SURVIVAL ANALYSIS
UNLV Project
Jieun Kang, Yejin Kim, SangHeon Lee, Juwon Lee
Content
Introduction of the survival analysis
Brief explanation of models
Project Detail
Result
Introduction
of the survival analysis
INTRODUCTION OF THE SURVIVAL ANALYSIS
What is survival analysis?
Survival analysis is a branch of statistics for analyzing the expected duration of
time until one event occurs, such as death in biological organisms and failure
in mechanical systems.
INTRODUCTION OF THE SURVIVAL ANALYSIS
What is survival analysis?
Brief explanation
of the models
BRIEF EXPLANATION OF THE MODELS
CNTSurv
Censored Neural Network
with Threshold Gradient Descent Regularization
BRIEF EXPLANATION OF THE MODELS
SurvTrace
Transformer-based deep learning model for survival analysis
BRIEF EXPLANATION OF THE MODELS
Cox-ResNet
This model uses gene data that transform into 2D images
2D images => ResNet18 => Cox Regression Layer
BRIEF EXPLANATION OF THE MODELS
CoxCNN
Project Detail
PROJECT DETAIL
Overview of Pipeline
PROJECT DETAIL
Data Exploration
Lung Squamous
Brain Lower
Breast Invasive
Stomach
Cell Carcinoma
Grade Glioma
Carcinoma
Adenocarcinoma
PROJECT DETAIL
Data Exploration
All of Datasets
have
Status(death or not)
and sensored months
LSCC : 495 Samples, 4786 Feature genes
BLGG : 515 Samples, 4788 Feature genes
BIC : 1092 Samples, 4789 Feature genes
SA : 407 Samples, 4789 Feature genes
PROJECT DETAIL
Data Preprocessing
Normalization
: Z-Score Normalization
Feature Selection
: KEGG Pathway
PROJECT DETAIL
Data Split
5 Stratified random sampling & Splitting 8:1:1
Validation set
Train set
Test set
PROJECT DETAIL
Model-specific hyperparameters
CNT
SurvTrace
Hyperparameters
Values
Hyperparameters
Values
Learning Rate***
Hidden Size
Tau*
Optimizer
Epochs**
[1e-4, 1e-5, 1e-6]
[10, 30, 50]
[0.6, 0.7, 0.8]
Adam
30000
Learning Rate***
Hidden Size
Transformer Layer
Multi Head
Optimizer
Epochs**
[1e-4, 1e-5, 1e-6]
[8, 16]
[2, 3, 4]
[1, 2, 4]
Adam
30000
* Threshold for The Threshold Gradient Descent Regularization Method
** Use early-stopping-method
*** Use the learning rate scheduler
PROJECT DETAIL
Model-specific hyperparameters
COX-Resnet
CoxCNN
Hyperparameters
Values
Hyperparameters
Values
Learning Rate***
Hidden Size
Dropout
Optimizer
Epochs**
[1e-4, 1e-5, 1e-6]
[10, 30, 50]
[0.2, 0.4, 0.6]
Adam
30000
Learning Rate***
Optimizer
Epochs**
[1e-4, 1e-5, 1e-6]
Adam
30000
PROJECT DETAIL
Evaluation
Loss Function
: Negative Partial Log-Likelihood
Evaluation
: C-Index
Results
RESULT
Learning curves of Training and Validation data
CNT
SurvTrace
RESULT
Learning curves of Training and Validation data
COX-Resnet
CoxCNN
RESULT
The C-index of test data
C-index for each datasets and trial with models
RESULT
The C-index of test data
Average C-index of models for each datasets
RESULT
The C-index of test data
Average C-index of models for all datasets
References
CNTSurv
Fan, Y.; Zhang, S.; Ma, S. Survival Analysis with High-Dimensional Omics Data Using a Threshold Gradient Descent RegularizationBased Neural Network Approach. Genes 2022, 13, 1674. https://doi.org/10.3390/ genes13091674
SurvTRACE
Zeng, Y., Zhang, X., & Xu, R. (2021). Adversarial Attack and Defense on Graph Data: A Survey. arXiv preprint arXiv:2110.00855.
Cox-ResNet
Diego Vallarino, "Machine Learning Algorithms for Survival Analysis: Advantages, Disadvantages, and Examples", International
Journal of Artificial Intelligence and Machine Learning, vol.4, no.1, pp.10, 2024.
CoxCNN
Yin, Q., Chen, W., Zhang, C. et al. A convolutional neural network model for survival prediction based on prognosis-related cascaded Wx
feature selection. Lab Invest 102, 1064–1074 (2022).
Thank you
감사합니다.