This week at the Women in AI in Healthcare Dinner we were joined by industry experts, both male and female, who came together to support women in STEM in healthcare and learn from industry experts.
As our host for the evening, Chief Science Officer at Unmind, Simmy Grover highlighted:
AI is an incredibly powerful tool, and as a result we are responsible for making sure it has a positive outcome and is a part of the healing process. AI can give us customised, affordable and scalable healthcare systems.
Insilico Medicine, who recently published a paper “proposing a blockchain-enabled solution to help people become aware of and take control over their medical data”, were kind enough to sponsor the champagne reception which kicked off the evening's activities, and Poly Mamoshina, lead contributor to their paper and attendee at the dinner thoroughly enjoyed the evening:
After the drinks reception, guests sat down for the three course meal, and Simmy kicked off the discussions by giving an overview of the landscape of AI in Healthcare and sharing some of the work Unmind are doing to provide a positive and proactive mental health platform to help businesses and oragnisations have happier and healthier teams. Simmy explained that traditionally the way in which we detect emotions of others if through elements such as voice and pitch, as well as facial expressions. This however isn’t the only way to determine if someone is say, happy or angry. We can ‘also analyse someone’s skin, their GPS location, and other elements to see if anything is out of the ordinary to identify if they might be feeling low or mentally unwell.’ She explained that in therapy you may be asked to rank your mental wellbeing on a scale from 1-10, but sometimes you may rank yourself incorrectly because you’re either unaware that you’re unhappy, or not ready to admit it yet. Through an amalgamation of different factors, AI can help identify if an individual may be struggling, and in turn can give us ‘a customised, affordable and scalable healthcare system’ with a positive outcome that will contribute to diagnosis and healing of patients.
After Simmy’s introduction, the guests continued networking over dinner before rotating seats to maximise the opportunity of discussing the industry with different attendees. We encouraged each table to select and write down an inspirational woman working in AI, and Anneclaire van Not from Cancer Research UK said:
At the RE•WORK table, we chose Daphne Koller from Calico who will be joining us at the Deep Learning Summit in San Francisco this January 25 & 26.
Farideh Jalali, Postdoctoral Research Associate and University of Manchester, was next to present, and discussed her work in applying machine learning methods for disease prediction and prevention in high risk groups. Currently, she’s working to predict more accurately which medicines are more likely to be most effective for which people and also identifying which patients are high or low risk. For example, psoriasis affects 2% of the population of the UK, and these patients are high risk. Roughly 30% of patients with psoriasis develop inflammatory arthritis, referred to as psoriatic arthritis. A study took 1400 patients with psoriasis and 1400 patients with psoriatic arthritis and and identified 3 potential varianty to be differently analysed in an attempt to predict 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. She explained that in doing this, a whole host of data is used and they’re currently testing the model with data from the UK bio band to make sure the results are consistent across different datasets. Farideh explained that there’s currently a problem with imbalanced datasets (i.e. the number of psoriasis patients is more than that of PsA patients), and they’re currently working to overcome this.
It’s very easy for AI to learn bias, and at Babylon we’re using generative Bayesian models to try and eliminate this.
People are inherently biased, and statistical bias exists in machine learning from algorithms, so these issues are present in healthcare. This comes from systematic inaccuracies true in psychology, statistics and all systems for example the underestimating of women and the overestimating of men! This isn’t just an ethical issue, but it’s inaccurate and Laura explained that at Babylon they’re aiming to identify and remove these inaccuracies. Often, they go unnoticed and get entered into machine learning models, for example in word vectors when you add king you get queen, but when you add computer programmer it’s female counterpart is ‘home maker’ which is terrible! Laura used an example of a model that was created in the US that ‘learned to be racist and was twice as likely to say that black people would reoffend when being analysed for crimes.’ This is discriminative, and needs to be overcome. When large amounts of biased data are loaded into a system it also teaches the machine to generalise and results get amplified. To overcome this, Babylon are using bayesian generative models to analyse risk factors that cause disease and symptoms, and these models give a way of generating solutions and also predicting outcomes. These models are very interpretable and easily traceable to their source which overcomes the issue of bias, and these algorithms do not amplify the data. However, all data is subject to bias and overconfidence, so this is something we need to take into consideration.
After Laura’s presentation, guests moved around the room once again to continue networking, and we had the chance to talk to several attendees about their evening and their opinions of Women in AI in Healthcare, and AI in Healthcare more generally:
Poly Mamoshina, Insilico Medicine:
I am glad that Insilico Medicine supports women in AI, and it’s always a good opportunity to meet so many like minded people!
Ekaterina Safonova, Cybertonica Ltd
In AI, for every group of boys, there's one girl, and usually she’s smarter!! Why are girls afraid? - because they’re under supported and we need to get better at supporting them and encouraging them into the industry.
We asked one of our male attendees, Steve Collins from Frontline Ventures how we can encourage more women into AI?:
It’s easy, we just need role models. The more role models we have, the more we can celebrate women that’s the way we can encourage young women into AI.
Alison Pouplin, Imperial College London
In AI in Healthcare, there are new improvements and advancements in the field every day like new articles about algorithms that can discover diseases and this is so exciting!
Our final speaker of the evening was Yinyin Yuan, Team Leader at the Institute of Cancer Research. Yinyin explained how she began her career in computer science where she was faced with the opportunity of ‘encountering types of data I never knew existed’. Her team uses techniques from a broad range of scientific fields to formulate unique approaches for linking genetic mutations, pathological observations and patient treatment to improve cancer research. She explained that pathology is a the centre of cancer diagnosis. In traditional methods, a pathologist uses a microscope to look at cells to determine how advanced the cancer is - ‘this is all very well, but the data is undervalued’. 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 and Yinyin’s research focuses on the emerging concept of tumours as ecosystems. Her team are currently working to develop machine learning approaches to classify and identify cells to differentiate between healthy and cancerous cells. The model provides you with rich spatial information to analyse making the analysis more accurate.
In an experiment at the Royal Marsden, samples from 1178 patients on a clinical trial were taken and they received homogeneous treatments - their cells were digitally analysed using single shot deep learning to detect and classify components and classify the cells into cancerous cells, immune cells, and ‘other’ cells.
The analysis Yinyin is using is 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 for breast cancers. 'If we only studied the cell abundance and not the spatial structure, we may not have been able to predict the recurrence of cancer in specific patients.’
If you’re interested in learning more about AI in healthcare, join RE•WORK at one of our upcoming events. Save 25% on all summit passes when you register before next Wednesday 29 November using the code CYBER25.