This article is a synthesis of an R&D discussion between Moichor team members:
Director of Pathology, Kyle Webb, DVM, DACVP (Clinical)
Director of Engineering, Van Duesterberg
Principal Computer Vision Scientist, Matthew Guay
CTO, Thanh Le
When we discuss hematology instruments, we’re largely referring to microscopes and laser flow cytometers. To date, when it comes to getting deeply analytical CBC results, there is no equivalent to a pathologist’s trained eye looking through a microscope at a blood smear.
If processing and pathologist time were not a limitation, then high throughput laser flow cytometry likely wouldn’t be ubiquitous. In the absence of a reliable and efficient manual solution, engineers in the mid 1990s figured out a way to apply high through-put laser flow cytometry. And since then, that has been the standard — built with throughput in mind and all analysis encapsulated within the device.
As we build the next-generation platform for diagnostics, it is important to explore how embracing a blend of hardware and cloud-enabled AI software solutions enables not only a deeper level of analysis and insight, but also higher throughput for the CBC.
The step from pathologists looking through microscopes to flow cytometry came with limitations. Those limitations included moving away from images to data represented as a dot plot.
In the dot plot format, clinicians are not able to calibrate the reported results with a visual confirmation without manually examining the blood smear, a time consuming and redundant process.
Additionally, software intended to address throughput provides little access to deeper analysis and intuitive controls. The processing and analysis are all encapsulated within the device itself — like an island. As a result, it’s difficult to transport results from this metaphorical island into an auxiliary system for further analysis.
One way to illustrate this point is to imagine a digital Casio watch from the 90s like this one. This watch has software built to tell time and calculate basic mathematical problems. It cannot connect to the internet and thus, to change the software to do something else, like play music when you press its tiny buttons, would involve a hacker to open up the back of the watch and manually reprogram the device.
In contrast, a smart watch today that is connected to the internet can update operating systems, download new software directly in the form of apps, and process information captured via the sensor, camera, and touch screen. It can send information back to the cloud allowing you to do many things beyond what was ever possible with the Casio watch.
To date, there remain few examples of a cloud-based approach to diagnostic software in the veterinary space. But this is changing for the reasons we explore in this previous R&D chat on The Advancements That Made Moichor's Technology Possible.
In the past, software only existed on devices — for example your computer, your thermostat, your phone, your car, or your digital watch calculator. This approach is important and can be very powerful. But it comes with some very fundamental constraints.
There are certain things that the software that lives on your device is good at, for example, moving optical components and focusing an electronic microscope, or controlling sensors that regulate the electrical system or combustion system in your car. Software that is built into hardware excels at tasks that are unlikely to change drastically.
But there are important limitations particularly when it comes to computational power and adapting to new information.
For example, while flow cytometry is good at counting intact and degraded cells in a high throughput way, it misses many of the key benefits of seeing the cells. One example of the limitations of this approach is that it can’t determine if there is a parasite in the cell or if there are atypical granules in the cell.
In flow cytometry-based solutions, information is processed in a closed system: the software component enables the automated counting of cells by passing a laser through the cells. The counting takes place at a localized software level on the device. Then, at a software level, there is a menu that has been calibrated with a finite list of species. Results are output as a table or dot plot that can be interrogated in a very limited and non-intuitive fashion. If further analysis is required, the process is done manually by a pathologist.
We envision a process that provides deeper analysis for every sample without affecting throughput. In this platform, information is exchanged to and from a software system in the computational cloud so that the hardware can access a large knowledge base of data and analyses — creating a positive feedback loop. Effectively, this is what makes a platform ‘smart’, being able to both provide information to the cloud and receive information back to the device that enables evolution of its operational approach.
As a result, in a high throughput image-based approach, a slide can be processed with the data instantly analyzed in the cloud. Depending on the findings of the AI, it can be flagged for further analysis by a pathologist or if not, it can be annotated by the AI. The pathologist can view the fields flagged by the AI and conduct rapid review facilitated by the AI viewer. There’s no need for the pathologist to look through a microscope because images have already been captured and uploaded to the cloud, and precise preliminary analyses were already performed.
By using an image based approach, high throughput and deeper analysis are achievable at the same time by using the same image data a pathologist would ultimately refer to. What is more, by isolating and organizing image data in a way that helps the pathologist to review samples, the technology eliminates the many different frictions that were part of the traditional approach that led to long processing times.
By modernizing the methods of building the tools we use for veterinary medicine, clinicians can expect more actionable insights. One of the simple benefits of a cloud-based solution is that we are able to connect to a network of pathologists for more advanced screening and seamlessly incorporate a path review.
A clear benefit of a cloud-based approach is that it enables your devices to have access to supercomputer computational capabilities without having to actually have a supercomputer built into the device itself.
A cloud-based approach acts as a bridge between a device and a centralized database that is able to continuously be adapted and improved over time. By embracing this approach, clinical devices will no longer be limited in what is possible to achieve on the device’s little processor alone. That enables improvements in the speed of analysis and even the addition of new tests using the same device over time.
Each rung on this ladder of hardware, to software enabled hardware, to cloud-based software enabled hardware, is about improving speed, efficiency, productivity, accuracy and connectivity.
Connectivity has been a relatively new trend in the human health space, in understanding the contextual data and environmental factors outside of disease and pathology alone.
These connections are data-points we are interested in tying together to access a more holistic picture of patient health.
We anticipate, through this process, that a more holistic approach towards collecting and organizing data will illuminate previously un-documented health insights simply because to date, the data has been isolated and siloed. As we begin to put the pieces all together we anticipate patterns will start to emerge.