Artificial Intelligence (AI) applications can help find an Alzheimer's drug, detect tumors faster or predict the effects of drugs down to the individual level, just to name a few research fields. With thanks to AI, breakthroughs in numerous medical applications can be expected in the coming decades.
And yet: when you talk to researchers who work with AI on a daily basis, there is still a large gap between expectations and reality. As often, the new technological possibilities run up against the limits of existing structure and culture. In this article, we take stock of the status of AI within medical science and provide an example of how not only science but also business and healthcare as a whole can benefit from it.
The development of AI technology is at a tipping point. More and more AI applications are making the transition from the lab to society. The technology should become more and more embedded over time. In order to strengthen the Dutch position in the field of artificial intelligence and to cash in on opportunities, the multi-year AiNed program (Growth Fund) has been established by the Dutch AI Coalition, a public-private consortium with more than 400 participating organizations.
The Growth Fund fits within a movement that has been going on for some time. After 2010, AI definitively and widely made the transition from the scientific world of the lab to practice. The number of AI patents is rising, investments in the technology are increasing and governments are also betting on it. However, there are still various recommendations for the government to support the embedding of AI in society, concludes the report 'Opgave AI. De nieuwe systeemtechnologie' of the Scientific Council for Government Policy (WRR).
AI technology itself is also advancing by leaps and bounds. The lightning-fast developments in AI, data science and digitization can accelerate scientific progress in all fields, from fundamental to applied research. For healthcare, it is no different. When it comes to medical scientific research, you could even say that AI has definitively established itself as a useful analysis tool for numerous studies.
Across the country, several AI labs are currently taking shape, such as ICAI labs and ELSA labs. ICAI stands for Innovation Centre for Artificial Intelligence and focuses on technology and talent development between knowledge institutions, industry and government. ELSA refers to Ethical, Legal and Societal Aspects. The goal of the ELSA Labs is to ensure that companies, government, knowledge institutions, civil society organizations and citizens work together to develop responsible applications of AI. This shows how many parties are working closely together to make steps forward.
In addition, a lot is already happening within South Holland's universities in the field of AI. Leiden, for example, offers a lot of expertise in the field of chemistry and small molecules. Delft is far along in automation and robotics, and Erasmus has many valuable datasets of patients with which AI systems can learn. In 2020, TU Delft launched the first "TU Delft AI Labs. These labs focus on collaboration between experts in the most advanced AI technologies with domain experts.
But what do we notice from this now in terms of concrete applications for healthcare?
A good example is the research of Dr. Sebastian van der Voort (Erasmus MC) on the use of artificial intelligence to predict the aggressiveness of brain tumors based on MRI scans. His dissertation on the use of AI for assessing MRI scans of patients with brain tumors was judged cum laude. It demonstrates how AI technology ultimately has the potential to improve healthcare.
Normally, a physician (pathologist) determines the aggressiveness of a brain tumor based on an analysis of a piece of tumor tissue obtained through surgery. Such an operation is not without risk. In addition, the patient has to wait for the operation and in the meantime he or she is uncertain about the continuation of the treatment.
Van der Voort's research shows that artificial intelligence is able to predict tumor aggressiveness based on MRI scans alone. In some cases, the model developed by Van der Voort predicted this even better than doctors looking at MRI scans alone. His expectation is that the use of artificial intelligence could reduce the number of biopsies in the future. In addition, a patient can already have more certainty about the course of the treatment immediately after the MRI scan, instead of having to wait for weeks.
A tool that Van der Voort developed for his research is now being used in research practice. Currently, scans of patients can only be compared if you have the same type of contrast images. But that is not how the database is set up in the clinic at the moment. Until the tool was available, it was therefore necessary to manually sort out which of the 12 MRI images per patient has which 'contrast'. A time-consuming task.
The new tool sorts the entire database, making it easier for the researcher to select. Analyses show that the tool is more than 98 percent accurate. Erasmus MC is not the only one using this tool now: many medical centers are using it. It is open source and therefore available to all scientists. Not only does this encourage wider use, it also helps to further improve the tool.
There are more examples like the one outlined above. Yet the application of AI technology is also often still far from clinical practice. One of the barriers to further development towards implementation is understanding and trusting this technology. A medical content guide can help healthcare providers assess the quality of the AI solution and help developers design and implement quality, trustworthy AI. With this, a non-technically trained healthcare professional can still make good judgments about how to use the AI application.
Since 2020, a broad group of experts and stakeholders have been working on such a guideline. The Guideline for high-quality diagnostic and prognostic applications of AI in healthcare has been developed by and for the field and is available to all.
Another condition for the further development of AI for healthcare is that AI applications can also make use of good, reliable data. Currently, data in healthcare is still mostly fragmented: data is organized by study and by application. At the moment Nictiz, the NL AI Coalition and the Ministry of Health, Welfare and Sport are mapping the requirements to improve the availability of data for AI in healthcare.
AI plays an increasingly important role in various scientific programs of the Medical Delta partnership. For example, in the Medical Delta AI for Computational Life Sciences program launched last year, or in the various Imaging programs to which Van der Voort's research also contributes. The potential of AI is clearly observable in the various studies. At the same time, it requires different disciplines to look across borders and work well together.
These developments show how complex the ecosystem around a broad application of AI in healthcare is. Technologically, an enormous amount is already possible and those possibilities are only increasing. But to really exploit the potential of AI for healthcare, good interdisciplinary cooperation between healthcare providers, data scientists, researchers, patient groups and governments is essential.
This article appeared in ICT&Health