An exploratory study of multi-channel CNN for early detection of lung cancer from longitudinal healthcare records

Lung cancer remains the leading cause of cancer-related mortality worldwide, with poor survival rates due to late-stage detection. Current low-dose computed tomography screening faces barriers including high costs and false-positive rates reaching 24%, while artificial intelligence offers opportunities to enhance early detection through longitudinal clinical data analysis. This study developed a Multi-Channel Convolutional Neural Network (MCNN) for lung cancer risk prediction using Taiwan’s National Health Insurance Research Database, encompassing 523,539 patients (2,809 lung cancer, 23,783 other cancer, and 496,947 non-cancer). The MCNN was designed as a lightweight model processing nine channels of diagnostic codes, medications, and medical orders over a three-year observation period. Systematic feature selection reduced estimated feature storage requirements by 99.8%, from approximately 1,184 GB for the full ICD feature space to approximately 2.11 GB for the selected features, while retaining clinical relevance. Model performance was assessed using stratified 10-fold cross-validation against seven machine learning baselines, and interpretability was examined through SHAP analysis. The MCNN achieved an F₁-score of 66.91%, precision of 84.47%, and recall of 59.79%. Ablation studies confirmed multi-modal integration benefits, with diagnostic codes providing primary predictive power. SHAP analysis revealed distinct temporal patterns validating the model’s ability to identify pre-diagnostic phases through healthcare engagement patterns. Findings are based on internal validation within a single national database, and key risk factors such as smoking history are not captured in administrative claims data; future evaluation in independent external cohorts is therefore warranted to confirm these findings. The model’s high precision minimizes false-positive rates while its computational efficiency and clinical interpretability support practical implementation as a complementary claims-based screening support tool for early cancer detection.

This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.