Answers are grounded in this paper and cite their sources.
Introduction Longitudinal CT 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, single-time-point CT-based methods cannot fully exploit temporal evolution across serial examinations, while the limited computational resources available in primary care also hinder the deployment of large-scale models. Methods To address these challenges, we propose LPNScreen, a lightweight longitudinal framework for pulmonary nodule malignancy prediction. LPNScreen consists of three complementary components: centercontext spatial attention, which enhances discriminative nodule and surrounding contextual representations; interval-aware temporal modeling, which incorporates actual follow-up intervals to characterize irregular temporal dependencies; and tri-pair change modeling, which explicitly captures both adjacent and long-range nodule evolution across longitudinal scans Results Experimental results showed that LPNScreen achieved AUCs of 82.13% and 94.11% on the NLSTLPN and DMULPN datasets, respectively. Extensive comparative experiments and ablation studies further demonstrated the effectiveness of the proposed framework and its robustness across datasets with different follow-up patterns. In addition, computational complexity analysis confirmed the lightweight nature of LPNScreen, which requires only 0.434M parameters and 0.577G FLOPs.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: