Motivation

FL offers a pathway to achieve this by allowing for the collaborative analysis of EHR across institutions while maintaining strict data privacy.

About EHR

As healthcare systems around the globe grapple with increasing amounts of data, there is a growing imperative to utilise the data in a way that is both secure and effective. Especially, Saudi Arabia is actively pursuing a transformative agenda within its healthcare sector as a core component of Vision 2030, aiming to embrace digitalisation and sophisticated data analytics.

Federated Learning

FL offers a pathway to achieve this by allowing for the collaborative analysis of EHR across institutions while maintaining strict data privacy. This process involves each institution developing local models based on its own EHR data, while a central parameter server coordinates the local efforts to synthesize a comprehensive global model. The trained FL models can make intelligent inferences in many ways, such as discerning patterns that signify high-risk patients, which can facilitate timely and targeted healthcare interventions, identifying correlations among patients' attributes, which can inform treatment strategies and healthcare policy decisions, and providing alerts for potential health condition changes, which supports preemptive healthcare measures.

Challenges

However, adopting the FL approach for EHR in Saudi Arabia is currently impeded by the following significant challenges:

Data Heterogeneity

Healthcare institutions often use varied formats and attribute sets for patient records, resulting in heterogeneous EHR data. This inconsistency poses a significant challenge to effective training of FL models.

Data Quality Disparity

Traditional, paper-based health records may have missed or inaccurately entered data, leading to variable data quality when these records are digitised.

EHR Dynamism

The ongoing digitisation of healthcare records and the evolving nature of patients' health conditions contribute to the dynamic quality of EHR, necessitating adaptable and responsive FL models.

Our Aim

By realizing these objectives, the project will aim to bridge the gap between the potential of FL and its practical, effective application in healthcare analytics, ultimately contributing to the improvement of patient outcomes and the efficiency of healthcare systems.

Design and Implement novel algorithms to address data heterogeneity issues in EHR

ensuring diverse representation across all data inputs are learned effectively

Create a comprehensive framework for the consistent FL model assessment

enhancing data quality within the FL model and ensuring the integrity of input data

Develop adaptive FL models that can efficiently process real-time EHR data updates

supporting continuous learning and reducing the need for frequent model retraining

Diagnostic Accuracy Improvement

improving the accuracy of health and disease diagnosis by creating algorithms that effectively manage the variability found in EHR data from multiple institutions

International Collaboration and Innovation

fostering a collaborative international effort, particularly between the UK and Saudi Arabia, to innovate and advance FL in healthcare analytics, setting a precedent for global scalability and cooperation

Sustainable Development Goals

The project outcomes are expected to contribute to several Sustainable Development Goals (SDGs)

Good Health and Well-being (SDG 3)

By enhancing the accuracy and reliability of disease diagnosis through Federated Learning with EHR, the project directly supports efforts to improve health outcomes and promote well-being at all ages

Quality Education (SDG 4)

The dissemination of research findings and the development of a prototype provide valuable educational materials for students and professionals in healthcare, informatics, and data science.

Industry, Innovation, and Infrastructure (SDG 9)

The project's innovative approach to handling healthcare data with advanced algorithms and machine learning models contributes to build resilient infrastructure and fosters innovation.

Reduced Inequalities (SDG 10)

By ensuring diverse and equitable representation in healthcare data, the project works towards reducing inequalities in health outcomes across different demographics.

Partnerships for the Goals (SDG 17)

The collaborative nature of the project, potentially involving multiple institutions and stakeholders, exemplifies the goal of revitalizing global partnerships for sustainable development.

Our Team

In recent years, our research team has laid substantial groundwork in federated learning (FL), graph processing and computational pathology, equipping us with the necessary expertise to tackle the challenges presented in our current proposal.

Lead Researchers

Dr Ligang He

Specialised in Parallel Computing and Federated Learning

Dr Mohammed Alghamdi

Specialised in Parallel Computing and Medical Analysis

Dr Hamman Alghamdi

Specialised in Medical Analysis

Federated Learning in Electronic Health Records

Developing Adaptive Federated Learning Frameworks for Heterogenous and Dynamic Electronic Health Records (EHR)