To introduce HiGATE, a framework that unifies cellular and tissue-level analysis in histopathological diagnosis through a dual-graph architecture, emphasizing the importance of this integration.
Key Findings:
Achieved 91.3% accuracy and F1-score of 0.896 for nuclei classification on the PanNuke benchmark.
Demonstrated 85.4% accuracy for tissue-type classification across 19 cancer types.
Exhibited exceptional cross-dataset generalization with high performance in MoNuSeg, DigestPath, and TCGA-BRCA tasks.
Maintained a precision of 0.87 at a recall of 0.95, reducing false positives by 10.1% compared to HACT-Net.
Received a mean diagnostic relevance score of 4.1/5.0 from a multi-reader study with pathologists.
Interpretation:
Limitations:
The study does not address potential biases in dataset selection or model generalizability beyond the tested datasets.
Conclusion:
HiGATE serves as a foundation for diagnostic AI in personalized medicine, with applications in domains requiring multi-scale relational reasoning, based on the findings presented.
