2024 World Conference on Lung Cancer (WCLC) - ePosters-EP.08F.04 Development of Learning-Based Predictive Models for Radiation-Induced Atrial Fibrillation in Non-Small Cell Lung Cancer

EP.08F.04 Development of Learning-Based Predictive Models for Radiation-Induced Atrial Fibrillation in Non-Small Cell Lung Cancer

Back to course

Pdf Summary

The study focuses on developing learning-based predictive models to predict radiation-induced atrial fibrillation (AF) in patients with non-small cell lung cancer (NSCLC) undergoing chemoradiotherapy (CRT). The research was carried out by a team from Yonsei University and Samsung Medical Center and aimed to integrate various types of input features to investigate their importance and validate the models with external data.<br /><br />The study utilized a data cohort consisting of 321 samples from Yonsei University and 187 samples from Samsung Medical Center. These were used to train and validate the models internally and externally. The features used in the models included clinical, dosimetric, and coronary artery calcium (CAC) scores, with the study emphasizing the importance of maximum heart dose and sinoatrial node (SAN) dose as significant predictors of AF.<br /><br />The machine learning (ML) models employed techniques like LASSO for feature selection and PCA for dimensionality reduction, with logistic regression used for classification. The deep learning (DL) model utilized a hybrid structure combining multi-layered perceptrons (MLPs) and long short-term memory (LSTM) networks.<br /><br />Validation results showed that DL models outperformed ML models in predictive capabilities, achieving a more consistent and robust performance. Internal validation showed an AUC (area under the curve) of over 0.8 for DL models, demonstrating superior performance compared to ML models both with and without intervention. External validation confirmed the utility of incorporating feature interventions in improving the model's predictive accuracy.<br /><br />The study concludes that deep learning predictive models significantly improve prediction accuracies for radiation-induced AF in NSCLC, with certain dosimetric features regarded as critical contributors to AF incidence. These findings underscore how advanced machine and deep learning methods can enhance clinical decision-making processes in radiation oncology.

Asset Subtitle

Hong In Yoon

Meta Tag

Speaker Hong In Yoon

Topic Local-Regional Non-small Cell Lung Cancer

Keywords

learning-based predictive models

radiation-induced atrial fibrillation

non-small cell lung cancer

chemoradiotherapy

clinical and dosimetric features

machine learning

deep learning

LASSO and PCA

multi-layered perceptrons

sinoatrial node dose