervised segmentation—an incredibly time-consuming and non-scalable process. This paper addresses the "segmentation gap" by proposing a weakly supervised framework that leverages slide-level binary labels (tumor vs. normal) to train high-quality spatial segmentation models.

Methodology: The Proxy-Labeling Meta-Algorithm

The core innovation is a meta-algorithm designed to convert slide-level diagnostic labels into patch-level training signals using a Multiple Instance Learning (MIL) strategy.

The four-step training pipeline: 1) Patch selection, 2) Inference, 3) Proxy label construction based on $\alpha$ and $\beta$ percentiles, and 4) Backpropagation.

The process follows four main steps:

  1. MIL Setup: A slide is treated as a "bag" of small image patches ("instances"). For a "normal" slide, all patches are assumed normal. For a "tumor" slide, at least one patch is tumorous.
  2. Inference: A ResNet50 backbone (pre-trained on ImageNet) performs a forward pass on a batch of patches.
  3. Proxy Labeling: Two hyperparameters, $\alpha$ and $\beta$, control the training signal. In a tumor-positive slide, the top $\alpha\%$ of patches with the highest predicted probabilities are assigned a "tumor" label (1), while the bottom $\beta\%$ are assigned a "normal" label (0). Patches in the middle are discarded to minimize noise.
  4. Optimization: Binary cross-entropy loss is computed only on these proxy-labeled patches, and the model is updated via backpropagation.

Experimental Setup and Datasets

The study utilized 6,481 flash-frozen WSIs from The Cancer Genome Atlas (TCGA), covering Kidney, Bronchus/Lung, and Breast tissue. Evaluation was conducted on:

  • In-Distribution Test Set: 130 slides with expert manual annotations.
  • Out-of-Location Test Set: 100 slides from unseen organs (Colon, Ovary, Uterus) to test generalization.
  • Benchmark: The PatchCamelyon dataset was used to compare against fully supervised methods.

Key Insights and Results

The model's performance is highly sensitive to the control parameters $\alpha$ and $\beta$. The parameter $\alpha$, representing the minimum assumed tumor area, acts as a lever for recall; increasing $\alpha$ forces the model to identify more tumor tissue, often increasing recall at the cost of precision.

Visual impact of varying $\alpha$ and $\beta$ on tumor segmentation maps. Black regions represent predicted tumor, white represents normal, and the ground truth is shown for comparison.

The best configuration ($\alpha=0.2, \beta=0.2$) achieved a pixel-wise AUC of 0.804 on the in-distribution test set. Remarkably, the model demonstrated "pan-cancer" generalization, maintaining an average AUC of 0.679 on unseen tissue types. On the PatchCamelyon benchmark, it reached an AUC of 0.802—a significant achievement for a model that never saw a pixel-level annotation during training.

Conclusion

This work demonstrates that "the power of extremes"—training on the most and least certain predictions—allows models to learn visual features of cancer through a self-correcting loop. While a performance gap remains compared to fully supervised models, this approach significantly lowers the barrier to entry for large-scale histopathological segmentation. The authors have released over 6,000 automatically generated tumor maps to support further research in the community.

Reference: Lerousseau, M., et al. "Weakly supervised multiple instance learning histopathological tumor segmentation." arXiv:2004.05024v4, 2021.

Summary updated

Abstract

Histopathological image segmentation is a challenging and important topic in medical imaging with tremendous potential impact in clinical practice. State of the art methods rely on hand-crafted annotations which hinder clinical translation since histology suffers from significant variations between cancer phenotypes. In this paper, we propose a weakly supervised framework for whole slide imaging segmentation that relies on standard clinical annotations, available in most medical systems. In particular, we exploit a multiple instance learning scheme for training models. The proposed framework has been evaluated on multi-locations and multi-centric public data from The Cancer Genome Atlas and the PatchCamelyon dataset. Promising results when compared with experts' annotations demonstrate the potentials of the presented approach. The complete framework, including $6481$ generated tumor maps and data processing, is available at https://github.com/marvinler/tcga_segmentation.
BibTeX
@inproceedings{lerousseau2020weakly,
  author = {Marvin Lerousseau and Maria Vakalopoulou and Marion Classe and Julien Adam and Enzo Battistella and Alexandre Carré and Théo Estienne and Théophraste Henry and Eric Deutsch and Nikos Paragios},
  title = {{Weakly Supervised Multiple Instance Learning Histopathological Tumor Segmentation}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2020},
  doi = {10.1007/978-3-030-59722-1\_45},
}
Figure