Pathology Artificial Intelligence (AI) as a Medical Device

Share this Post:

September 10, 2018 – A pathology AI system with an intended use for clinical diagnostics (Dx), prognostics (Px) or companion diagnostics (CDx) can be commercialized as a medical device.

Commercialization as a medical device requires:

  1. The pathology AI system to be manufactured as a medical device,
  2. Clinical studies to be conducted,
  3. The appropriate regulatory approvals are obtained, and
  4. A (global) distribution channel is established.

With the adoption of digital pathology, clinical laboratories are using slide scanners from different manufacturers. As those slide scanners are a major investment in digital pathology, any pathology AI system needs to be slide scanner-agnostic, and be compatible with slide scanners commonly found in clinical laboratories. No organization wants to purchase a slide scanner just to be able to use a new pathology AI system. As in radiology, pathology now needs to be able to plug-and-play with different modalities and devices to enable widespread adoption. Pathology AI systems need to be their own medical devices with a clear and easy path to work with commodity slide scanners. A pathology AI system could be a medical device by itself that delivers reports for pathologists and oncologists, but often clinical laboratories would like to see an integration with their existing LIS/LIMS, which can be done fairly easily with most LIS/LIMS systems.

A pathology AI system can be considered software-as-a-medical-device (SAMD), using commercial-off-the-shelf (COTS) hardware. To manufacture medical device software, a company must have a software engineering team and a QA team with an established quality management system (QMS) based on ISO 13485, ISO 14971, IEC 62304 and FDA QSR.

Pathology AI Dx and Px products can be developed by “just” technology companies; no In-vitro Diagnostics (IVD) manufacturer needs to be involved when common H&E staining or already approved IHC assays are used. The key is to get access to patient outcome data for the development and the clinical trials, which typically implies partnerships with academic medical centers. The entire burden on the development and commercialization is on the technology company.

Conversely, pathology AI CDx products are driven by pharmaceutical companies developing therapeutics that need a test for patient selection as part of a drug label, thereby requiring a continuous path from their clinical trials to a CDx. The pharmaceutical companies typically provide the data for the development and conduct the clinical trials required for the regulatory approvals as part of a therapeutic/diagnostic co-development. However, CDx products typically also involve a partnership with IVD manufacturers who provide the wet assays and have the appropriate distribution channels.

Building a pathology AI system from scratch is quite an endeavor, as a lot of expertise and experience needs to go into its development. But often, as seen in the case of a CDx, some existing technology has shown clinical utility in an exploratory setting and “just” needs to be manufactured as a medical device with a simplified user interface and locked-down algorithms. At this moment, it becomes important to control the software, relying on in-house software development, as opposed to using third-party software. Using a validated prototype (e.g. LDT under CLIA/CAP) as the basis for a medical device development simplifies and de-risks the development lifecycle considerably, which at that point can become a simple waterfall model. Being able to manufacture a medical device that provides the same data as its prototype during a clinical trial opens the door for a product development parallel to the clinical trials, such that a simple bridge study can be used to allow the medical device to use the study data from the clinical trial for its regulatory approval.

Once a pathology AI system has been manufactured as a medical device it provides a platform to which new applications for Dx, Px, or CDx can be added very quickly (basically only the algorithms); ultimately building a comprehensive pathologist cockpit.

To run clinical trials and get the appropriate regulatory approvals for a medical device, a company must have the expertise to conduct clinical trials and to obtain regulatory approvals for IVDs globally, ideally having established standard operating procedures (SOP) for good clinical practices (GCP) for clinical trials.

Technology companies need to build their own global distribution channels into the clinical laboratories, or work with partners, like IVD manufacturers, that already have the appropriate distribution channels.

The commercialization of a medical device is very costly. The US anatomic pathology market for a medical device based on the current reimbursement model is relatively small; we estimate it to about $11 million, which then again is shared by multiple manufacturers. The problem is that the anatomic pathology market is very segmented by subspecialties, which correspond to different tissue types (e.g. breast). Each tissue type requires different tests (e.g. H&E diagnosis, IHC Her2, ER, PR), that typically correspond to different stains. This creates a myriad of tissue–stain–clinical outcome-specific tests that each must become a separate medical device. The business case to commercialize a pathology AI system as a medical device is a challenge.

Interestingly, all existing tissue image analysis medical devices for pathology (computer assisted IHC Her2, ER, PR, …) have been developed by digital pathology manufacturers, but with a completely different business case in mind, to be able to market their digital pathology equipment to the clinical market. No real pathology AI medical device exist today.

Are you (planning to) building a Pathology AI system as a medical device? What is its intended use: Diagnostics (Dx), Prognostics (Px) or Companion Diagnostics (CDx)? How do you integrate your system with the existing Digital Pathology infrastructure? What global distribution channel are you going to use? What is your business case?

Flagship has been developing our own pathology AI system over the last eight years to solve the most challenging real-world tissue image analysis problems across the entire pharmaceutical industry. We have everything in place to manufacture our pathology AI system as a medical device (e.g. QMS for medical device software), to conduct the clinical studies (e.g. SOPs for GCP for clinical trials) and to obtain the appropriate regulatory approvals (e.g. FDA PMA), thereby providing our pharmaceutical customers with a continuous path from their clinical trials to a CDx. With our technology and capabilities, and the right business opportunity or strategic partnership, we are well-positioned to commercialize our pathology AI system as a medical device across the full spectrum of intended uses including Dx and Px.

If you want to learn more about the different aspects about pathology AI, view our series of brief lectures for a broad non-technical audience on LinkedIn.

Holger Lange, PhD
Chief Technology Officer
Flagship Biosciences