Researchers have developed LPNScreen, a lightweight longitudinal deep learning framework that predicts pulmonary nodule malignancy risk using serial computed tomography examinations, achieving area under the curve values of 82.13 percent on the NLSTLPN dataset and 94.11 percent on the DMULPN dataset.
Longitudinal computed tomography provides valuable information for assessing pulmonary nodule malignancy risk and may improve access to lung cancer screening and follow-up care in resource-constrained primary-care settings. However, conventional prediction methods based on single-time-point computed tomography scans cannot fully exploit temporal evolution across serial examinations.
Deploying advanced deep learning tools in primary care presents practical challenges, as limited computational resources frequently hinder the deployment of large-scale predictive models. The researchers designed LPNScreen to address both temporal modeling requirements and computational efficiency constraints.
The LPNScreen architecture incorporates three complementary components.
These include center–context spatial attention, interval-aware temporal modeling incorporating actual follow-up intervals to characterize irregular temporal dependencies, and tri-pair change modeling capturing nodule evolution across longitudinal scans.
In experimental testing, comparative experiments and ablation studies demonstrated the effectiveness of the framework and confirmed its robustness across datasets featuring different clinical follow-up patterns.
Computational complexity analysis confirmed the lightweight architecture of LPNScreen, which requires 0.434 million parameters and 0.577 giga floating-point operations, indicating its potential for practical deployment in resource-constrained primary-care settings.
Subject of Research: Lightweight longitudinal deep learning for pulmonary nodule malignancy risk prediction
Article Title: LPNScreen: a longitudinal pulmonary nodule malignancy risk prediction model based on temporal interval and feature evolution modeling
Article References: Wang, M., Li, X., Feng, M., & Li, Y. (2026). LPNScreen: a longitudinal pulmonary nodule malignancy risk prediction model based on temporal interval and feature evolution modeling. Frontiers in Medicine, 13, Article 2011576. https://doi.org/10.3389/fmed.2026.2011576
Image Credits: AI Generated
DOI: 10.3389/fmed.2026.2011576
Keywords: pulmonary nodules, lung cancer screening, longitudinal CT, LPNScreen, malignancy risk prediction, deep learning, center-context spatial attention, interval-aware temporal modeling, tri-pair change modeling, NLSTLPN dataset, DMULPN dataset, computational complexity, primary care, feature evolution modeling, follow-up intervals, serial CT examinations, ablation studies
News Source: Ophelia Keating. (November 12, 2026). Lightweight LPNScreen Model Evaluates Pulmonary Nodule Malignancy Across Serial CT Scans. Scienmag.



