Abstract
This work investigates the generalizability of state-of-the-art deep learning models within a medically informed ordinal framework for lung nodule malignancy risk assessment. A three-class ordinal learning approach is used to evaluate the algorithms, reflecting the indeterminate nature of screening pulmonary nodules and mirroring radiological decision-making. Convolutional neural networks, vision transformers, and capsule networks were implemented in both 2D and 3D. These models were trained on the well-known LIDC-IDRI dataset and externally validated under identical conditions on the P-ELCAP dataset, a novel well-curated lung cancer screening cohort aligned with standard clinical guidelines for lung nodule management. The 2D vision transformer and 3D capsule network achieved the best and second-best external validation performance on P-ELCAP for distinguishing benign from malignant nodules, with the 3D capsule network producing the fewest indeterminate predictions, only one of which corresponded to a truly malignant. However, these results did not extend to nodules subsequently diagnosed as lung cancer at 1–3 years follow-up. The ordinal ⋆ Both authors share Senior Authorship. 2 Cobo Cano et al. framework can enhance indeterminate pulmonary nodule characterization in low-dose CT imaging, contributing to improved risk stratification. The findings reflect the challenges of generalization and the critical need for external assessment of deep learning models prior to clinical deployment. The dataset and the code are available at https://doi.org/ 10.5281/zenodo.21894313 and https://github.com/MiriamCobo/Deep_ ordinal_lung_nodule_evaluation_framework, respectively.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TIA_010.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{CobMir_Ordinal_MICCAISAT2026,
author = { Cobo Cano, Miriam AND Serrano, Diego AND Barranco, Jennifer AND Pasquier, Andrea AND de-Torres, Juan Pablo AND Ezponda, Ana AND Seijo, Luis M. AND Bastarrika, Gorka AND Silva, Wilson AND Lloret Iglesias, Lara AND Montuenga, Luis M.},
title = { { Ordinal deep learning for pulmonary nodule risk assessment: an external validation study on a novel screening cohort } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17269},
month = {pending},
page = {pending}
}