Using Pretreatment PETCT, DeepMTS Deep Multi-task Learning for Survival Prediction in Patients with Advanced Nasopharyngeal Carcinoma

PROJECT TITLE :

DeepMTS Deep Multi-task Learning for Survival Prediction in Patients with Advanced Nasopharyngeal Carcinoma using Pretreatment PETCT

ABSTRACT:

Nasopharyngeal Carcinoma, also known as NPC, is a form of malignant epithelial cancer that originates in the nasopharynx. Predicting whether or not a patient will be able to live through their disease is a significant source of anxiety for oncology professionals and patients alike. Deep survival models, which are based on Deep Learning, have recently demonstrated the potential to outperform traditional radiomics-based survival prediction models. These models have been developed in recent years. As the input, deep survival models typically make use of image patches that either cover the entirety of the target regions (such as the nasopharynx for NPC) or contain only segmented tumor regions. However, models that use the entire target regions will also include information that is irrelevant to the background, whereas models that use segmented tumor regions will ignore potentially prognostic information that exists outside of primary tumors (e.g., local lymph node metastasis and adjacent tissue invasion). We propose a 3D end-to-end Deep Multi-Task Survival model (DeepMTS) in this study for the purpose of joint survival prediction and tumor segmentation in advanced NPC based on pretreatment PET/CT scans. Our innovation consists of the implementation of a hard-sharing segmentation backbone to direct the extraction of local features related to the primary tumors. This helps to reduce interference from background information that is not pertinent to the problem at hand. In addition, we also introduce a cascaded survival network to capture the prognostic information existing out of primary tumors and to further leverage the global tumor information (for example, tumor size, shape, and locations) derived from the segmentation backbone. Both of these innovations were carried out in this study. Using two different clinical datasets, we were able to demonstrate that our DeepMTS has the ability to consistently outperform both traditional radiomics-based survival prediction models as well as existing deep survival models.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here