Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tissue Aware Nuclei Detection and Classification Model for Histopathology Images

Kesi Xu, Eleni Chiou, Ali Varamesh, Laura Acqualagna, Nasir Rajpoot

TL;DR

TAND tackles nuclei detection and classification in histopathology by leveraging tissue context through a tissue-aware conditioning mechanism. A frozen Virchow-2 tissue segmentation branch provides tissue priors, while a DINOv3-based detection–classification backbone yields nucleus centers and class logits; multi-scale Spatial-FiLM modulates the classification stream using tissue cues, leaving detection untouched. The approach demonstrates strong performance on the PUMA dataset under point supervision, with notable gains in tissue-dependent cell types and a reduced annotation burden compared to dense mask supervision. This work provides a practical pathway to incorporate tissue context into nucleus analysis, improving reliability for clinical workflows without requiring full instance annotations.

Abstract

Accurate nuclei detection and classification are fundamental to computational pathology, yet existing approaches are hindered by reliance on detailed expert annotations and insufficient use of tissue context. We present Tissue-Aware Nuclei Detection (TAND), a novel framework achieving joint nuclei detection and classification using point-level supervision enhanced by tissue mask conditioning. TAND couples a ConvNeXt-based encoder-decoder with a frozen Virchow-2 tissue segmentation branch, where semantic tissue probabilities selectively modulate the classification stream through a novel multi-scale Spatial Feature-wise Linear Modulation (Spatial-FiLM). On the PUMA benchmark, TAND achieves state-of-the-art performance, surpassing both tissue-agnostic baselines and mask-supervised methods. Notably, our approach demonstrates remarkable improvements in tissue-dependent cell types such as epithelium, endothelium, and stroma. To the best of our knowledge, this is the first method to condition per-cell classification on learned tissue masks, offering a practical pathway to reduce annotation burden.

Tissue Aware Nuclei Detection and Classification Model for Histopathology Images

TL;DR

TAND tackles nuclei detection and classification in histopathology by leveraging tissue context through a tissue-aware conditioning mechanism. A frozen Virchow-2 tissue segmentation branch provides tissue priors, while a DINOv3-based detection–classification backbone yields nucleus centers and class logits; multi-scale Spatial-FiLM modulates the classification stream using tissue cues, leaving detection untouched. The approach demonstrates strong performance on the PUMA dataset under point supervision, with notable gains in tissue-dependent cell types and a reduced annotation burden compared to dense mask supervision. This work provides a practical pathway to incorporate tissue context into nucleus analysis, improving reliability for clinical workflows without requiring full instance annotations.

Abstract

Accurate nuclei detection and classification are fundamental to computational pathology, yet existing approaches are hindered by reliance on detailed expert annotations and insufficient use of tissue context. We present Tissue-Aware Nuclei Detection (TAND), a novel framework achieving joint nuclei detection and classification using point-level supervision enhanced by tissue mask conditioning. TAND couples a ConvNeXt-based encoder-decoder with a frozen Virchow-2 tissue segmentation branch, where semantic tissue probabilities selectively modulate the classification stream through a novel multi-scale Spatial Feature-wise Linear Modulation (Spatial-FiLM). On the PUMA benchmark, TAND achieves state-of-the-art performance, surpassing both tissue-agnostic baselines and mask-supervised methods. Notably, our approach demonstrates remarkable improvements in tissue-dependent cell types such as epithelium, endothelium, and stroma. To the best of our knowledge, this is the first method to condition per-cell classification on learned tissue masks, offering a practical pathway to reduce annotation burden.

Paper Structure

This paper contains 16 sections, 5 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. The Proposed Method
  3. Overview
  4. Tissue Segmentation Branch
  5. DINOv3 Detection-Classification Backbone
  6. Spatial Feature-wise Linear Modulation
  7. Training Protocol
  8. Experiments
  9. Dataset and Protocol
  10. Comparative Evaluation
  11. Ablation Study
  12. Visualisation Result
  13. Discussion and Limitations
  14. Conclusions
  15. Compliance with ethical standards
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Overall architecture of the proposed Tissue Aware Nuclei Detection and Classification (TAND) framework. A frozen Virchow-2 encoder predicts tissue masks probability map. The red arrow shows feature maps modulate only the nuclei classification stream via multi-scale Spatial-FiLM at $1/16$, $1/8$, and $1/4$ scales, while the nuclei detection path remains unmodulated.
  • Figure 2: Per-class F1 comparison between the state-of-the-art methods and the proposed TAND model.
  • Figure 3: Qualitative visualization of TAND's predictions. (Center) TAND's nuclei overlay on a test sample. (Right Column) Zoomed-in (dashed box) comparing the DINOv3 U-Net baseline (top) and Ground Truth (bottom). (Left Column) Corresponding ground truth (top) and TAND's predicted (bottom) tissue segmentation.