Time-to-event modeling in lung cancer prognosis: From classical survival Analysis to machine and deep learning approaches

Huihu College of Pharmacy, Xi’an Jiaotong-Liverpool University, Suzhou 215123, Jiangsu Province, China

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Lung cancer is still the top cause of death from cancer worldwide. Accurate time-to-event models are important for prognosis and treatment decisions. Older survival analysis methods depend on strict assumptions perform poorly on high-dimensional multi-modal data. Previous studies don’t have a systematic comparison of three different types of approaches: classical methods, machine learning, and deep learning. This leaves a gap between prediction performance and clinical interpretability. This review gives a systematic summary of different time-to-event modeling techniques that are used for lung cancer prognosis. It looks at how methods work, what good things techniques have, and how well models work in real clinical settings, split into three main groups: classical statistics, machine learning, and deep learning. Classical approaches like Kaplan-Meier and Cox regression provide the basic groundwork for analysis, but have inherent limitations. Machine learning methods deal well with nonlinear relationships and complex data points. Deep learning lets people get features all the way from start to finish, right from medical images and multi-omics data. To pick the right method, researchers have to match it up with data type and research goals. This review provides practical guidance on method selection and supports personalized prognostic decision-making for lung cancer.