AnatomicalNets: Deep Learning Pipeline for Lung Cancer T-Staging
A research paper on arXiv introduces AnatomicalNets, a multi-stage deep learning pipeline for lung cancer T-staging. Unlike conventional black-box classification models, AnatomicalNets reformulates staging as a measurement and rule-based inference problem. It uses three encoder-decoder networks to segment lung parenchyma, tumor, and mediastinum, with diaphragm boundary estimated via a lung-contour heuristic. The approach aims to align with explicit anatomical criteria from clinical guidelines, enhancing interpretability and accuracy.
Key facts
- Title: AnatomicalNets: A Multi-Structure Segmentation and Contour-Based Distance Estimation Pipeline for Clinically Grounded Lung Cancer T-Staging
- Published on arXiv with ID 2511.19367
- Announce type: replace-cross
- Proposes a multi-stage pipeline for tumor staging
- Uses three dedicated encoder-decoder networks for segmentation
- Reformulates staging as measurement and rule-based inference
- Focuses on lung cancer T-staging
- Aims to improve interpretability over deep learning classification
Entities
Institutions
- arXiv