Unlocking the true potential of digital pathology

The promises of digital pathology are myriad – for starters, the reduction and potential elimination of the physical footprint occupied by dusty glass slides, soon to become obsolete. There is the ability of clinical labs to have a specialized pathologist in, let’s say, San Diego, diagnose tissue from a patient in New Jersey. Going even further, you could supplement the efficiency of that pathologist via a GenAI model which not only diagnoses the sample, but factors in the patient’s entire medical history and presents a personalized treatment plan.

A key takeaway from this year’s HIMSS (Healthcare Information and Management Systems Society) conference in Orlando, FL, was that the above vision should sound something like the concept of an UberEats delivery to someone in the 90s listening to the crackle of dial-up connection. We are only beginning to scratch the surface on this journey.

The message was aptly delivered and visualized by Dr. Jen McKay, Senior Clinical Specialist at Google, during the panel discussion I participated in (Transforming Public Sector Healthcare). She showed a chart with an upward sloping line, illustrating that on the path from Efficiency (automating admin tasks) to Effectiveness (augmenting provider decisions) to Innovation (actually making healthcare more personalized and affordable), we are just barely approaching Efficiency. When she presented this, there were excited whispers in the room and several heads nodding vigorously in acknowledgement.

I was asked what Iron Mountain had seen in the market that would inform the direction that GenAI developers should take. It was a complicated question, because in fact, we are currently in the process of figuring this out. I mentioned an example of a Digital Pathology customer who requested (and paid for) us to scan each of their glass slides on two separate machines, each of which produces images in fundamentally different ways. Machine number one uses two dimensional scanning, most common with pathology scanners today, which produces images in a single layer, via a lens that adjusts in height to accommodate various focal points as it works across the sample. Machine number two deploys volumetric scanning, capturing focal points throughout the depth of the sample before reassembling them to produce an optimal image. Why would a customer want both images? To run analytics on both to then obtain a more robust dataset, and to make optimal investment decisions before pursuing a larger project.

Iron Mountain never planned for this scenario as a solution or as a use case; it came up along the way, and we reacted accordingly to meet the requirements. More importantly, we learned about a unique need that had business value, which in turn poses important questions:

• What is this customer learning from running their AI on both images? Likely this could be for the validation of a desired end result; if the AI is meant to aid in the detection of a specific disease, and achieves this result with equal accuracy and speed in both cases, which is important information from a commercial point of view.
• Are those learnings only achievable in this way, or could GenAI models be trained to circumvent this need? The two-scanner approach is expensive and inefficient, versus being able to achieve the same end result of robust data with GenAI.
• What is the interplay between the hardware being used and the models being deployed on the output of that hardware? Are there limitations that should be addressed? There are scanners which come with their own AI models, which may or may not suit the needs of the customer.

Iron Mountain is able to formulate and potentially answer these questions due to our role in this space as a broker and an integrator, as a vendor providing a truly end-to-end solution. On an existing foundation of operational and technical expertise and physical and digital storage services, we have added an ecosystem of partners who provide scanning hardware, viewing software, and AI analytics. While we do not develop the software ourselves, we make integrations seamless, allowing our customers to opt for the tools that meet their business needs while maintaining the ease and efficiency of having a single provider.

Case in point: Customers are often surprised when they hear that our storage solution, Iron Cloud, does not charge egress fees to access their data. While this is compelling on its own, we are able to take it a step further and provide this option with a solution that includes physical pick up of slides from a lab, delivery of those slides to a storage facility that includes scanning upon intake, and upload of the corresponding images to Iron Cloud. Images can then be routed to specific storage tiers based on their use case, with higher value images placed in hot storage for immediate access and longer-term legacy data routed to lower-priced cold storage.

On the flip side of an “end-to-end solution” is flexibility and vendor-neutrality - if a customer’s existing storage service must be maintained, images can be delivered via data agent to an SFTP server and integrated with any downstream system. Likewise, our viewing platform runs cloud native, we can architect a solution or integrate with other viewers if that is not an option. We have carefully constructed our standard offering for a reason, but remain agile as we meet requirements across a broad and global customer base.

From this truly unique vantage point, and as we continue to refine our solution, we are able to understand the needs and challenges of all players in the market, and ask questions such as those noted above in the “two scanner” scenario. This will have two sets of impacts: the way we choose to develop our solution in terms of capabilities and features, and the influence we have on our partner vendors.

As an example of the first, we understand the business needs of our customers to then decide which workflows to incorporate into our solution. Referrals and second opinions, for instance, are a key requirement for many pathology departments. It is of critical importance, then, that we choose to partner with an image viewing platform that can be enabled with such workflows and integrated with our end-to-end solution in terms of architecture, hardware, and other services such as logistics and storage.

The second point, regarding our ability to influence the partners we work with, speaks to our position in the bi-directional flow between customers and the other vendors we choose to partner with. Labs and hospital systems are sitting on a potential goldmine of pathology assets for researchers and AI developers. As a guardian of those assets, we are positioned to capitalize on this value for our customers by partnering with marketplace facilitators that connect buyers and sellers. A key challenge we have discovered is that many customers, having grown inorganically to massive scale, face tremendous difficulty in even being able to identify which assets they have and their marketable characteristics. In relaying the nuances of these challenges to our partner, they are able to refine their approach and accommodate the situation - based on the customer segment (CRO, AMC, etc.), we can work together to identify the likely “low-hanging fruit” to begin on the monetization journey, rather than requiring an analysis of the organization’s entire data set.

This flow between players in the market – hospitals providing images to AI developers, clinical labs integrating their information systems with image viewing platforms, marketplace platforms running AI on the slides and blocks on-hand with hospital systems – illustrates the challenges and opportunities ahead of us, and the unique position Iron Mountain sits in. As a broker and integrator offering an end-to-end solution, we are learning, adjusting, and accommodating, working towards Efficiency with everyone else. Let’s enjoy the dial-up days while we’re here.