Loading...
Survival Analysis for Cancer Patients Using Interpretable Deep Learning
Ridzuan, Muhammad Firdaus Ahmad
Ridzuan, Muhammad Firdaus Ahmad
Files
Supervisor
Department
Machine Learning
Embargo End Date
2025-12-01
Type
Dissertation
Date
2024
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
This thesis explores advancements in survival analysis for cancer prognosis, focusing on challenges posed by censored data and unknown event times. It proposes innovative methods to address these issues and improve prognostic accuracy. Key contributions include:
1. SurvCORN: A method using conditional ordinal ranking networks to predict survival outcomes, improving time-to-event prediction accuracy.
2. SurvRNC: A regularizer for deep survival models that enhances performance by ordering latent features based on survival times.
3. HuLP: A framework that allows clinicians to interact with and intervene in the model s predictions, representing a significant advancement in the field of prognostic modeling and enhancing model reliability and interpretability.
4. MAC-MIL: An attention-based, multiple-instance learning approach to automatically select important patches and predict biochemical recurrence from large images."
Citation
M. F. Ridzuan, "Survival Analysis for Cancer Patients Using Interpretable Deep Learning", PhD Dissertation, Machine Learning, MBZUAI, Abu Dhabi, UAE, 2024