Just in time for Christmas…
On 15 December 2020, the Data plus Pizza meetings returned for the first time since the coronavirus outbreak. The organising Amsterdam Medical Data Science (AMDS) network – with 1,700 members and counting – has co-hosted various online events in the past few months to keep developing the connection between healthcare and data science.
“COVID-19 is no joke, we are busy,” says one of the founders of AMDS, Dr Paul Elbers, who as an intensivist at Amsterdam UMC is a front-line medical professional. However, the quest to apply data science to improving patient care has also continued. For example, ICU units in the Netherlands began working more closely together during the pandemic’s first wave and this has now evolved into a game-changing project: ‘Dutch intensive care units collaborate on ‘world first’ data-sharing project’. “We are really the first country to share ICU data at this scale,” says Dr Elbers.
Seeding innovation in AI and healthcare
Under the name Data without Pizza #1, this online version of an AMDS meetup continues to confront the various challenges in applying data science and AI to healthcare – while discussing the latest innovations and research.
Now, AMDS is also instigating research. Two months ago, a call for proposals was made for seed projects that could fuel collaboration between data scientists and medical professionals. A total of four pitches have now been chosen to receive a €5,000 seeding grant based on the relevance of the medical problem, the relevance of the data science aimed at solving that medical problem, group expertise, and the appropriateness of the budget. This evening, the winners gave short presentations about their respective projects.
From ‘Scientific Tinder’ to gastroenterology
Under the snappy name ‘Scientific Tinder for AMDS’, Dennis Diederix of the Amsterdam Data Collective presented a project to “find those needles in the haystack of important research articles on PubMed, so practitioners can stay on top of the most important developments in their field.” In short, “key opinion leaders” can swipe right if they like a particular article – and articles with the most swipes are then shared with the larger community. With €5,000, Diederix hopes to begin rolling out the service with Amsterdam UMC as its test client.
Meanwhile, ‘Real-time polyp classification in colonoscopy imagining data’, presenter Leon Cullens of the University of Amsterdam says their project will use the grant to hire in a student to start labelling the hundreds of raw videos of colonoscopies that can be used to teach a model to recognise whether a polyp is cancerous – and if so, at what stage it is. Any remaining cash will be used towards developing a proof of concept.
From VR surgery to predicting mental health
Chris Hordijk is founder of MedicalVR, and part of the team that scooped one of the seed funds for its ‘Visualisation of coronary artery calcification’ project. He explained how they want to train a neural network to recognise differences – and eventually visualise in virtual reality – the CT-scans of artery calcification. This would allow surgeons to practice operations on a patient’s digital reproductions before carrying out the actual operation. This technique has already been clinically validated for removing lung tumours – saving up to 50% more healthy lung tissue for the patient.
“We’re out to dive deep into the causes of adult psychopathology,” says Dirk Pelt, a postdoctoral researcher at Vrije Universiteit Amsterdam. He presented the project ‘Predicting adult mental health from an integrated data-driven childhood perspective’.
Pelt and his team hope to harness more insights by bringing together different data sets – such as combining the rich genetic and familial information that can be found in the National Twin Register with the socio-economic nitty gritty that can be gleaned from postal code databases. “We want to use the grant to link as many relevant databases as possible,” says Pelt. “Obviously, this is a complex and multileveled question that involves an interplay between different genetic and environmental zones.”
Predicting admission risk for heart patients
Olivier Witteman is a recent graduate of the TU Delft Aerospace Engineering programme. “But battery pack development was not my future,” he says to open the presentation: ‘Predicting admission risk for heart patients’. Having shifted to computational intelligence, Witteman now works at the VuMC on integrating a Cardiology Hospital Admission Risk Predictor (CHARP) with the EPIC patient health record environment.
“This is work that makes a difference – you can benefit people directly,” says Witteman, explaining his newfound passion. Indeed, cardiovascular disease remains the leading cause of death globally, and is the costliest of noncommunicable diseases since it often involves hospitalisation. By being able to identify the most high-risk patients, the project aims to decrease both mortality and costs.
“There’s just too much data for doctors to access properly,” says Witteman. “But we can use machine learning to reduce all this data into a single risk value that doctors can use within their decision-making process. We can also use it to assemble a list of high-risk patients and do feature analysis to potentially find the cause behind the hospitalisation.”
By using a “relatively simple” deep neural network, Witteman’s model already outperforms traditional risk assessments. Now, he wants to further improve it by applying Long Short-Term Memory (LSTM) networks which are better capable of learning long-term dependencies. Meanwhile, he hopes to have CHARP fully integrated into EPIC by next summer.
Earlier that day, Witteman heard that CHARP was granted approval to use with the ODISSEI supercomputer. “Now we’ll be able to add other features from CBS Statistics Netherlands such as mortality to the data set. It also means we have the computer power for intensive Shapley-value based features analysis,” enthuses Witteman. “This really opens up whole new possibilities!”
AMDS & Smart Health Amsterdam
The data and pizza meetings are supported by The Right Data Right Now consortium and organized by AMDS; The Amsterdam Medical Data Science Group (AMDS)
Smart Health Amsterdam is the network for data- and AI-driven innovation in Amsterdam’s life sciences and health sector. Join our community here
This contributes to the development of the Amsterdam Metropolitan Area as the European Life Sciences & Health hub