Applying AI in a clinical setting: the long but rosy road

As Amsterdam knowledge institutes commit 1 billion euros to AI research in the coming decade, those already in the data science trenches remind us: while the potential of applying data science for the benefit for patients seems almost infinite, there’s still a lot of work to be done.

Deck the halls with a billion euros

Christmas came early at the latest Medical Data plus Pizza meeting at the Amsterdam UMC on 17 December 2019. Presenter Dr. Lucas Fleuren opened the evening by highlighting the article: ‘Amsterdam knowledge institutions invest 1 billion in AI’. You don’t need an algorithm to tell you that’s an amazing amount of cash.

“Under the banner ‘AI technology for people’, leading knowledge institutions in Amsterdam have committed to invest 1 billion euros over the next 10 years, to employing at least 800 researchers, to training 5,000 bachelor’s, master’s and PhD students, to having 10,000 students follow an AI-minor and to helping develop 100 spin-offs and 100 startups,” according to this extended press release from University of Amsterdam.

“In other words, it’s a good time to be in AI in Amsterdam,” says Fleuren.

Getting to the heart of right ventricular morphology

The presentation ‘Machine learning to determine regional changes in right ventricular morphology from MRI images’ came courtesy of Azar Kianzad, a PhD candidate for pulmonary medicine.

Kianzad is working with the UK researchers behind the 2017 study ‘Machine Learning of Three-dimensional Right Ventricular Motion Enables Outcome Prediction in Pulmonary Hypertension: A Cardiac MR Imaging Study’ to apply their model at Amsterdam UMC.

The purpose of the study was “to determine if patient survival and mechanisms of right ventricular failure in pulmonary hypertension could be predicted by using supervised machine learning of three-dimensional patterns of systolic cardiac motion.” Basically, the answer is yes: that right ventricular function is a major determinant of prognosis.

With pulmonary hypertension, the pulmonary vessel narrows which leads to increased vascular load on the right ventricle (RV). The RV adapts by changing form, which results in an increase in wall stress and oxygen consumption – with a potential final and usually fatal stage: “uncoupling”. Unfortunately, the mechanisms behind chronic RV adaption are still not fully understood.

Meanwhile: while the algorithm may be efficient in predicting survival based on markers in the right ventricle’s 3D motion, it works in a black box and therefore it’s not clear what exact features are being used to calculate probabilities.

Kianzad hopes to help develop the study to see if motion analysis can also be applied to predict the causes behind pulmonary hypertension, to detect early clinical deterioration and to better understand the mechanism behind RV adaption. “But it’s all very complex,” she says. The audience nods in agreement.

Turning billions of words into real-time actions

Azar’s work is largely inspired by a single study. Now imagine if millions of articles can be mined for correlated information. Elsevier, “the global information analytics business”, publishes close to half a million articles a year in more than 2,500 journals. Its archives include over 16 million documents and 30,000 eBooks. How can this fountain of medical knowledge be focused and fed directly to doctors to support clinical decision-making?

Elsevier’s Dr Ewoud Pons is working to turn these lofty goals into reality. Or, as he puts it in his presentation ‘Semantics-driven decision support and patient pathway identification’: “How do we leverage all this content and automate fact extraction to inspire actionable insights?”

Pons certainly has the right background: after obtaining his medical degree, he started a multidisciplinary PhD in radiology and medical informatics. During his studies, he used natural language processing and machine learning on electronic health records to generate tailored imaging advice to clinicians at the point-of-care.

Pons believes that adapting scientific literature into real-time guidelines will first involve human-assisted curation, but that ultimately the automation of guideline creation is possible. But yes, other challenges are also involved. For example, how do you insert these guidelines and models at the right time during the clinical process? And, perhaps more importantly, how do you convince doctors to use these models?

“This all relates to acceptance, so it’s quite complicated,” says Pons. “I think it should be soft recommendations where a ranked list of guidelines are offered that are the most relevant to the case history. And then over time and through daily use, we can then hopefully create a positive feedback loop.”

Then, in the name of creating another positive feedback loop, pizza was served. Several small groups discussed how far 1 billion euros can take the local AI and data science scene. Well, to put the number into perspective, that amount could buy around 150 million pizzas.

That’s a whole lot of networking.

About Amsterdam Medical Data Science

The Medical Data plus Pizza meeting aims to bridge the gap between health professionals and data scientists by bringing both together in an informal setting for presentations and pizza. At Amsterdam UMC data scientists, medical professionals and researchers discuss how AI (Artificial Intelligence) and medical data can be used to benefit patients and improve medical practices. Since its 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. For more reports from previous Medical Data plus Pizza Meet-ups, click here.

The Amsterdam Medical Data Science Group meetings are supported by The Right Data Right Now consortium, which includes Amsterdam UMC, OLVG, Vrije Universiteit, Pacmed and the Amsterdam Economic Board.

#slimgroengezond

This contributes to the development of the Amsterdam Metropolitan Area as the European Life Sciences & Health hub