Deep3DSCan
ML — deep residual networks + morphological descriptors for lung cancer classification and 3D segmentation.
Research · Medical imaging · ML
Classify and segment lung cancer in 3D
Deep3DSCan combines deep residual networks with morphological descriptors for lung cancer classification and 3D segmentation — bridging CNN feature learning with shape priors used clinically.
System An end-to-end medical imaging pipeline: volumetric CT inputs → residual CNN backbone → morphological descriptor fusion → dual outputs for malignancy classification and 3D nodule segmentation.
Problem Pure deep models can miss clinically meaningful shape cues; handcrafted morphology alone lacks representational power. Lung screening needs both accurate class decisions and spatially precise segmentations.
Our contribution Fused residual deep features with morphological descriptors in a unified framework; framed classification and 3D segmentation as joint goals; demonstrated a practical architecture for volumetric lung-cancer analysis.
Achievements
- Published medical-imaging ML system with dual-task framing
- Showed how morphology + deep residuals improve nodule characterization
- Artifact available via project PDF on this site