Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Modeling Long Sequences in Bladder Cancer Recurrence: A Comparative Evaluation of LSTM,Transformer,and Mamba

Runquan Zhang, Jiawen Jiang, Xiaoping Shi

TL;DR

This work tackles predicting bladder cancer recurrence from long time-series data by comparing three long-sequence architectures—LSTM, Transformer, and Mamba—and integrating them with the Cox proportional hazards model. The LSTM-Cox approach delivers strong predictive performance and interpretable risk factors, outperforming some newer architectures in several settings, and highlighting predictors such as treatment stop time, maximum tumor size at recurrence, and recurrence frequency. The study uses both real bladder cancer recurrence data and simulated Weibull-based data, employing LIME, gradient analysis, and t-SNE to elucidate feature contributions and risk stratification. The results support the practical integration of LSTM-based temporal feature extraction with survival analysis to improve clinical risk prediction and patient management.

Abstract

Traditional survival analysis methods often struggle with complex time-dependent data,failing to capture and interpret dynamic characteristics adequately.This study aims to evaluate the performance of three long-sequence models,LSTM,Transformer,and Mamba,in analyzing recurrence event data and integrating them with the Cox proportional hazards model.This study integrates the advantages of deep learning models for handling long-sequence data with the Cox proportional hazards model to enhance the performance in analyzing recurrent events with dynamic time information.Additionally,this study compares the ability of different models to extract and utilize features from time-dependent clinical recurrence data.The LSTM-Cox model outperformed both the Transformer-Cox and Mamba-Cox models in prediction accuracy and model fit,achieving a Concordance index of up to 0.90 on the test set.Significant predictors of bladder cancer recurrence,such as treatment stop time,maximum tumor size at recurrence and recurrence frequency,were identified.The LSTM-Cox model aligned well with clinical outcomes,effectively distinguishing between high-risk and low-risk patient groups.This study demonstrates that the LSTM-Cox model is a robust and efficient method for recurrent data analysis and feature extraction,surpassing newer models like Transformer and Mamba.It offers a practical approach for integrating deep learning technologies into clinical risk prediction systems,thereby improving patient management and treatment outcomes.

Modeling Long Sequences in Bladder Cancer Recurrence: A Comparative Evaluation of LSTM,Transformer,and Mamba

TL;DR

This work tackles predicting bladder cancer recurrence from long time-series data by comparing three long-sequence architectures—LSTM, Transformer, and Mamba—and integrating them with the Cox proportional hazards model. The LSTM-Cox approach delivers strong predictive performance and interpretable risk factors, outperforming some newer architectures in several settings, and highlighting predictors such as treatment stop time, maximum tumor size at recurrence, and recurrence frequency. The study uses both real bladder cancer recurrence data and simulated Weibull-based data, employing LIME, gradient analysis, and t-SNE to elucidate feature contributions and risk stratification. The results support the practical integration of LSTM-based temporal feature extraction with survival analysis to improve clinical risk prediction and patient management.

Abstract

Traditional survival analysis methods often struggle with complex time-dependent data,failing to capture and interpret dynamic characteristics adequately.This study aims to evaluate the performance of three long-sequence models,LSTM,Transformer,and Mamba,in analyzing recurrence event data and integrating them with the Cox proportional hazards model.This study integrates the advantages of deep learning models for handling long-sequence data with the Cox proportional hazards model to enhance the performance in analyzing recurrent events with dynamic time information.Additionally,this study compares the ability of different models to extract and utilize features from time-dependent clinical recurrence data.The LSTM-Cox model outperformed both the Transformer-Cox and Mamba-Cox models in prediction accuracy and model fit,achieving a Concordance index of up to 0.90 on the test set.Significant predictors of bladder cancer recurrence,such as treatment stop time,maximum tumor size at recurrence and recurrence frequency,were identified.The LSTM-Cox model aligned well with clinical outcomes,effectively distinguishing between high-risk and low-risk patient groups.This study demonstrates that the LSTM-Cox model is a robust and efficient method for recurrent data analysis and feature extraction,surpassing newer models like Transformer and Mamba.It offers a practical approach for integrating deep learning technologies into clinical risk prediction systems,thereby improving patient management and treatment outcomes.
Paper Structure (18 sections, 14 equations, 4 figures, 5 tables)

This paper contains 18 sections, 14 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methods and Materials
  3. Variable Definitions
  4. Dataset Description
  5. Data Preprocessing
  6. Model Design and Feature Examination
  7. LSTM Model Settings
  8. Transformer Model Settings
  9. Mamba Model Settings
  10. Construction and Training of the Cox Proportional Hazards Model
  11. t-SNE Visualization Analysis
  12. LIME and Gradient Test Analysis
  13. Validation on Simulated Dataset
  14. Comparison of Traditional Survival and Recurrence Models
  15. Software and Tools
  16. ...and 3 more sections

Figures (4)

  • Figure 1: (a) Overall research design (b) Specific structure of the LSTM model
  • Figure 2: Kaplan-Meier curves for high-risk and low-risk groups predicted by the LSTM-Cox,Transformer-Cox,and Mamba-Cox models
  • Figure 3: t-SNE visualizations of patient risk profiles based on features extracted by the LSTM-Cox,Transformer-Cox,and Mamba-Cox models
  • Figure 4: LIME and Gradient analysis displaying the frequency and impact of features on LSTM-Cox model predictions