Siemens Healthineers’ Mario Orsini discusses the impact of artificial intelligence on clinical medicine, challenges around adoption of AI solutions, and measuring ROI.

What initially sparked your interest in medicine and subsequently, digital health and data science?

I wanted to be a doctor from a very young age. I have always been fascinated by science and motivated to use my abilities for societal good.

As for digital health and data science, this interest came much later and is more complicated. In my clinical career, I practiced in many different environments. There has been a relentless increase in imaging exam volumes and imaging exam complexity over the years. A Head CT used to be 18 images on a single sheet of film and is now several hundred images. An Abdominal CT might have been 80-90 images on a few sheets of film, but we now have studies like CT Urogram that can be several thousand images. With this increased technology and complexity has come tremendous improvements in diagnostic capabilities, but it is difficult to consume that much data.

At my former practice, I was part of the management team trying to improve efficiencies so our radiologists could provide quality care in the face of this data deluge. I worked the night shift as an ER Radiologist my final few years of practice. Many of the systemic strains of radiology are amplified in the ER setting, even more so overnight. After earning an MBA from IMD Business School in 2019, I decided to help solve these and other healthcare challenges. When I saw the opportunity to join Siemens Healthineers, I jumped. In my current role, I can positively impact the entire clinical value chain, potentially helping clinicians care for billions of people worldwide.

Your primary focus is on applying AI to improve imaging data and to integrate clinical information. In which clinical domains are you seeing the greatest impact?

These are components but, to be more precise, my primary focus is developing AI solutions that improve healthcare outcomes.

We see time, effort, and financial resources directed at medical imaging, which is one crucial piece of the puzzle. However, the ultimate goal must be to integrate clinically relevant data from the entire patient journey to understand disease processes better, intervene earlier, and develop robust predictive analytics. I want AI to assist clinicians in determining the best treatment options for a specific patient, not a generalized population; this is precision medicine.

What are the main challenges you have observed in the adoption of AI solutions, and how can they be mitigated?

Medicine is a risk-averse industry, and I believe this is good. However, this inherent risk aversion, combined with misperceptions of what AI can and cannot do, has contributed to the slow adoption rate.

Another challenge is that radiologists are over-tasked. Workflow changes must be intuitive and optimally efficient; anything less is met with skepticism, even if the promise is long-term efficiency gains.

There is also a lack of comprehensive solutions. It is difficult for radiologists to interact with three or four different user interfaces to cover a meaningful range of clinical applications. In my opinion, companies that offer a broader range of solutions, either developed by them or via a multi-vendor but unified platform, will be most successful.

Finally, there is the question of cost and payment. AI solutions are typically Software as a Service, which is good on the one hand because it is scalable, but it requires a shift of thinking. As more healthcare institutions use and prove the value of AI solutions, adoption will increase, and we will “Cross the Chasm,” so to speak.

A major consideration for healthcare providers looking at AI solutions is around ROI. In your experience, is this strictly about impacting the bottom line, or are there other outcomes which represent value?

I think there are many ways to view ROI when it comes to AI solutions.

Time savings for the radiologist can translate to improved job satisfaction and more time to spend on clinically challenging cases. Reducing mundane tasks such as lung nodule detection and measurement can reduce fatigue, affording the radiologist more time to think, and decreasing inter-reader variability.

A second-look algorithm can reduce missed findings, which helps our patients and reduces radiologist stress and legal exposure. Algorithms can also quantify information that would not typically be possible such as coronary artery calcium and bone density. Capturing this information into something actionable can positively impact long-term health outcomes through earlier treatment intervention.

It is important to me that we are relentless in our focus on clinical outcomes. I want to create environments where radiologists can apply their medical expertise to improve patient care maximally.

So, my view is that focusing only on radiologist productivity metrics would be incredibly myopic. We can transform healthcare through the intelligent development of AI solutions; this is particularly important in a world of increasing populations, advancing age, and a chronic undersupply of healthcare professionals. It may be even more crucial in areas of the world that are truly medically underserved.