A 10-Year Probability Deep Neural Network Prediction Model for Lung Cancer

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Lung cancer is a leading cause of death in Taiwan, with high incidence rates, particularly among men.
• Early detection is crucial for improving patient outcomes and survival rates.
• Existing diagnostic methods can be enhanced with predictive modeling.

Purpose of the Study:

• To develop a 10-year probability deep neural network (DNN) prediction model for lung cancer.
• To identify key diseases and symptoms associated with lung cancer risk.
• To support clinical decision-making and facilitate earlier lung cancer diagnosis.

Main Methods:

• A cohort study utilizing the National Health Insurance Research Database (NHIRDB).
• Analysis of patient disease histories up to ten years prior to lung cancer diagnosis.
• Development of a nine-layer DNN model trained with 1000 iterations, batch size 100, SGD optimizer, learning rate 0.1, and momentum 0.1.

Main Results:

• Identified 13 diseases as significant predicting factors for lung cancer.
• The DNN model achieved 85.4% accuracy, 72.4% sensitivity, and 85% specificity.
• The model demonstrated an Area Under the ROC Curve (AUROC) of 87.4%, indicating strong predictive performance.

Conclusions:

• The developed DNN model offers a valuable tool for predicting lung cancer risk up to a decade in advance.
• The model identified novel predictive features beyond established clinical knowledge.
• Early prediction and diagnosis of lung cancer can be significantly improved, benefiting at-risk patients.