Skip to main content

At the Bayer Press Day last March, Guido Gebhardt spoke with Michael McDermott, Senior Director and Head of Research and Development Innovation at Bayer, to know more about the company’s innovation strategy and its plans to harness the power of digital tools. 

Bayer has been very active in the field of digital radiology solutions for some time now. How important will AI be for you in the future? 

While AI is an important part of our future, within our radiology business, we have four main pillars, which represent the areas or product paradigms that we innovate in and around. First and foremost is contrast media –  where we follow the classical pharmaceutical pathway.

Another area is medical devices, primarily with our injection systems, which help to deliver the contrast media precisely. Third the digital space where Calantic Digital Solutions combined with our acquisition of Blackford Analysis have propelled us to the forefront of AI integration in diagnostic imaging. Another focus in digital removes around the use of AI together with contrast media to optimize the administration of contrast agents and/ or subsequent image quality.

One of our newest areas of innovation is around data. Medical imaging data holds an incredible amount of power, and we’re only just starting to unlock that power now. That’s also a key area of focus for us. Data is something everybody wants to have, but the big question is how can we effectively utilize it?

(left to right) Michael McDermott speaking with Guido Gebhardt

Where do you get your data from?

There are different pathways, and it depends on which type of data and what we want to use it for. Our physical injection systems have a very rich ecosystem of data, and together with the consent of our customers we can use that data for example to know precisely how much contrast media is being administered.

This can help with tasks like optimizing contrast media use, and especially in times where contrast media shortage was all over the news, this proved to be very valuable for many of the clinics we support.

But when you talk about imaging data, there are numerous sources, be it via clinical trials, cooperations with leading healthcare institutions, or industry partners. Most importantly, the sources and data types depend on what we plan to do with it. If we’re developing a new digital product, we need to make sure that the data we’re capturing will be accepted by health authorities to eventually receive market authorization for that product.

We are also taking steps to support not just our own data needs, but we are in a unique position to blaze the trail for other industry partners. Our goal is to enable the acceleration of new AI products and also new medicines to more patients. With the launch of our new Clinical Imaging Core Lab Services at ECR 2024, we have created a framework for Bayer and our healthcare industry partners to leverage structured, AI-ready, and easily accessible sources of reusable, sustainable data.

Data is area specific, but algorithms have to be trained outside of this area. What do you think?

If they aren’t, they just don’t work. That is one of the challenging things about algorithms. For each area of the body you are interested in regarding AI-supported diagnosis, for each imaging modality used to acquire images, you require complete new datasets for training. Added complexity comes when adhering to the expectations of health authorities, which we also see as diverging. What the FDA is expecting in terms of data populations and variation may not 100% overlap with what is expected here in Europe. 

The same goes with China. Often times, authorities want data with their own patient populations. They also require data produced with different scanner manufacturers and including a wide variety of diseases. It is something that has to be tackled to bring a product to the market. We’re also hoping that regulators will come along with us on this journey to bring these digital products to market in different regions and countries.

What about pattern recognition using AI in the development of pharmaceuticals?

One of the areas we are looking at here includes radiomics, which involves extracting quantitative features from medical images. The idea is that you can also do this with thousands and thousands of medical images, and leverage pattern recognition to infer various conclusions about the patient. For example, combining imaging data with clinical follow-up can allow us to develop models to identify patients who will respond best to certain treatments based on unique characteristics found in the images. In practice, this would allow us to look at a future patient’s images andhelp decide which treatment they will most effectively respond to.

That also helps in drug development, for example to know, in a clinical trial, which molecules will be predicted to produce the most positive response in a given patient population.

This technique can also be used not just for therapeutic intervention, but also for diagnostic support. For example, based on a given image pattern surrounding a lesion we can infer information about the micro environment of the tumor, and ultimately more accurately predict staging and prognosis for example in hepatocellular carcinoma. So it’s quite an interesting technology, but it’s very early and also requires a ton of data.

Michael McDermott, Senior Director and Head of Research and Development Innovation at Bayer

What is special for Bayer in radiology?

I would start with what kind of challenges does radiology face. We see three main challenges that determine our area of interest for how we can support our main customers, the radiology departments.

If this weren’t enough, as the demand medical images accelerates with chronic diseases on the rise, they must deal with increasing workload burden, qualified staff shortages, high turnover rates, and high rates of burnout. The ground burns beneath their feet as they’re trying to solve all of these challenges. We have the utmost respect for how they manage this, and we are working every day to support them in this.

First, our customers are fighting working hard to defend and grow their position in the healthcare decision making landscape, to sit more prominently at the center of medical decision teams as opposed to becoming just merely a service provider of images. At the same time, they’re also having to manage being at the tip of the spear for integration of AI in healthcare, and figuring out how best to embed this safely and effectively within their clinical routines.

We see digital transformation is as really the key. That is, supporting our customers with digital solutions to reduce the manual labor that’s necessary for staff, help streamline workflows, improve efficiency, and save time. In addition, we aim to support by adding even more robust insights in the diagnosis to be able to make radiology departments even more valuable to their referrers and the healthcare landscape. And I think that we, as a company with innovation across the four pillars I mentioned, are well suited to deliver on that.

If radiologists start not using AI, is there a risk that radiology departments will be taken over by other specialists that use AI?

We see this competition as well, and we’re very aware of it. You do see more and more imaging being performed by other specialists, who are happy to grab onto AI because they may not have the legacy knowledge that a radiologist has. And so it helps them do their job more effectively. 

I think still the biggest burden challenge to this will be the regulatory hurdles. The adoption of AI in clinical routine at scale is really a challenge when you think about the barriers that companies have to face. For small startups to be able to achieve market authorization can be quite challenging. A relevant proportion of AI in radiology is currently still homegrown in clinics, not produced by companies who are selling products, primarily as the barrier to entry is so high. That’s starting to change and the regulators are working more and more together with industry to facilitate these value-adding AI solutions to enter the market helping providers and benefiting patients.

Why do you think AI is not as successful as it should be?

I think about it in terms of a traditional hype cycle. In engineering and technology, we talk a lot about hype cycles. And it’s this classic shape where you have a fast rise of expectations with new technology, a peak of excitement, and then a steady decline as the high expectations fail to be met in unrealistic timeframes. At the bottom of that bell curve you reach what is called the trough of disillusionment as disappointment reaches its maximum, but shortly after that the curve rises again as the technology matures and stabilizes with broader adoption.

This is true with nearly all technologies, and AI in healthcare is no different. In the early years excitement was so high that some even claimed within a few years that radiology departments would cease to exist. We then quickly realized that it is hard to implement AI broadly and at scale in healthcare. However, I believe we are on the path to maturity and stabilization, as industry and healthcare providers are committed to improving the scalability and adoption to support radiologists, not to replace them.