Multi-Scale Lung Histopathology Classification

A pathology model that fuses 20x and 40x whole-slide image views with cross-scale attention for malignancy, subtype, and differentiation grading.

prototype Pathology Histopathology Attention Classification
Project Website Source Code

project status: prototype

Lobe Ranger is a multi-scale ordinal pathology foundation network for whole-slide lung histopathology.

The model pairs 20x and 40x image views so the network can preserve both architectural context and cytologic detail. A shared ViT backbone feeds bidirectional cross-scale attention and multi-task heads for malignancy, subtype, and ordinal differentiation classification.

Core pieces:

  • Patient-wise 20x and 40x magnification pairing
  • Shared ViT feature extraction
  • Bidirectional cross-scale attention fusion
  • Multi-task classification for malignancy, subtype, and differentiation grade
  • Attention audit utilities for explainability checks