Authors: Jing Teng, Lan Yang, Shan Wang, Jing Yu
Published: 2025-02-13
Source: Full article
AbstractAnalyzing survival rates for lung cancer presently grapples with two significant hurdles. Insufficient available data is the first one, which is exacerbated by a large amount of censored information, thereby obstructing the effective employment of this data for accurate predictions. Second, lung cancer patient survival data often exhibit complex temporal feature associations, suggesting patient‐specific traits on survival outcomes. To address these issues, the dataset are augmented by integrating semi‐supervised learning, which allows for more effective use of the event data and mitigates overfitting issues. Subsequently, a Transformer‐based predictive model is developed, trained by assorted survival time groups, intending to expand the comprehension of data feature associations. The accuracy of the proposed model is meticulously assessed using a variety of validation metrics, including the Time‐dependent C‐statistic, Integrated Brier Score (IBS), the Time‐dependent receiver operating characteristic (ROC) curves, and survival curves. These rigorous evaluations offer an intricate insight into the model's performance across different time intervals and event occurrences, thus obtaining a holistic comprehension of the model's precision. The experimental findings convincingly demonstrate the superiority of the proposed framework compared to state‐of‐the‐art methods.