Deep Learning and Radiomics Based Outcome Prediction for Cancer Patients

The accurate prediction of cancer patient treatment outcomes is essential for personalized treatment planning and improved treatment outcome. The use of machine learning methods, such as deep learning (DL) and radiomics, has been gaining attention in the field of cancer research for predicting treatment outcomes. In this dissertation, we present a comprehensive study on developing deep learning and radiomics-based models for outcome prediction in various types of cancer patients. The first part of this study focuses on developing a deep learning model for joint vestibular schwannoma enlargement prediction and segmentation using initial diagnosis MR images to assist in patient management when a tumor is first discovered. The second part of the study aims to identify high-risk head and neck cancer (HNC) patients for locoregional recurrence (LRR) before radiotherapy using clinical data and PET/CT imaging. In addition, we developed a DL segmentation model to guide the extraction of radiomics features for HNC recurrence-free survival (RFS) prediction using data collected pre-treatment. These approaches allow for efficient and accurate prediction of LRR and RFS, which is essential for patient counseling and have the potential to enable clinicians to tailor treatment plans accordingly. The third part of the study focuses on pancreatic ductal adenocarcinoma (PDAC), a highly aggressive form of cancer. We developed delta-radiomics (DRF) based models for overall survival (OS), disease-free survival (DFS), and surgical margin prediction for PDAC patients using clinical and imaging data collected after neoadjuvant therapy. These models can assist clinicians in intra-treatment decision-making about pancreas tumor surgery. Finally, we developed an HNC locoregional recurrence prediction method using early surveillance images and investigated a prediction uncertainty estimation method to quantify the reliability of predictions with the AI model. This method helps to identify HNC patients at risk of recurrence and provides clinicians with useful information for planning follow-up care. Overall, the deep learning and radiomics-based models developed in this dissertation offer promising tools for predicting outcomes in various types of cancer patients. These models have the potential to provide clinicians with important information for treatment planning and can aid in improving patient treatment outcomes.

Author dc:creator

• Wang, Kai

Contributors dc:contributor

• Lu, Weiguo
• Wang, Jing
• Jia, Xun
• Gu, Xuejun
• Nguyen, Dan

• Deep Learning
• Head and Neck Neoplasms
• Magnetic Resonance Imaging
• Neuroma, Acoustic
• Outcome Assessment, Health Care
• Radiomics

Language dc:language

en

Identifier

1522122389

OAI identifier oai:identifier

oai:utswmed-ir.tdl.org:2152.5/10588