nadeemlab/CIR

Provides voxel-to-mesh deep learning models for end-to-end lung nodule segmentation and clinically-grounded malignancy prediction, extending Voxel2Mesh with spike classification (spiculation/lobulation) that preserves sharp surface features. Includes a curated dataset of ~1000 radiologist-annotated nodules from LIDC-IDRI and LUNGx with voxel-level clinical feature labels, enabling direct attribution of predictions to actionable radiological findings rather than relying on post-hoc explanation methods. Built on PyTorch and PyTorch3D with Docker support for reproducible end-to-end workflows.

No commits in the last 6 months.