search for




 

A multi-modal CNN approach for high-dimensional survival data
Journal of the Korean Data & Information Science Society 2025;36:115-26
Published online January 31, 2025;  https://doi.org/10.7465/jkdi.2025.36.1.115
© 2025 Korean Data and Information Science Society.

Juyoung Kim1 · Vu Tuan Anh2 · Il Do Ha3

123Department of Artificial Intelligence Convergence, Pukyong National University
3Department of Statistics and Data Science, Pukyong National University
Correspondence to: This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2023-00240794).
1 Master student, Department of Artificial Intelligence Convergence, Pukyong National University, Busan 48513, Korea.
2 Master student, Department of Artificial Intelligence Convergence, Pukyong National University, Busan 48513, Korea.
3 Corresponding author: Professor, Department of Statistics and Data Science, Pukyong National University, Busan 48513, Korea. E-mail: idha1353@pknu.ac.kr
Received December 13, 2024; Revised December 31, 2024; Accepted December 31, 2024.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
High-dimensional survival data, which have a large number of input variables (p) compared to the sample size (n), are typically analyzed using penalized survival analysis based on the Cox proportional hazards (PH) models. However, these modelling approaches assume linearity in the hazard function with respect to the input variables, which prevents them from effectively learning non-linear and interaction patterns among the input variables. To address this issue, this paper proposes a multi-modal CNN (MCNN), which combines deep neural networks (DNN) with convolutional neural networks (CNN) allowing for Attention, based on the Cox-PH model. To train the MCNN survival model, a loss function based on Breslow’s penalized log-likelihood is used. The predictive performance of the proposed MCNN model was evaluated using three practical high-dimensional survival datasets. Experimental results show that the MCNN outperforms existing various survival methods in terms of C-index and integrated Brier score.
Keywords : High-dimensional survival data, deep learning, CNN, Cox-PH model, MCNN