Company Spotlight: Scopio X100 - Automating the Blood Smear for Hematology

Sam: Imagine running 1200 blood smear slides through an automated digital scanner, avoiding manual microscopes, and hitting 97.88% overall agreement on red blood cell color. That is the promise we are looking at today.

Maya: Right, and today we are unpacking the FDA 510(k) summary for the X100 and X100HT with Full Field Peripheral Blood Smear Application, published by the US Food and Drug Administration on June 27, 2025. And as a quick reminder, our voices are AI-generated, and this is a summary of a publicly available document.

Sam: And to be totally clear up front, this is a Company Spotlight episode, but we are entirely independent. This is not sponsored, and we are working exclusively from the FDA clearance summary, not from company marketing materials or websites.

Maya: Exactly. So let us get into it. The sponsor here is Scopio Labs Ltd., based in Tel Aviv, Israel. The device is categorized under the regulation name Automated cell-locating device, and it is a Class II medical device under the Hematology product panel. The product code is JOY.

Sam: JOY. I always love the random FDA product codes. But what does this machine actually do in a hospital laboratory?

Maya: The document describes it as a digital cell morphology solution. It presents high-resolution digital images of fixed and stained microscopy peripheral blood smears. So, instead of a lab technologist hunching over a microscope looking at a glass slide, the machine locates and displays images of white cells, red cells, and platelets.

Sam: Which is a big deal for lab workflows. The summary mentions it assists a qualified technologist in conducting a WBC differential, RBC morphology evaluation, and platelet estimate. But how exactly does it get the images?

Maya: It features fully automated scan and image acquisition. The document says it captures multiple images under a plurality of illumination conditions and reconstructs a 100X magnification image of the viewed area, without the need for immersion oil.

Sam: Wait, really? It reconstructs a 100X image without immersion oil? Anyone who has worked in a lab knows oil immersion is messy, it slows you down, and it ruins the slide if you try to go back to a lower magnification. Skipping the oil is a huge physical workflow change.

Maya: It definitely is. And the document gives us a sense of the physical footprint of these machines, too. There are two models. The X100 is 32cm wide, 32cm long, and 35cm high, weighing 14Kg. The X100HT is bigger, at 39cm wide, 42cm long, and 55cm high, weighing 33Kg.

Sam: So the X100 is about the size of a heavy microwave, and the HT model is more like a mini-fridge you would put on a benchtop. They run off a standard 120/100 to 240V power source, so they just plug into the wall. Once a slide goes into this box, what is the software doing?

Maya: The analysis uses what the summary calls standard mathematical methods, including deterministic artificial neural networks. The system streamlines the analysis with a four-step review workflow.

Sam: Let us break down those four steps. What is step one?

Maya: Step one is simply the full field review. You look at the overall picture.

Sam: Makes sense. Step two?

Maya: Step two is the white blood cells review. The document notes that white blood cells are located and counted by moving according to the battlement pattern, which ensures that each cell is counted only once.

Sam: Ah, the battlement pattern. For those who might not know, that is the classic technique where you move the view up, across, down, across, like the top of a castle wall, to make sure you do not double-count cells. It is interesting that the digital system replicates that exact physical microscopy technique.

Maya: Yes, and for the white blood cells, the neural networks are trained to distinguish between classes of cells. The system pre-classifies them, and then the user reviews the suggested classification and accepts or reclassifies the images.

Sam: Okay, so the system takes a guess, and the human confirms it. What about step three?

Maya: Step three is the red blood cells review. And step four is the platelet review. Platelets are also located and counted using that same battlement pattern.

Sam: Now, the X100 and X100HT were actually cleared by the FDA previously, right? The document mentions predicate clearances.

Maya: Yes, they were cleared under 510(k) submissions K201301 on October 2, 2020, and K220013 on May 3, 2022. This new document, K243144, is for a modification to that software.

Sam: So if the hardware is the same, and the four steps are the same, what exactly is the new update that required a whole new FDA clearance?

Maya: The sponsor added a new Decision Support System, or DSS, specifically to steps three and four. They added system-suggested red blood cell morphological pre-gradings, and system-suggested platelet clump indications.

Sam: Decision support for red cells and platelets. How does that interface actually look to the technologist sitting at the screen?

Maya: According to the summary, it is presented to the user by means of a dotted line around a suggested grading selection box.

Sam: A dotted line. That sounds very subtle.

Maya: It is, and the FDA summary is extremely clear on why it is subtle. It states repeatedly, quote, notably, the user is still required to review the slide and actively mark the final grading, exactly as performed in the cleared workflow review.

Sam: That is the crux of the substantial equivalence argument, isn't it? The FDA clearance hinges on the fact that this is not replacing the technologist. It is just drawing a dotted line to say, hey, I think this is a platelet clump, but you still have to click the box to report it.

Maya: Exactly. There is a great comparison table in the document. For the predicate device, it says, the device presents an overview image, and the examiner manually grades RBC morphology. For the modified device, it says the exact same thing, but inserts one sentence in the middle: the system presents to the user RBC morphology grading suggestions.

Sam: So the human is still the ultimate diagnostic engine. The AI is essentially a highly advanced highlighter pen. I think that is a crucial point for hospital leaders buying these systems. You cannot fire your lab staff because you bought this machine. The indications for use explicitly say it assists a qualified technologist. It is for in-vitro diagnostic use only, and for professional use only.

Maya: Right. And to prove that this new dotted line highlighter actually works and does not mislead the technologist, the company had to submit performance data. They conducted a Method Comparison Study.

Sam: Let us get into the numbers. How big was the study?

