Distributed System Architecture for Decentralized Learning in Telemedicine
Internship under the supervision of
Sophia Ben Mokhtar, LIRIS- CNRS
Carole Frindel, CREATIS - INSA Lyon
Topics: Decentralized learning / Telemedicine
Machine learning approaches are gaining momentum in the medical domain, especially to inform and drive clinical processes through decision support systems. With the development of large scale telemedical systems providing care to thousands of patients at home, machine learning technology has the potential to improve the efficiency of daily health data assessments in the telemedical center.
As part of our research projects, we investigate the applicability of machine learning for telemedical care in atrial fibrillation (AF), which is among the main causes for unplanned hospitalizations in Europe and drastically increase the risk of stroke. Our database includes 84 long-term ECG recordings of subjects with atrial fibrillation: record durations vary but are typically 24 to 25 hours and were manually annotated by clinicians to label the changes in cardiac rhythm.
This type of applications is particularly appealing in the context of smart cars (this use case is supported by the company Renault Group), where detecting irregularities in hear signals may be detrimental for drivers’ safety as well as their environment (other cars/pedestrians/bicycles in the vicinity).
Since the recorded patient data is subject to high data privacy concerns, the machine learning system must be designed with data minimization and patient anonymity in mind. On the other hand, the trained ML model should ideally be public as to benefit all affected AF patients regardless of whether or not they are currently enrolled in a telemedical care program.
Recent developments in the field of federated learning indicate the feasibility of implementing machine learning systems in which user data remains private. Moreover, in decentralized learning, no central aggregation server is needed. However, regarding the real-world application of such architectures in the medical domain many questions remain, especially regarding the dependability requirements for medical care systems.
Objective: This master project focuses on architecture design and evaluation of decentralized learning from a distributed systems perspective.
High-level project goals include:
- Designing a distributed architecture to ensure the fault tolerance and dependability of the system, including the collaboratively trained model
- Creating a prototypical implementation for use in a lab environment, incorporating peer-to-peer networking and possibly distributed ledger technologies
Prerequisites and expected skills:
- Efficient programming level in Python
- Knowledge of machine learning and TensorFlow and/or Pytorch libraries
- Interest in DL mathematical theory would be a plus
- Good collaboration and communication skills (written / oral)
Hosting structure: This internship will be hosted in Lyon on “La Doua” Campus.
About CREATIS: CREATIS is a joint research unit (CNRS, INSERM) in medical imaging with about 200 people whose research areas are at the crossroads of two major areas: major health issues and the theoretical challenges in signal and image processing. CREATIS responds to these challenges with a transdisciplinary approach involving four research teams belonging to information and communication sciences and technologies, engineering sciences and life sciences.
About LIRIS: LIRIS is a joined research unit (CNRS) in computer science with about 330 members. LIRIS research addresses a broad spectrum of computer science within its 12 research teams structured in 6 poles of expertise. The research conducted addresses the challenges of the digital world, including those posed by artificial intelligence (AI), big data analytics, computer vision, cyber security, digital transformation and human learning.