HiGATE: hierarchical graph attention for multi-scale tissue encoder in computational pathology

1

HiGATE is a novel framework that integrates cellular and tissue-level analysis using a dual-graph architecture for improved histopathological diagnosis.
2

The framework employs a bidirectional Cross-Level Attention mechanism, allowing dynamic communication between cellular details and tissue organization.
3

HiGATE achieved state-of-the-art performance on the PanNuke benchmark with 91.3% accuracy in nuclei classification and 85.4% in tissue-type classification.
4

The model demonstrated strong cross-dataset generalization, achieving high accuracy in segmentation and classification tasks across multiple datasets.
5

A multi-reader study confirmed HiGATE's clinical relevance, with pathologists rating its integrated multi-scale explanations highly for diagnostic relevance.

1