Keywords

Data science; machine learning; metastasis informatic; predictive system; multiple metastasis

Abstract

As lung cancer has a high mortality rate, particularly due to the development of metastases, the goal of this study was to develop a predictive system that is capable of predicting the likelihood of metastasis in bone, brain, and liver using multilabel classification models. This system aims to support lung cancer decision-making by providing personalized risk assessments based on patient-specific clinical data. This study utilized and evaluated multilabel classification (MLC) models for simultaneously predicting multiple metastasis sites (bone, brain, and liver) in lung cancer patients, using data from the Surveillance Epidemiology and End Results (SEER) cancer registry. Various MLC methods including problem transformation, algorithm adaptation, and ensemble methods were used to enhance prediction accuracy. The effectiveness of these models was assessed through a comprehensive analysis of label dependencies, feature importance, and performance metrics. Feature importance analysis revealed critical factors such as AJCC staging and age. Although label powerset random forest had a slightly higher sub-accuracy (74.2%) and lower Hamming loss (0.116), RakEL random forest outperformed label powerset random forest in terms of F-1 score micro-precision and micro-recall. The models demonstrated varied effectiveness, with some excelling in capturing label interdependencies and others optimized for specific clinical conditions in terms of high precision and recall metrics. Overall, our findings show that multilabel classification models not only offer valuable insights into metastasis prediction but also reveal significant label dependencies and highlight influential features in lung cancer patients.

Download

Share

COinS