Deep Predictive Clustering of Irregular Time Series Data for Prostate Cancer Risk and Its Explainability
Principal Investigator
Name
Vanessa Tschichold
Degrees
M.Sc.
Institution
Norwegian University of Science and Technology
Position Title
Master Thesis Student
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-996
Initial CDAS Request Approval
Jun 24, 2022
Title
Deep Predictive Clustering of Irregular Time Series Data for Prostate Cancer Risk and Its Explainability
Summary
To detect and classify the risk of prostate cancer, screening of prostate-specific antigen (PSA) levels is done by doctors. In this thesis, we want to improve this risk classification by applying unsupervised machine learning methods and explainability methods. The goal of this thesis is to improve the risk classification of patients for developing prostate cancer by finding a suitable and well performing deep clustering method for clustering irregular PSA time-series data into different risk classes. This also includes representation learning and finding significant features. By comparing the results to the risk classes given by the doctors and adding an explainability layer after the clustering algorithms, we want to find if there is another yet hidden factor in the data which is significant for developing (mortal) prostate cancer.
Aims
* representation learning of irregular time series PSA data
* explainability of learned embeddings
* clustering of patients into risk of prostate cancer
Collaborators
Prof. Massimiliano Ruocco, NTNU
Dr. Alexander Marx, ETH Zurich