Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Cells to Survival: Hierarchical Analysis of Cell Inter-Relations in Multiplex Microscopy for Lung Cancer Prognosis

Olle Edgren Schüllerqvist, Jens Baumann, Joakim Lindblad, Love Nordling, Artur Mezheyeuski, Patrick Micke, Nataša Sladoje

TL;DR

The paper addresses prognosis in NSCLC by leveraging the tumor microenvironment captured in multiplex imaging. It introduces HiGINE, a hierarchical graph isomorphism network with edge features that encodes local and global cell interactions, incorporating cancer stage as multimodal information. Across two public NSCLC datasets, HiGINE achieves superior risk stratification compared with baselines, with ablations showing benefits from edge weighting, hierarchical core modeling, and stage fusion. The approach emphasizes interpretability and robustness, offering a practical framework for prognosis based on spatial cell inter-relations in the TME.

Abstract

The tumor microenvironment (TME) has emerged as a promising source of prognostic biomarkers. To fully leverage its potential, analysis methods must capture complex interactions between different cell types. We propose HiGINE -- a hierarchical graph-based approach to predict patient survival (short vs. long) from TME characterization in multiplex immunofluorescence (mIF) images and enhance risk stratification in lung cancer. Our model encodes both local and global inter-relations in cell neighborhoods, incorporating information about cell types and morphology. Multimodal fusion, aggregating cancer stage with mIF-derived features, further boosts performance. We validate HiGINE on two public datasets, demonstrating improved risk stratification, robustness, and generalizability.

From Cells to Survival: Hierarchical Analysis of Cell Inter-Relations in Multiplex Microscopy for Lung Cancer Prognosis

TL;DR

The paper addresses prognosis in NSCLC by leveraging the tumor microenvironment captured in multiplex imaging. It introduces HiGINE, a hierarchical graph isomorphism network with edge features that encodes local and global cell interactions, incorporating cancer stage as multimodal information. Across two public NSCLC datasets, HiGINE achieves superior risk stratification compared with baselines, with ablations showing benefits from edge weighting, hierarchical core modeling, and stage fusion. The approach emphasizes interpretability and robustness, offering a practical framework for prognosis based on spatial cell inter-relations in the TME.

Abstract

The tumor microenvironment (TME) has emerged as a promising source of prognostic biomarkers. To fully leverage its potential, analysis methods must capture complex interactions between different cell types. We propose HiGINE -- a hierarchical graph-based approach to predict patient survival (short vs. long) from TME characterization in multiplex immunofluorescence (mIF) images and enhance risk stratification in lung cancer. Our model encodes both local and global inter-relations in cell neighborhoods, incorporating information about cell types and morphology. Multimodal fusion, aggregating cancer stage with mIF-derived features, further boosts performance. We validate HiGINE on two public datasets, demonstrating improved risk stratification, robustness, and generalizability.

Paper Structure

This paper contains 14 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and previous work
  3. Data
  4. Method
  5. Graph Construction
  6. Model Architecture
  7. Training
  8. Multimodal Information Fusion
  9. Evaluation
  10. Metrics
  11. Baselines
  12. Results
  13. Ablation Study
  14. Discussion and Conclusion

Figures (3)

  • Figure 1: Each cell in the mIF image (7 channels, here visualized as RGB) is segmented, and its quantitative features are extracted. The core is divided into subsample graphs using a sliding window approach. Each subsample graph is represented as a single node in the core graph.
  • Figure 2: At the first level, the model classifies local cell graphs as short or long survival. Embeddings from the penultimate layer are used as node features of the graph at the second level, for core-level short- vs. long-term survival predictions.
  • Figure 3: Kaplan-Meier curves for patients from Dataset 1 split into short/long survival, predicted by HiGINE when (left) incorporating CS in the model input, and (right) relying on mIF image data alone. The $p$-values are based on the log-rank test.