Cell-Graph Dataset for Nucleus Classification in Pan-Cancer Histology

Graph-PanNuke is a node-level classification dataset derived from the PanNuke pan-cancer histology dataset, using all slides at 40× magnification. Each tissue patch is converted into a cell-graph where nodes represent detected cell nuclei, with the task of predicting cell type across 5 classes. Node features describe cell morphology and texture.

Use Cases

• Train a graph neural network on node features describing cell morphology and texture to perform 5-class nucleus classification.
• Analyze spatial proximity relationships encoded in cell-graph edges for tissue structure modeling in pan-cancer histology.
• Benchmark node classification models on cell-graphs derived from the PanNuke dataset at 40× magnification.

Strengths

• Derived from the established PanNuke pan-cancer histology dataset, providing a foundation in cancer research.
• Focuses on a specific, high-resolution (40× magnification) view of tissue slides for detailed cell analysis.
• Structures data as cell-graphs, a format suitable for modern graph-based machine learning methods.

Limitations

• Key quantitative details like the number of graphs, nodes, or edges are not provided in the input.
• The dataset's specific license, file formats, and exact size are unknown, complicating integration planning.
• Relies on the PanNuke source; any inherent limitations in that dataset (e.g., class imbalance, staining variation) may propagate.

Provenance

Source

Derived from the PanNuke pan-cancer histology dataset.

Collection Method

Tissue patches from histology slides at 40× magnification were converted into cell-graphs where nodes are detected nuclei and edges encode spatial proximity.

Time Range

null

Freshness

null

Geography

null