Using data science in neurosurgery and radiology
The final Amsterdam Medical Data Science meetup of 2018 continued the event’s success by bringing together data scientists and medical professionals at Amsterdam UMC to share insights, experiences and research in medical data science. The event aims to bridge the gap between health professionals and data scientists by bringing both together in an informal setting for presentations and pizza. Since it's launch in August 2018 it has steadily grown in popularity with more and more people attending to hear the latest news and innovations in medical data and spark new collaborations.
The session’s host, Dr Lucas Fleuren, introduced another full meetup by asking members to identify two senior members of staff at Google – Jeff Dean, its head of artificial intelligence, and David Feinberg, its new head of health. “It’s interesting that Google has hired a healthcare CEO, and he actually reports to the CEO of artificial intelligence,” Dr Fleuren remarked. “I think that shows that outside of the medical and academic world a lot of big companies are investing in artificial intelligence in the healthcare domain.”
Using AI to develop a ‘drug atlas’
The first talk of the night was from Dr Bart Westerman, who works at Amsterdam UMC’s neurosurgery department. Dr Westerman introduced a project that he worked on which aimed to design a platform to help predict the best synergistic drug combinations to treat glioblastoma, an aggressive form of brain cancer which is highly likely to spread. Glioblastoma has a median survival rate of just of 14 and a half months, and standard therapies, including surgery, radiotherapy and chemotherapy, are often ineffective. Personalised treatments using synergistic combinations of drugs have proved to be the most effective treatment for patients with glioblastoma, yet often prove challenging to prescribe as effective drug-synergy is hard to predict. In total, 400 FDA-approved anti-cancer drugs can be combined into around 15 million combinations for each patient when they are taking three of the drugs together.
Despite this enormous challenge, Westerman showed how the project had used machine learning algorithms to connect four layers of phenotypic and chemogenomic data into a ‘drug atlas’ that can be used to predict the best course of combination treatment for patients. It used data from the ‘Genomics of Drug Sensitivity in Cancer’, a database containing drug sensitivity data for almost 75,000 experiments, describing the response to 138 anti-cancer drugs across almost 700 cancer cell lines. Dr Westerman showed that the drug atlas had been validated using cell culture experimentation and benchmarked it against known data sets. He said that one advantage of the drug atlas is that “it doesn’t need combination data to learn because we made a prediction model based on literature”. Dr Westerman finished his presentation by showing that therapy combinations recommended by the drug atlas had helped to prolong the lives of mice and revealed that he was also working on a ‘gene atlas’ that could help predict the evolution of tumours based on genetic interactions.
How AI can help to diagnose and treat stroke
Henk Marquering, who works within two of Amsterdam UMC’s departments (radiology and nuclear medicine and biomedical engineering and physics), gave the event’s next talk. Marquering trained as a seismologist before moving into the medical field around 16 years ago. For the last seven years his work has focused on stroke, and his talk focused on AI in neuroradiology to help identify and treat strokes.
Marquering introduced a research project which developed AI to identify and detect thrombus – or a blood clot – in the brain. An occurrence of thrombus can lead to stroke, which is the second-highest cause of death in epidemiology and affects 41,000 people a year in the Netherlands. As well as diagnostic value attained from the study, it also helped to discover a new measure – thrombus perviousness – which is used as an imaging biomarker to estimate clot permeability from CT imaging. The more pervious a thrombus, the more fluid it lets through, meaning an improved outcome for patients.
He also discussed the trial of a new intra-arterial treatment for stroke called MR CLEAN, in which retrievable stents were placed into arteries to extract thrombus and help patients recover from acute ischemic strokes. He showed that the treatment, which is now standard in medical practice, had helped increase the number of patients who could live independently after having a stroke. After the trial a registry was also launched, which registered every patient treated with thrombectomy. Marquering said that this resulted in more data for data scientists and doctors to use in research projects.
Marquering also showed how a French student he worked with had developed convolutional neural network deep-learning methods to analyse CT-scans and quantify the amount of blood in patients. This helped to predict the prognosis of patients who have suffered a haemorrhagic stroke, including the likelihood of delayed cerebral infarction. Another study Marquering discussed also created machine learning models using just six image features in CT-scans to make similar predictions. He said these models are improving, but admitted it was difficult as doctors often found it difficult to interpret the machine learning techniques and results. He ended his talk by introducing Nico.lab, a company created using some of the research he was involved in, which uses artificial intelligence algorithms for accurate, quick analysis of relevant biomarkers from stroke imaging, including NCCT, CTA, dynamic CTA and follow-up imaging.
Reflecting on a successful 2018
As always, after the talks, attendees shared ideas, stories and connected over pizza and soft drinks. Reflecting on the event’s success as 2018 draws to a close, Dr Fleuren said that he had been amazed at the interest in the meetups. “We have been overwhelmed by the enthusiasm of everyone involved and we actually already have speakers scheduled for a lot of upcoming meetings as people are excited to share their research and projects with us,” he said. “I remember starting out in the summer and we didn’t know whether people would be motivated to come – but as soon as we set up the Meetup account, the level of interest was huge. Looking to 2019 our aim is to set up a concrete collaboration or project.” The meetup takes place on the third Tuesday of every month at Amsterdam UMC.
About Amsterdam Medical Data Science
The Amsterdam Medical Data Science meetups are supported by The Right Data Right Now consortium, which includes Amsterdam UMC, OLVG, Vrije Universiteit Amsterdam, Pacmed and the Amsterdam Economic Board.
For more reports from previous Medical Data plus Pizza Meet-ups, click here.
Want to read more articles like this one?
Follow us daily on LinkedIn, Twitter and sign up for our Board Update.
The merger of our Smart Health Amsterdam initiative with Amsterdam AI became ...
Smart Health Amsterdam is merging with Amsterdam AI. Geert ten ...
Bringing people together to accelerate important developments: the Amsterdam Economic Board and ...