Clinical radiology Artificial Intelligence (AI): A blended learning programme (January, online): Fully booked

Timing

Topic

Speaker

09:30

Introduction

Dr Susan Shelmerdine, Consultant Paediatric Radiologist, Great Ormond Street Hospital

09:40

Overview of Online Learning Material

Discussion of use cases and applications for AI for different areas of radiology including the basic jargon and development of AI algorithms with opportunity to ask questions related to the topics from the online e-modules.

Dr Sarah Hickman, Radiology Registrar, Barts Health NHS Trust

10:30

Comfort break

10:45

A primer to data curation and processing for machine learning in clinical radiology

A practical interactive workshop using Google Colab workbook to walk delegates through the basics of AI model building, using an exercise in Python.

Led by: Dr Mitch Chen, National Institute for Health and Care Research (NIHR) Clinical Lecturer in Radiology, Imperial College London

Facilitators:

Dr Mathew Storey, Radiology Registrar, St George's Healthcare NHS Trust

Dr Amanda Isaac, Musculoskeletal Diagnostic and Interventional Consultant, Guy’s and St Thomas’ NHS Foundation Trust

Dr Sarah Hickman, Radiology Registrar, Barts Health NHS Trust

12:15

Comfort break

12:30

Navigating the Complexities of Open-Source Datasets and AI Grand Challenges: Mitigating Bias and Unintended Outcomes in AI Models

Discuss the limitations and benefits of open-source datasets. What are grand challenges and what biases happen when we implement tools designed this way?

Dr Kristofer Linton-Reid, Schmidt Futures AI in Science Research Fellow, Imperial College London

13:15

Addressing Small Data Challenges in AI for Radiology: Exploring the Potential of GANs, Transfer Learning, and Federated Learning

How can we leverage small datasets or rare diseases to make AI models?

Cato Pauling, UCL Great Ormond Street Institute of Child Health (GOS ICH)

14:00

Summary and Q&As

Dr Susan Shelmerdine, Consultant Paediatric Radiologist, Great Ormond Street Hospital

14:15

Close for a day

Module Title

Learning Objectives/ Outcomes

1.     Introduction to AI in Radiology and Healthcare

By the end of this introductory session, learners will have an understanding of the fundamental principles underlying AI; describe different AI methods/approaches, their strengths and limitations and justification for these approaches

Learning Objectives:

·       Reiterate learner’s current knowledge and understanding of AI / clarify learner’s doubts about what AI, ML, DL means and the differences between these.

·       Explain the differences between what is meant by AI, radiomics, CAD.

• Provide an overview of AI techniques and applications

Understand what is meant by:

• A ‘use case’ (what makes a good use case)
• What do we mean by ‘ground truth’
• What are traditional machine learning methods
• Understand different forms of learning: Supervised, semi-, and non-supervised, active, transfer learning
• Understand the concept of Artificial neural networks

·       Justify the use of specific AI models and limitations.

Provide learners with examples of:

• AI for medical image analysis and imaging (classification, segmentation, object detection, improved image reconstruction/ reduced dose etc)
• AI for data analytics and data mining (e.g. administrative tasks, augmenting clinical workflows)

Suggested exercise for learners will be to search the literature for AI use cases in their particular subspecialty area of interest and to brainstorm new ideas that could be developed that they view as common problems in their field which AI could potentially solve.

2.    Building an AI : Key Concepts

By the end of this session, learners will have a general understanding about how AI models are developed/built; with a particular focus on data collection, annotation and issues surrounding information governance and data security.

Issues surrounding open source datasets, awareness of AI grand challenges (e.g. Kaggle) and how bias and unintended outcomes may be introduced into AI models.

Learning Objectives:

·       Understand the process of producing an AI algorithm (generic stages of development, without coding).

• Understand the common pitfalls and considerations when building AI models

Data sources

• Understand how to prepare imaging data for AI development and data annotation
• Have an awareness of information governance issues surrounding data sharing and data security/privacy e.g. GDPR, data protection guidelines
• Awareness of how data scientists can work with limited data (e.g. GANs, recap of transfer learning as concepts, federated learning as a concept)
• Understand the importance of collaboration and open datasets to build robust AI
• Have an awareness of AI grand challenges and how these help to progress development of AI, and steer the use-cases for AI in healthcare.

Bias

• Have an awareness regarding the difficulties of generalizing AI models
• Understand how algorithms can perpetuate or exacerbate existing biases in datasets and society
• Explain the need to build bias-mitigating or bias-free systems, and the unintended consequences of not doing this.