On Thursday, 8 April 2021, Amsterdam Data Science (ADS) hosted its Women in Data Science (WiDS) event in collaboration with Amsterdam Medical Data Science. The afternoon featured experts sharing their latest innovations and research related to applying AI to the medical field.
“It’s about bringing different worlds together,” says Prof. Lynda Hardman in her introductory talk. As director of ADS she represents the importance of a multidisciplinary approach with a career that cuts across all the realms – from academia to business.
“That’s what ADS is all about: to help with next steps and connect you with the people to make those steps.”
Diversity in action: going beyond the hype
WiDS is indeed all about diversity and inclusion. The organisation aims to inspire and educate data scientists worldwide, regardless of gender, and to support women in the field. And in her opening academic keynote ‘AI and healthcare: Going beyond the hype’, Prof. Clarisa Sánchez was certainly inspiring.
As the Professor of AI and Health at the Faculty of Natural Sciences, Mathematics and Computer Science (FNWI) of the University of Amsterdam (UvA), and scientific director of two ICAI labs, Prof. Sánchez has been actively working to solve current challenges.
“While there is an exponential rise over the last year of articles being published that show how AI solutions can be highly effective, it’s rather scary and frustrating to see how few healthcare workers are actually using these!”
Even in a field like radiology where it’s already established that machines can outperform humans in recognising anomalies, the uptake is low and slow.
“We’re not being clear enough that this technology is ready to use. At the same time, we need to co-evaluate these solutions together with the main stakeholders – involve those that will use the actual applications from the beginning.”
Following on the study of ‘Automated Assessment of COVID-19 Reporting and Data System and Chest CT Severity Scores in Patients Suspected of Having COVID-19 Using Artificial Intelligence’, Prof. Sánchez co-led the study together with Dr. Lessman and Prof. van Ginneken. They worked actively with healthcare professionals and commercial partners. These included radiologists across the Netherlands, companies such as Thirona, and AI-researchers in Amsterdam, Nijmegen and Germany. Their common goal was to apply these findings to directly help patients. Barely a year later, the resulting CAD4COVID-CT is already certified and being sold to automatically screen for COVID-19 on CT and X-ray images.
“Domain knowledge is essential not only for the proper evaluation of these solutions but also for their design and development,” concludes Prof. Sánchez.
Breaking out: from deep tech to crossover
Next up were two very different examples from postdoctoral researchers. Giulia Bernardini is a postdoctoral fellow for CWI’s Life Sciences and Health department. Her rather technical pitch ‘Incomplete Directed Perfect Phylogeny in Linear Time’ confronted one of evolutionary biology’s biggest challenges: how to reconstruct a species’ history when information is missing – a common problem for anyone working with real data. Bernardini is currently finetuning an algorithm to do just that. “It’s actually based on a quite simple operation – no black box required,” she says. We’ll take her word on that.
With a background as a hospital pharmacist, Joanna Klopotowska has become a specialist in the crossover approach. And her speech dovetails nicely with the day’s recurring theme: only through intense collaboration across sectors and disciplines will you be able to bring AI innovations to the bedside faster and more effectively.
As an assistant professor at Amsterdam Public Health research institute of AMC-UVA and VUmc-VU, she’s working ‘towards a learning medication safety system – timely detection and prevention of adverse drug events’, as her pitch was called.
She’s a great believer in the Learning Health System approach, which seeks continual improvement in healthcare by embedding knowledge generation in daily practice. After gruelling years of gaining access and anonymising data, her team is now using the data of 100,000 IC patients from 15 hospitals to develop models that predict acute kidney damage caused by adverse drug events.
“We’re always looking for all sorts of people to join our team. It’s about diversity. It’s not only about medical expertise. Sometimes communications or psychological expertise is equally important.”
Deep dive into the data gap
The industry keynote speaker Lydia Mennes offered a suitable conclusion to the afternoon: ‘How to make AI fly in the clinical setting?’. As a healthcare NLP expert at CTcue, a company out to make Electronic Health Records (EHR) data accessible and useful for clinicians, she has both feet firmly in the real world.
She also bemoans how all this promising research is not yet being effectively applied. She blames a “data gap”, with medical data often being inaccessible, uninterpretable, poor quality, fractured, incompatible and/or inconsistent.
“And when bad data goes into a model, bad info comes out,” says Mennes. “But data collecting is improving!” Indeed, AI is not only being applied to improve patient outcomes but in interpreting patient records. However, there’s still a long road ahead.
“But imagine a future when we solve the gap! Then AI models can be implemented a lot easier and used a lot more. Imagine a medical app store…”
Meanwhile: back to work.