Advance Machine Learning for Medical Survival Analysis

In this Mitacs Cluster project, we aim to investigate and improve performance of deep survival models developed by Ortho Biomed Inc., our industrial collaborator. The main objective is to predict clinical outcomes, such as time of death, and the time of organ failure before and after surgery, more accurately and more efficiently. Architecturally, Ortho Biomed Inc. has developed two Deep Neural Network (DNN)-based model types, One2Seq and Seq2Seq, where the former (the One2Seq model) can accept one observation per patient. The Seq2Seq model, on the other hand, can accept longitudinal observations, i.e., multiple observations over time per patient. The outputs of both model types are interpretable and can be a probability distribution function (PDF), regression value, or binary occurrence of an event. Because of superior performances and interpretability of their models over other state-of-the-art solutions, Health Canada selected Ortho Biomed Inc. to build a prototype product. However, the company targets further improving performance of their underlying models using recent advancements in DNN architectures.

Faculty Supervisor:

Arash Mohammadi

Student:

Partner:

Ortho Biomed

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

Concordia University; University of Calgary

Program: