Beyond the Benchmark: Bridging the Clinical AI Implementation Gap

Speaker: Dr Brad Segal
Time: 18:00-19:00
Levett Room, Wolfson College & Online (Teams link).

Biography:

Dr Brad Segal is a clinician-engineer pursuing a DPhil in Biomedical Engineering at Oxford's Computational Health Informatics Lab under Professor David Clifton. His research focuses on robust deployment strategies for clinical machine learning systems, with particular emphasis on resource-constrained environments. As a practicing physician in South Africa's public health sector, he has unique insights into the practical challenges of AI implementation across diverse healthcare settings.Brad's work spans both technical development and clinical implementation, having co-founded healthcare ventures in predictive analytics and clinical decision support that now serve millions of patients across sub-Saharan Africa. As both a technology developer and clinical end-user, he brings practical perspective to the challenges of transitioning machine learning systems from research environments to clinical practice.

Abstract:

Despite remarkable advances in machine learning benchmarks for healthcare applications, the gap between research performance and clinical utility remains a critical challenge. Through a series of case studies drawn from implementations across diverse healthcare settings, this talk will analyse how common machine learning practices can lead to unexpected failure modes in clinical environments. We will explore how traditional evaluation metrics can mask critical shortcomings, as well as examining how the misalignment between model optimization objectives and clinical decision-making requirements can compromise real-world implementation.

The discussion will cover fundamental challenges in clinical ML deployment, including the impact of population-specific disease presentations on model generalization, technical constraints of clinical workflow integration, and trade-offs between interpretability and performance. Drawing from experiences implementing systems across various clinical contexts, we will explore potential approaches for identifying and mitigating these challenges, considering both theoretical and practical aspects of building clinical AI systems that better align with real-world healthcare needs.

Scalable and low-cost federated learning in the NHS using micro-computing

Speaker: Andrew Soltan
Time: 18:00-19:00
Levett Room, Wolfson College & Online (Teams link).

Biography:

Clinical academic at Oxford, Profile

Paper

Podcast

Abstract:

Training fairer medical AI needs diverse data, but hospitals are restricted in what they can share for privacy reasons. Here, I will discuss our new, easy-to-deploy way for hospitals to take part in AI development without sharing data, and our learnings from a pilot deployment across 4 NHS Trusts. Federated learning (FL) was first developed by researchers at Google as a way to train AI models without moving data. Researchers at NVIDIA, Rhino Federated Computing and University of Pennsylvania have since deployed FL in to hospitals to develop clinical models, but deployment relied on specialist technical expertise at every hospital taking part. Using cheap micro-computers, we built a platform for any hospital to easily take part in training and testing AI models without needing to share patient data. We developed software for FL and loaded it on to Raspberry Pi 4B devices, delivering ‘ready to go’ federated clients to hospitals. Using our approach, four NHS hospital groups developed and evaluated a COVID-19 screening test while retaining full custody of their data throughout, together building a more performant model. By making it easier to train models without moving data, we hope our new full-stack federated learning approach may lead to better and fairer models, while respecting patient privacy and data sovereignty.