Maya: They collected a total of 1200 anonymized peripheral blood smear slides from the laboratory routine workload of three medical centers.

Sam: Three medical centers and 1200 slides is a solid sample size. Though I will note, the summary just says collected from the routine workload and anonymized. It does not explicitly state if this was a prospective clinical trial where they followed patients over time, or just a retrospective pull of old slides. Based on the wording, it sounds retrospective.

Maya: That is a fair observation. The document does not specify. But it does break down the agreement between the modified X100 software and the established reference standard across different categories. They measured Overall Agreement, Positive Percent Agreement, and Negative Percent Agreement.

Sam: Positive Percent Agreement, or PPA, is basically how often the software agreed when the feature was actually there. And Negative Percent Agreement, NPA, is how often it agreed when the feature was missing. What were the results for the red blood cells?

Maya: For RBC Color, the Overall Agreement was 97.88%. For RBC Shape, Overall Agreement was 96.22%. And for RBC Size, Overall Agreement was 95.58%.

Sam: Those are very high numbers. Mid-to-high nineties across the board for color, shape, and size. But I am looking at the data table here, and RBC Inclusions is an interesting one.

Maya: Yes, for RBC Inclusions, the Overall Agreement was still high at 97.90%. But the Positive Percent Agreement was noticeably lower, at 86.73%.

Sam: And the Negative Percent Agreement for inclusions was 98.41%. So if I am reading that right, the software is very good at knowing when an inclusion is not there, but it is slightly less sensitive at catching an inclusion when it is actually present, agreeing only 86.73% of the time.

Maya: That matches the numbers in the document. And the summary does not offer a medical explanation for why inclusions are harder for the neural network to flag than color or size. It simply states that the results met the pre-defined acceptance criteria.

Sam: What about the platelets? Step four of the workflow. The new feature was suggesting whether platelet clumps were detected.

Maya: For PLT Clumping, the numbers were a bit lower overall. The Overall Agreement was 87.08%. The Positive Percent Agreement was 86.11%, and the Negative Percent Agreement was 87.39%.

Sam: So with positive percent agreement at 86.11% and negative percent agreement at 87.39% for platelet clumps. Again, not as high as the 97.88% for red blood cell color, but apparently high enough to meet the FDA's bar for a system that just offers a dotted-line suggestion.

Maya: Exactly. Because the user must confirm or adjust the information, the FDA concluded that this difference does not raise different questions of safety or effectiveness.

Sam: Beyond accuracy, lab equipment has to be reliable day to day. Did they test for that?

Maya: They did. The sponsor ran Precision Studies according to CLSI guidelines. First was a Repeatability study, which was a 20x2x2 study using 27 test samples representing different red blood cell morphological features at various levels of severity, as well as the presence of platelet clumping.

Sam: A 20x2x2 design usually means running it over 20 days, with 2 runs per day, and 2 replicates per run. So they hammered those 27 slides over and over to see if the machine gave the same answers.

Maya: Right. And they also ran a Reproducibility study. That was a 3x5x5 study using 16 test samples.

Sam: A 3x5x5 would typically mean 3 sites or operators, 5 days, 5 replicates. It tests if different machines or different users get the same results. Did the document give the variance percentages for those?

Maya: It actually does not list the raw data for precision. It simply states that for both the repeatability and reproducibility studies, the results met the pre-defined acceptance criteria.

Sam: Which is classic for a 510(k) summary. They do not have to show every data point to the public, they just have to show the FDA that they passed the test for substantial equivalence.

Maya: There is also a note on daily quality control in the lab. The daily QC involves testing a standard blood smear slide. The user reviews the images and verifies that at least 95% of the white blood cells in the scanned image were identified and located correctly by the system.

Sam: That is a great practical detail. Every morning, a tech runs a known slide and makes sure the X100 is finding at least 95% of the white blood cells before they process any real patient samples.

Maya: The document also covers software verification and validation testing. The FDA classified this software at the Basic Documentation Level.

Sam: Basic Documentation Level. Why is it basic and not a higher risk tier?

Maya: The summary explains it is because, quote, a failure or flaw in the software would not present a hazardous situation with a probable risk of death or serious injury, either to a patient, user of the device, or others in the environment of use.

Sam: Which brings us right back to the human in the loop. If the software draws a dotted line around a piece of dust and calls it a platelet clump, the technologist is sitting right there looking at the high-resolution image to say, nope, that is dust, and they do not check the box. The human is the safety net.

Maya: Exactly. The intended use explicitly states it assists a qualified technologist. It does not replace them.

Sam: So, pulling this all together, what is the big takeaway for our listeners? For the lab directors out there dealing with severe staff shortages, this device seems to offer a way to get rid of oil immersion microscopy and speed up the manual review process.

Maya: Yes, by reconstructing a 100X image digitally and using neural networks to pre-classify cells and suggest morphology grades, it centralizes the workflow to a screen rather than a microscope eyepiece.

Sam: And for the digital health builders, it is a masterclass in how to clear an AI update through the FDA. By framing the AI as a decision support system that only offers a dotted line suggestion, and by mandating that the user actively clicks to agree, Scopio avoided raising new questions of safety and effectiveness.

Maya: Right. The overall agreement numbers were strong, hitting 97.88% for red blood cell color, even if they dipped to 87.08% for platelet clumps. The system proved it was substantially equivalent to its own previous versions.

Sam: It is a fascinating look at the steady, step-by-step automation of the hematology lab.

Maya: That wraps up our look at the FDA 510(k) summary for the Scopio Labs X100 and X100HT. You can find a link to the full document in the show notes.

Sam: And remember, we are discussing a public regulatory document for educational purposes. This is not medical advice. Thanks for listening.