The application of Artificial Intelligence (AI) and Machine Learning (ML) techniques across a vast array of problems in science, medicine and technology is fueling a revolution in our capabilities for processing and extracting knowledge from data. Biomedical imaging is one area where AI/ML processes are offering particularly exciting opportunities and demonstrating important results.

This workshop will bring together a set of speakers who are experts in the use of AI/ML in the medical domain with an audience of researchers, academics and research software engineers to demonstrate some of the research and developments going on in the field.

The talks will be followed at around 5pm by drinks and pizza.

There will be an opportunity to chat with the speakers and network with other attendees from a range of institutions who are interested in the application of AI and Machine Learning technologies.

The event is open to anyone interested in advanced techniques for large-scale biomedical image analysis and will be of interest attendees from both the Computer Science/Software Engineering community and the Medical domain.

The workshop is being organised by the “Data-driven, Machine Learning-based biomedical image analysis” networking group funded under the Francis Crick Institute’s Partnership Networking Fund. The group brings together a group of academics, researchers and Research Software Engineers from The Francis Crick Institute, Imperial College London, University College London and King’s College London.

Speakers include:

• Joe Ledsam, Deepmind
• Nick Pawlowski, Imperial College London
• Emma Robinson, King's College London
• Carles Bosch, The Francis Crick Institute
• Martin Jones, The Francis Crick Institute

Schedule:

Arrival and registration from 13:30

14:00 Welcome / Introduction

14:10 Artificial Medical Intelligence at Google Health and DeepMind, Joe Ledsam, Deepmind

14:50 Probabilistic Modelling for Detecting Outliers in Medical Images, Nick Pawlowski, Imperial College London

15:10 Coffee Break

15:40 Predicting Cognitive Disorders through Machine and Deep Learning, Emma Robinson, King's College London

16:10 Learning across 8 million patients, Marc Modat, King's College London

16:25 Segmentation of volume electron microscopy data using crowdsourced annotations as ground truth, Martin Jones, The Francis Crick Institute

16:40 Lightning talks:

• Philip Noonan, King's College London
• Tim Hogenboom, Imperial College London

16:50 Closing remarks

17:00 Networking - Food and drinks available

Talk details:

Artificial Medical Intelligence at Google Health and DeepMind, Joe Ledsam, Deepmind

Artificial intelligence has a profound potential to improve patient outcomes. A holistic, patient centered approach is required to achieve this. The applications of artificial intelligence to research topics in healthcare cover a broad range of areas including algorithm development, translational research, clinical validation and even new biomarker discovery. Google Health and DeepMind are active across this diverse research landscape, and this talk will describe their approach for developing and implementing artificial intelligence in clinical practice, with particular emphasis placed on the work of DeepMind in Medical Imaging. The talk will be given by Joe Ledsam, a clinician scientist at DeepMind, who works on applications of machine learning to challenges in health and science.

Probabilistic Modelling for Detecting Outliers in Medical Images, Nick Pawlowski, Imperial College London

Advances in machine learning for medical image analysis are largely driven by data-hungry deep learning methods. Normative modelling offers an approach that allows for detecting abnormal conditions as outliers. This talk will introduce different ways of performing normative modelling using deep learning methods and explore the application to brain as well as histopathology images.

Predicting Cognitive Disorders through Machine and Deep Learning, Emma Robinson, King's College London

Machine and Deep Learning are having a significant impact on image processing of neuroimaging data but we still face some challenges when it comes to precision modelling of phenotypes, particularly those pertaining to cognition or complex neurological/psychiatric conditions of the cortex. This talk will discuss the challenges, and present some results of new methods trying to address these problems.

Understanding the brain with imaging data. Current challenges in systems neuroscience, Carles Bosch, The Francis Crick Institute

The architecture of neural connections can be densely imaged, potentially to the extent of full circuits, enabling an array of experimental avenues to explore how the brain works. However, the challenge that emerges has multiple sides: entire neural circuits must be recorded and imaged, so as to interrogate their function and their structure, in order to generate mechanistic understanding. This talk will explore the throughput and bottlenecks in data generation, processing, analysis and interpretation of correlative multimodal imaging pipelines in systems neuroscience.

Segmentation of volume electron microscopy data using crowdsourced annotations as ground truth, Martin Jones, The Francis Crick Institute

Whilst deep learning methods have revolutionised many fields of bioimage analysis, the major bottleneck in almost all projects is the lack of availability of ground truth training data. Our “Etch a Cell” project (etchacell.org) achieves expert-level ground-truth annotations from thousands of non-expert volunteers, via the Zooniverse citizen science platform. Consensus is achieved by aggregating multiple non-expert annotations per image. A U-Net deep-learning system trained on this aggregated data produces expert level predictions on large volume electron microscopy datasets.

----

Registration for this event is being handled by a group member based at Imperial College London. Attendee data received from Eventbrite will be held in accordance with Imperial's Events Privacy Policy.