A multi-parameterized artificial neural network for lung cancer risk prediction

Area of Science:

• Oncology
• Medical Informatics
• Biostatistics

Background:

• Lung cancer remains a leading cause of cancer mortality worldwide.
• Early detection and risk stratification are crucial for improving patient outcomes.
• Predictive models utilizing readily available health information can enhance screening strategies.

Purpose of the Study:

• To train and validate a multi-parameterized artificial neural network (ANN) for predicting lung cancer risk.
• To assess the sensitivity and specificity of the ANN model using diverse personal health information.
• To evaluate the potential of the ANN as a cost-effective, non-invasive clinical tool for lung cancer risk stratification.

Main Methods:

• Utilized the 1997-2015 National Health Interview Survey adult dataset.
• Included demographic and health-related features: gender, age, BMI, diabetes, smoking status, emphysema, asthma, race, Hispanic ethnicity, hypertension, heart diseases, vigorous exercise habits, and history of stroke.
• Trained and validated a multi-parameterized artificial neural network (ANN) model, identifying 648 cancer and 488,418 non-cancer cases.

Main Results:

• The ANN achieved a sensitivity of 79.8% and specificity of 79.9% on the training set (AUC: 0.86).
• On the validation set, the model demonstrated a sensitivity of 75.3% and specificity of 80.6% (AUC: 0.86).
• The model exhibited high specificity and modest sensitivity for lung cancer detection.

Conclusions:

• An ANN model trained on personal health information can effectively stratify lung cancer risk.
• The developed ANN serves as a promising, non-invasive, and cost-effective tool for clinical risk assessment.
• Further research can refine the model for improved sensitivity in lung cancer screening protocols.