Conformalized Survival Analysis from Whole Slide Histopathological Images
Principal Investigator
Name
Qian Zhao
Degrees
Ph.D
Institution
Department of Statistics, Xi'an Jiaotong University
Position Title
Associate Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1232
Initial CDAS Request Approval
Apr 12, 2024
Title
Conformalized Survival Analysis from Whole Slide Histopathological Images
Summary
In recent years, survival analysis from whole slide histopathological images (WSIs) using deep learning, especially for lung cancer, has attracted increasing attention. Currently, most existing deep learning methods can only provide point predictions for the survival time of patients. However, in real clinical applications, the prediction uncertainty is also of great interest. To this aim, in this project, we will try to construct new deep learning methods that can also provide prediction uncertainty. Our main idea is to introduce the recently developed conformal prediction techniques, that can produce prediction intervals in a distribution-free way, and thus naturally provide uncertainty measure with a predefined confidence level. We will also try to develop new deep-learning models to better meet our goal of conformalized survival analysis.
Aims
1. Develop new methods for WSI-based survival analysis that can provide prediction uncertainty.
2. Develop new deep learning models for a better generalization performance for the studied problem.
3. Verify the effectiveness of the proposed methods on WSI datasets, including the NLST dataset.
4. Compare the proposed methods with existing methods.
Collaborators
Xiangyong Cao, Xi'an Jiaotong University
Yuntao Shou, Xi'an Jiaotong University
Peiqiang Yan, Xi'an Jiaotong University
Xingjian Yuan, Xi'an Jiaotong University