HIStalk Interviews Anjum Ahmed, MBBS, Chief Medical Officer, Agfa HealthCare

Anjum Ahmed, MBBS, MBA, MIS is chief medical officer, clinical safety officer, and global director of AI and innovation of Agfa HealthCare of Mortsel, Belgium.

Tell me about yourself and the company.

Agfa HealthCare is a global solution provider of imaging IT solutions. It is part of the Agfa-Gevaert Group, which has been in the industry for over 150 years. Our prime focus over the last year has been transitioning our customers from the traditional PACS approach towards enterprise imaging. That strategy of consolidating imaging service lines has evolved across the industry. We launched our flagship platform for enterprise imaging few years ago. We were first in the industry to build a platform from the ground up. The company has R&D centers across the globe in Canada, Belgium, Austria, and China.

My role with the company is global chief medical officer. I’m also head of the portfolio for innovation, for artificial intelligence, and in making sure that we are successful in rolling out these new innovations to our strategic customers.

What maturity level does enterprise imaging have in the US, and what benefits does it offer?

The rest of the world is looking at what the US is doing. If you look back at how the consolidation of electronic health records started in the US — that transition from paper to digital and from digital to electronic health record – it made CIOs and hospital systems across the US realize that now is the opportunity to think about imaging as a service line. How the consolidation that they did with electronic health records could transform the care that they are delivering to their communities. That’s one of the reasons I would say that the US as a region was one of the early adopters of the enterprise imaging strategy. It made sense because they realized gains from electronic health record consolidation. 

The question was, why not when it comes to imaging? There were multiple approaches that the health centers and health systems in the US took. The initial approach was with vendor-neutral archives that could be a starting point for consolidating imaging from service lines that go beyond traditional radiology and cardiology into oncology, point-of-care ultrasound imaging, mammography, and breast imaging use cases. That was one aspect.

Here in North America, including where I’m based in Canada, there was also another aspect, which was that we have consolidated the imaging records, but our health systems or hospitals are on multiple PACS technologies. How do we go about bringing a uniform viewing layer? That’s where the universal viewing component for enterprise imaging also came about. We have the VNA and we have consolidated the archive of images, but how do you use or visualize that data? Besides VNA, universal viewer also became an important component for not only beginning the journey for consolidation, but also the visual layer in terms of consolidation of imaging and how you view those images.

The US is pretty mature, I would say, in comparison to what’s happening in the rest of the globe, where the enterprise imaging strategy may initially be focused on bringing the new technology into radiology or cardiology, point-of-care ultrasound imaging, and GI endoscopy. Multimedia images related to surgical procedures is also something that is being spoken about.

The next wave in enterprise imaging will be led by digital pathology. If you think about holistic clinical care in terms of oncology, and along with a lot of talk about precision health and precision medicine, bringing in histopathology, digital pathology data, and seamless collaboration with other imaging records is something that we are already hearing about in the US as a region when it comes to enterprise imaging adoption.

A recent KLAS report noted that Europe is leading the adoption of digital pathology. What are the opportunities and challenges of rolling it out in the US?

I have noticed that as well. We saw the rollout of digital pathology for certain use cases in Europe in 2015 or 2016. Obviously there are regulatory challenges when we compare North America to what was done in Europe, but the biggest challenge is that there are no standards that have been adopted for digital pathology, unlike what we had in radiology with DICOM imaging and all those standards.

The other challenge with pathology was the use of scanners to scan the glass slides and convert those glass slides into digital data. That is unlike radiology imaging, where you have modalities that are generating digital data. In pathology, you still use microscopes that are being read manually. Every scanner vendor generates proprietary formats for data ingestion. That was a challenge with some of these labs that were transitioning from glass slides to digital. Should they stick to one scanner vendor, or if they have multiple clinical use cases, they might be in a multiple scanner environment, which means multiple storage solutions for each of those scanners. That is where they started exploring whether a data management strategy would be an entry point into digital pathology with enterprise imaging. That is something that UK also took when these new RFPs or tenders were coming out over the last couple of years.

Data management became a very relevant ask. Rolling out enterprise imaging outside radiology, how would you manage data from these multiple scanners that generate proprietary data in the absence of DICOM standards? That challenge had to be addressed. VNA is vendor-neutral, so there must be a strategic approach in how that data could be managed with digital pathology acquisition.

Besides the data management aspect, there is also the departmental workflow when you go digital with pathology, similar to radiology and cardiology workflows. Pathology has its own requirement in terms of the departmental model. The question was, how are we going to develop these modules within enterprise imaging similar to radiology in the pathology workflow?

The third aspect is the visual layer. Should it be a universal viewing platform? Should it be a radiology desktop kind of a solution for pathology?

This is how the industry evolved. We have seen recently in our regulatory clearances that have been coming out in the US certain use cases to consider for digital pathology. That’s one of the reasons I’m saying that there are lots of lessons learned in how Europe started with their adoption of digital pathology based on certain clinical use cases, data management acquisition, and the visualization layer. Those are the three components that will help drive the adoption of enterprise imaging further into digital pathology.

EHRs made it possible for clinicians to work from anywhere. How is the profession of radiology changing as their work becomes digital and enterprise imaging becomes more prevalent?

We witnessed that during the pandemic. Enterprise imaging is a modular platform. As part of that modular platform, we have the image exchange portfolio. Besides image exchange, there is the federated image exchange network, so that you don’t need to physically move the data. Our customers started asking us when they started working from home how they could access this desktop on their home environment with the all the tools they require. Little did the customers realize that when they invested in that enterprise imaging platform, which brought them image exchange and collaboration capabilities, it took just a click of a button to enable those collaborative workflows.

