Predicting patient health trajectories from big health record data using advanced AI methods (postdoctoral)

Professor Yue Li is hiring one Postdoc for the machine learning and AI research in healthcare. The position will be held at the School of Computer Science.

The broad adoption of electronic health records (EHR) systems has created rich digital healthcare data and opportunities for conducting transformative health informatics research. However, irregular patient visits, heterogeneous data modalities, restricted EHR data access, and data distribution shifts among hospitals hinder their full utility. Rapidly advancing machine learning (ML) research offer great promise including transformer-based deep generative models and federated learning framework.

In Li lab, we develop explainable AI (xAI) methods that adapt the state-of-the-art AI methods for real-world life science applications pertinent to human health. In a funded project, we proposed 3 aims:

Aim 1: Predicting patient health trajectories using AI. Through our collaboration with the co-PIs at the McGill University Health Centre (MUHC), we have access to irregular longitudinal patient records for over one million patients from Montreal metropolitan area. We will develop a foundational “HealthGPT” model with a focus on predicting future health trajectories of patients by learning the hidden clinical semantics from pretraining on the massive EHR data. We will adapt the recently developed time-series transformer architectures to effectively account for the unequally spaced time-series time in our EHR data.

Aim 2: Robust federated learning in healthcare sectors. Training large neural networks to learn clinical semantics requires big data. Through our extended network, we can gain secure access to patient data on-site beyond MUHC. By using federated learning (FL) framework, we can aggregate model parameters in a central server without sharing the patient data from each hospital. To build a robust FL model, we will tackle data distribution shifts or out-of-distribution (OOD) challenges among hospitals. We will use deep density estimators to model the distributions within health centres and train OOD-aware AI models that are tailored to each hospital in their prediction tasks while harnessing the information across hospitals.

Aim 3: building real-world deployable software healthcare system for federated learning. While many FL algorithms in healthcare were developed, very few have been deployed in the real-world healthcare systems. This aim addresses the software engineering problem in building secure, robust, and efficient network communications between hospitals to deploy state-of-the-art AI methods in the health systems using/adapting existing FL frameworks such as Flower (https://flower.dev/).

Name of the immediate Supervisor:           Yue Li

Work schedule:         Five working days (Monday through Friday)

Working hours:         40 hours per week, 9 a.m. to 5 p.m.

Work location:           Trottier building at McGill Downtown campus

Salary information:   50,000 / year + benefits

 Planned start and end dates of appointment (if applicable):

The position can start on September 1, 2023, and ends on September 1, 2024. The position can be extended by 1 year.

To apply:  Send your CV and cover letter to : yueli@cs.mcgill.ca

The postdoc candidate will have the following duties under supervision of Dr. Li:

1. developing and implementing ML models pertinent to the described aims
2. processing in-house healthcare data
3. manuscript preparations
4. grant writing
5. mentoring graduate students in Li lab
6. attend the Li lab weekly lab meeting and present on a regular cycle.

The postdoc research associate must have a PhD degree in computer science or equivalence. He or she should have strong background in machine learning especially deep learning and hands-on experience in programming. Backgrounds in health informatics are also desired.