Women in AI in Healthcare Dinner schedule

Simmy Grover is the Chief Science Officer at Unmind. Her role is to ensure the academic and scientific validity of our platform and interventions. Simmy is an organisational psychologist, who is completing a PhD in Organisational Psychology at University College London, under the supervision of Prof. Adrian Furnham, with research including individual differences, interpersonal relationships, organisational processes and structure and their impact on individual and organisational performance. Her most recent academic publication was a systematic review of the effectiveness of coaching in organisational settings and she presented these findings at the SGCP 4th International Congress of Coaching Psychology 2014. Prior to her career in occupational psychology, Simmy spent a number of years in financial services. She started her career as an equity trader at Morgan Stanley but her most recent role was to establish a European business for Aite Group, a US research/consulting organisation, and Simmy was honoured as one of Financial News’ “40 Under 40 Rising Stars of Trading and Technology“ in 2011. Together with being widely quoted in the press, in publications including Bloomberg, Reuters, CNBC, The Financial Times, and Financial News, she also spoke at key industry events across Europe and Asia, including TradeTech Europe, TradeTech Japan, International Trader Forum, FPL Japan and FPL EMEA..

Simmy holds a Masters degree in Biomedical Engineering from Imperial College London, as well as a Masters degree with Distinction in Social Cognition from University College London. She is fully accredited in Hogan Personality Assessments and has graduated from the Meyler Campbell Business Coach Program.

Application of Machine Learning Methods for Disease Prediction in High Risk Groups

Genetic variation within the major histocompatibility complex (MHC) is associated with susceptibility to many autoimmune diseases. However, the genes within this region are highly polymorphic with extensive linkage disequilibrium (LD) making feature selection using statistical models very challenging. Machine learning methods are particularly suited to this challenge where feature selection approaches try to find a subset of the original variables that enable more accurate prediction by the elimination of irrelevant and confusing information. Accurate feature selection is essential for predicting disease in high risk groups. For example, approximately 30% of patients with psoriasis develop an inflammatory arthritis referred to as psoriatic arthritis (PsA). Accurate prediction of psoriasis at high risk of developing PsA would allow early intervention and limit the impact on a patient’s quality of life. We focus on the filter feature selection methods based on information theoretic criteria that are classifier independent methods that provide the ranking of genetic features for differentiating PsA from psoriasis. The ensembles predictive models were trained and evaluated ‘with feature selection’ and ‘without feature selection’.

Farideh is a postdoctoral research associate and teaching associate fellow in arthritis research UK centre for genetics and genomics school of biological sciences faculty of biology, medicine and health at Manchester University. Her work aims to bring insight from the field of machine learning to solve clinical problems that can provide clinicians with actionable information. Before joining the arthritis research centre, Farideh applied machine learning techniques to develop an objective patient’s voice-assessment model that is now being used in hospitals. After completing her MSc in microelectronics engineering with distinction level at Newcastle University she was awarded PhD studentship from the school of computer science at Manchester University. During her PhD, she was awarded people’s choice at the University of Manchester for her three minute thesis presentation.

Bias: Statistical and Significant

Cognitive bias exists in people, statistical bias exists in machine learning algorithms, both exist in healthcare. Given it is easier to remove biases from algorithms than from people, AI has the potential to create a future where important decisions, such as hiring, diagnoses and legal judgements, are made in a more fair way. However, if we don’t actively try to model and remove theses biases, we can end up simply propagating them into future. This is the path we are currently on. The current gold standard word vectors are inherently sexist and there are courts in the US using an algorithm which is inherently racist. Of course bias isn’t just about race and gender, those are just some of the easiest places to notice injustice. In healthcare there exist biases for all sorts of reasons, both in the data and the algorithms, I will talk about ways we can go about understanding and removing these biases through machine learning.

Laura is a Research Scientist at babylon health. Her current research uses Probabilistic Inference and Bayesian Networks to model the diagnostic process of a GP. Previously, she has researched ways to predict patient outcomes for a health tech start-up in Singapore, and built state of the art Natural Language Processing tools for another London based start-up. She holds an MSc in Machine Learning from UCL and a Masters (Part III) in Maths from the University of Cambridge.

Deep Learning the Ecological Niches of Cancer Cells for Combating Treatment Resistance

Tumours consist of not only cancer cells, but also normal cells such as immune cells that can be critical in eliminating cancer cells. These different types of cells co-exist in different parts of the same tumour with profound clinical implications. Just like in ecology where spatial organisation of animals, their predators and habitats is central for understanding the ecosystem and make prediction, It is becoming increasingly evident that we need to use a similar spatial approach to evaluate tumour heterogeneity.

My team at the Institute of Cancer Research develops machine learning and deep learning approaches to identify different types of cells in digital pathological images of tumour sections based on their differences in appearance. Such automated image analysis allows us to map their spatial distribution within the tumour of a patient. The next step is to quantify spatial variability of these cells, usually in the order of millions, using spatial statistics.

Our recent study on breast cancer and lung cancer underscored the importance of examining spatial heterogeneity of the tumour. We studied how immune cells are spatially arranged within the tumours, and detected the so-called immune hotspots, which are tumour regions that contain spatial clustering of immune cells. This uses a spatial statistical method called Getis-Ord Hotspot analysis, which is commonly used for detecting crime hotspots in cities. High amount of immune hotspots, but not the amount of immune cells, correlates with high probability of cancer recurrence. This study provides a new way to predict patient prognosis, and open the door to new therapeutic opportunities using immunotherapy across cancer types.

Yinyin Yuan joined the ICR in 2012 as the leader of the Computational Pathology and Integrative Genomics team. Currently, her team is part of the Centre for Evolution and Cancer and the Division of Molecular Pathology. Yinyin was trained in computer science and bioinformatics. She obtained her academic degrees in computer science during her education at the University of Science and Technology of China (BSc 2003) and University of Warwick (MSc by research 2005, computer vision and steganography; PhD 2009, machine learning and bioinformatics).