When I talk about collaboration, I talk about real-time collaboration. One benefit of building that enterprise imagine platform strategy is that you’re not sending data across to external systems, where you could be exposed to someone interfering or accessing that information. Because you have created this secure system with enterprise imaging on a single platform, you are enabling access to your users if they’re at home to leverage the same capabilities with the same viewing platform on a thin client. We have Xero Universal Viewer, which is cleared for diagnostic reading. It has built-in capabilities and real-time chat collaboration similar to WhatsApp. Within this tool, you can see your colleagues who are online, you can share interesting cases with them, and you can share securely, including with other users who may not be part of your enterprise. It generates a secure image exchange kind of a workflow.

Another thing I spoke about was the federated image exchange. Federated image exchange means that you do not need to push and pull images from one storage to another archive. We could  set up Xero universal nodes so that users are able to view our stream images from an external, non-Agfa PACS, for example. That’s one of the benefits that we have seen our customers appreciating — they were able to build these networks of communication and collaboration not only within their Agfa enterprise imaging environment, but also outside Agfa’s enterprise imaging portfolio, so that they can view those images on a common viewing platform.

The clinical community, radiologists in this case, have realized that these tools are actually much more helping and facilitating in terms of how they view cases and how can they be more productive if they are not on premise. From an IT perspective, we have gone live at certain hospitals in the US during the peak of the pandemic in a rollout of the technology that was also managed remotely. That’s where we saw a lot of collaboration, not only from a clinical perspective, but between the IT segment of the community as well with our customers, where our IT and project management got involved with the customer IT to remotely deploy some of these solutions.

The hype a couple of years ago was that AI would replace radiologists, which has moderated into thinking about how AI can support radiologists. What is the most promising use of AI in imaging to improve patient outcomes?

We started working on AI in 2015 and 2016, when there was all this discussion about whether AI would be of any use in medical imaging. We partnered with some early adopters and explored certain clinical use cases. My first question to our customers was, what problem are you trying to solve? Let’s park AI on the side and first identify those clinical challenges that your healthcare organization is trying to address. Then we can decide whether it is AI or whether it is deep learning, machine learning, automation, or pixel intelligence. What kind of technology could we apply in helping you address those clinical challenges?

We identified certain use cases associated with chronic diseases such cancer care, where we thought — and customers agreed with us — that automation could perhaps help to them in early disease detection or even automating some of the manual tasks that radiologists are performing in some of these clinical applications. When we announced our AI strategy, we called it augmented intelligence, the intersection of machine learning and advanced applications where clinical knowledge and medical data converge on a common platform. AI replaces clinical knowledge or clinicians, while augmented intelligence works with the clinical audience and facilitates their work.

We worked with our clinical users and early adopters to say, let’s define KPIs and see what outcomes we are able to improve. At Agfa, we want to focus on the workflow side of the things. We are an enterprise imaging solution provider and our customers would expect us to use AI data from several AI applications that are being developed in the market and leverage that data to do something. Some of those companies that were creating hype around replacing physicians with AI have disappeared from the market because the claims that they were making were not addressed in use cases.

There are 100-plus AI startups out there. We decided to focus on workflow, because in developing our own AI applications, we realized that a lot more needs done than just reading pixels and images. An AI algorithm developer has developed something very nice, so how can we as Agfa utilize it? We developed this framework for AI that we call RUBEE, whose goal is to embed clinical intelligence into the user’s workflow from five perspectives.

Number one is that AI generates a lot of data. How do you utilize that data and how do you visualize it? How do you show it to the clinical user? Instead of having a radiologist or a clinician use multiple applications or viewers, we have embedded those visual findings from AI into the enterprise imaging portfolio, whether it is the desktop or whether it is the Xero Universal Viewer that I spoke about.

The second and third aspects are the workflow orchestration and triage. With workflow orchestration, AI generates abnormality findings, abnormality scores, measurements, or some other aspects. With the RUBEE engine, we are able to orchestrate certain workflows and automate certain tasks that radiologists are spending time today doing.

When we released our AI package to one of our first early customers, they said that reading a particular CT scan went from taking 15 minutes to being finished in seven to eight minutes. With RUBEE, all the tasks that they were doing have been automated. They know that at the top of the list, these are the abnormal cases that they need to start their work with, these are the measurements that the AI algorithm has generated. With RUBEE, they can see where those specific cases are. We can distribute some of those cases to certain groups of radiologists who are concerned about that specific clinical scenario. That’s where the visualization, workflow orchestration, and triage help achieve certain productivity.

The fourth aspect is automation of all hanging protocols. Radiologists spend a lot of time — when they are reading certain exams, currents, priors, and certain cases — going back and looking at certain prior scans in comparison with what they are seeing now. RUBEE, based on AI findings, automates certain hanging protocols so that radiologists do not need to find a relevant prior scan for this particular patient. Early adopters told us that this is useful and they have appreciated the time savings.

The fifth element of our RUBEE strategy is, how do you communicate reports and results? AI is generating visual findings and you are orchestrating and triaging. How can you save me some time in generating reports? That varies in North America versus UK and Europe, where the use cases are different. In the North American region, we have seen customers are using specific reporting solutions, so we can provide a feed from report that is generated by AI to the reporting engine. In Europe, where customers are using the built-in module for reporting with enterprise imaging, we have created structured reporting within radiology, so that we can extract certain drop-down menus within the report itself. It becomes then easy for radiologists to do a one-click signoff to agree with the report or disagree with the report and generate their own findings.