Models

This section documents the modeling layer of the project.

Main model families

• TransnnMIL v2.0 — custom multi-branch WSI architecture with attention, hierarchical pooling, and topology-aware graph processing.
• AttentionMIL / CLAM / TransMIL-style baselines — multiple-instance learning baselines for WSI and patch-based experiments.
• Foundation model encoders — pretrained feature extractors used for pathology representation learning.

Validation status

Model documentation describes research implementations. Reported benchmark numbers should be interpreted with their dataset, split, hardware, and validation context.