Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Generalized Category Discovery for Brain Tumor Classification in Digital Pathology

Matthias Perkonigg, Patrick Rockenschaub, Georg Göbel, Adelheid Wöhrer

TL;DR

This work tackles the problem of brain tumor classification when not all subtypes are labeled by leveraging Generalized Category Discovery with a respect for hierarchical tumor taxonomy. HGCD-BT combines a contrastive learning branch with a semi-supervised hierarchical clustering loss implemented via a soft binary decision tree, enabling simultaneous learning of known and novel categories. Evaluations on Stimulated Raman Histology (OpenSRH) and H&E-stained whole-slide images (DBTA) show superior patch-level and slide-level performance, with robust discovery of unseen tumor categories and cross-modality generalization. The approach promises faster, more interpretable intraoperative guidance and a pathway toward continual, hierarchy-aware discovery in biomedical imaging.

Abstract

Accurate brain tumor classification is critical for intra-operative decision making in neuro-oncological surgery. However, existing approaches are restricted to a fixed set of predefined classes and are therefore unable to capture patterns of tumor types not available during training. Unsupervised learning can extract general-purpose features, but it lacks the ability to incorporate prior knowledge from labelled data, and semi-supervised methods often assume that all potential classes are represented in the labelled data. Generalized Category Discovery (GCD) aims to bridge this gap by categorizing both known and unknown classes within unlabelled data. To reflect the hierarchical structure of brain tumor taxonomies, in this work, we introduce Hierarchical Generalized Category Discovery for Brain Tumor Classification (HGCD-BT), a novel approach that integrates hierarchical clustering with contrastive learning. Our method extends contrastive learning based GCD by incorporating a novel semi-supervised hierarchical clustering loss. We evaluate HGCD-BT on OpenSRH, a dataset of stimulated Raman histology brain tumor images, achieving a +28% improvement in accuracy over state-of-the-art GCD methods for patch-level classification, particularly in identifying previously unseen tumor categories. Furthermore, we demonstrate the generalizability of HGCD-BT on slide-level classification of hematoxylin and eosin stained whole-slide images from the Digital Brain Tumor Atlas, confirming its utility across imaging modalities.

Hierarchical Generalized Category Discovery for Brain Tumor Classification in Digital Pathology

TL;DR

This work tackles the problem of brain tumor classification when not all subtypes are labeled by leveraging Generalized Category Discovery with a respect for hierarchical tumor taxonomy. HGCD-BT combines a contrastive learning branch with a semi-supervised hierarchical clustering loss implemented via a soft binary decision tree, enabling simultaneous learning of known and novel categories. Evaluations on Stimulated Raman Histology (OpenSRH) and H&E-stained whole-slide images (DBTA) show superior patch-level and slide-level performance, with robust discovery of unseen tumor categories and cross-modality generalization. The approach promises faster, more interpretable intraoperative guidance and a pathway toward continual, hierarchy-aware discovery in biomedical imaging.

Abstract

Accurate brain tumor classification is critical for intra-operative decision making in neuro-oncological surgery. However, existing approaches are restricted to a fixed set of predefined classes and are therefore unable to capture patterns of tumor types not available during training. Unsupervised learning can extract general-purpose features, but it lacks the ability to incorporate prior knowledge from labelled data, and semi-supervised methods often assume that all potential classes are represented in the labelled data. Generalized Category Discovery (GCD) aims to bridge this gap by categorizing both known and unknown classes within unlabelled data. To reflect the hierarchical structure of brain tumor taxonomies, in this work, we introduce Hierarchical Generalized Category Discovery for Brain Tumor Classification (HGCD-BT), a novel approach that integrates hierarchical clustering with contrastive learning. Our method extends contrastive learning based GCD by incorporating a novel semi-supervised hierarchical clustering loss. We evaluate HGCD-BT on OpenSRH, a dataset of stimulated Raman histology brain tumor images, achieving a +28% improvement in accuracy over state-of-the-art GCD methods for patch-level classification, particularly in identifying previously unseen tumor categories. Furthermore, we demonstrate the generalizability of HGCD-BT on slide-level classification of hematoxylin and eosin stained whole-slide images from the Digital Brain Tumor Atlas, confirming its utility across imaging modalities.

Paper Structure

This paper contains 23 sections, 6 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Contribution
  3. Related work
  4. GCD
  5. Artificial Intelligence for Stimulated Raman Histology
  6. Artificial Intelligence for FFPE brain tumor classification
  7. Method
  8. Contrastive learning loss
  9. Semi-supervised hierarchical clustering loss
  10. Experiments and Results
  11. Data
  12. Implementation details
  13. SRH setting
  14. DBTA setting
  15. Baselines
  16. ...and 8 more sections

Figures (4)

  • Figure 1: Overview of our proposed method HCGD-BT consisting of a hierarchical clustering branch and a contrastive learning branch.
  • Figure 2: Results for GCD, DCCL and HGCD-BT for patch-level classification on OpenSRH.
  • Figure 3: Results for GCD, DCCL and HGCD-BT for slide-level classification on DBTA.
  • Figure 4: Different settings for the tree level $T$ for patch-level classification on OpenSRH