Hilab is a point-of-care hematology system that combines microfluidic blood processing, artificial intelligence image analysis, and cloud-based telepathology to deliver a complete blood count from a fingerstick sample, outside a traditional laboratory. Developed in Brazil, it connects a compact bedside device to a remote team of licensed health professionals through the internet, creating a hybrid workflow where AI does the initial heavy lifting and a human specialist signs off on every result. Validation studies comparing Hilab to conventional hospital-grade analyzers have shown strong correlations across most standard blood parameters, making it one of a handful of platforms trying to move routine hematology closer to the patient.
How the System Works
A conventional complete blood count, or CBC, requires a tube of venous blood drawn by a phlebotomist and sent to a central lab where a large automated analyzer counts and classifies cells. Hilab condenses much of that process into a small device that accepts a few microliters of capillary blood from a fingerstick. The sample is loaded into a disposable capsule containing reagents and a microfluidic channel, which handles the staining and preparation that a lab technician would normally perform manually.
Once the blood cells are stained and spread through the microchannel, the device captures high-resolution images using an integrated optical system. Those images are then transmitted via the internet to Hilab’s cloud platform, where an AI algorithm analyzes cell morphology, counts different cell types, and generates preliminary results. But the process does not stop there. A licensed health professional working on Hilab’s interactive platform reviews the AI-processed images, checks the automated results, and makes corrections if needed. If the human analyst and the AI disagree, the human’s judgment always wins. After this dual verification, the signed report reaches the patient by email or text message.
1Scientific Reports. Hilab system, a new point-of-care hematology analyzer supported by the Internet of Things and Artificial IntelligenceThis telepathology loop is central to the system’s design philosophy. Pure AI analysis, no matter how accurate on average, can miss unusual cell morphologies or be thrown off by sample artifacts. The remote human review acts as a safety net, and the dual-verification approach is intended to minimize the kinds of analytical errors that plague decentralized testing when only a machine or only an undertrained operator is making calls.
2medRxiv. Analytical Accuracy and Clinical Agreement of a Novel Internet of Things and AI-based Point-of-Care Testing LaboratoryAccuracy Compared to Standard Laboratory Analyzers
The core question for any point-of-care blood analyzer is whether its numbers agree with the reference-standard machines used in hospital labs. For Hilab, the primary validation study compared results against a Sysmex XE-2100, a well-established automated hematology analyzer. Across all tested blood parameters, the correlation coefficients were at or above 0.8. Hemoglobin came in at 0.95, hematocrit at 0.96, white blood cells at 0.99, and platelets at 0.95. Kappa coefficients, which measure agreement beyond what you would expect by chance, were also above 0.8 for every analyte. The one exception worth noting was the eosinophil/basophil count, which still met the 0.8 correlation threshold but showed a statistically significant difference between the two methods.
1Scientific Reports. Hilab system, a new point-of-care hematology analyzer supported by the Internet of Things and Artificial IntelligenceThese numbers are encouraging, but context matters. Other point-of-care hematology devices have shown similar strong correlations for the “big five” parameters (hemoglobin, red blood cells, platelets, white blood cells, and neutrophils) while stumbling on certain differential counts. A clinical evaluation of the HemoScreen, another portable CBC analyzer, found strong linear correlations for hemoglobin, red blood cells, platelets, white blood cells, and absolute neutrophil count, but hematocrit and mean cell volume ran significantly lower than the reference instrument, and monocyte counts were roughly halved.
3PubMed Central. Clinical Evaluation of a Novel Point‐of‐Care Hematology Analyzer for Complete Blood Count With DifferentialThe pattern across point-of-care devices is consistent: the parameters that matter most for urgent clinical decisions, like hemoglobin and white blood cell count, tend to agree well with central lab results. The differential counts for less abundant cell types, particularly monocytes, eosinophils, and basophils, are where portable devices struggle. That is partly a numbers game. When a cell type makes up only a few percent of the total white cells, small counting errors translate into large percentage differences.
The Hilab Lens and Pediatric Oncology
A more recent iteration of the technology, called the Hilab Lens, has been clinically validated in a particularly demanding setting: pediatric oncology. Children undergoing chemotherapy need frequent blood counts to monitor for dangerous drops in white blood cells or platelets, and each standard venous blood draw is painful and stressful for a small child. A fingerstick-based point-of-care device that delivers reliable results at the bedside would be a genuine improvement in quality of care.
Precision testing of the Hilab Lens showed that the vast majority of measurement variability came from biological differences between samples, not from inconsistencies in the device itself. Part-to-part variability, meaning the natural differences among patient samples, accounted for 98% or more of total variance across all analytes tested. Reproducibility, the variability introduced by different operators running the same sample, had minimal impact. In practical terms, that means two different nurses running the same device on the same blood sample would get essentially the same answer, which is the benchmark you need for a device to be useful outside a controlled laboratory environment.
4Communications Medicine. Clinical validation of the Hilab lens for AI supported point of care CBC testing in pediatric oncologyOne hemoglobin precision metric slightly exceeded the acceptable coefficient of variation set by the European Federation of Clinical Chemistry and Laboratory Medicine, but overall performance for total error and bias remained within limits. For a point-of-care instrument being compared against central laboratory standards, this is considered acceptable. POC devices are generally held to somewhat wider precision tolerances than the large analyzers they are being measured against, because the clinical utility of getting an answer in minutes rather than hours offsets modest reductions in precision.
The Role of AI in Reading Blood Cells
Behind the scenes, the AI that powers systems like Hilab draws on a rapidly maturing field of computer vision applied to microscopic blood images. The basic task is straightforward to describe but hard to automate: look at a stained blood smear image, find every white blood cell, identify what type it is, and flag anything abnormal. For decades this was done by trained laboratory scientists peering through microscopes. Now deep learning models can do the initial classification at scale, though the accuracy varies by cell type and by how challenging the image is.
A deep learning framework for leukocyte segmentation and classification recently achieved roughly 98% accuracy in identifying and categorizing white blood cells from microscopic images.
5PubMed Central. Deep learning-based image annotation for leukocyte segmentation and classification of blood cell morphology Those numbers sound almost too good, and it is worth understanding the caveat: performance drops when the model encounters interference factors like overlapping cells, debris, or unusual staining. An ensemble detection model designed to handle such real-world interference achieved a mean average recall of about 92%, meaning it missed roughly 8% of leukocytes under challenging conditions. It performed best on neutrophils, the most common white blood cell type, and worst on basophils, which are rare.
6Scientific Reports. A deep learning model for detection of leukocytes under various interference factorsWhen the task gets harder still, such as classifying across 40 different leukocyte categories rather than the standard five, accuracy falls to roughly 77% at the top prediction.
7PubMed. Fine-grained leukocyte classification with deep residual learning for microscopic images This is why Hilab’s architecture includes the human analyst in the loop. The AI handles the high-volume, routine classification work quickly, and the specialist catches the edge cases that would trip up the algorithm. It is a pragmatic design choice rather than a concession of failure: even in well-funded central labs, automated analyzers flag certain results for manual review by a pathologist.
Microfluidics and Sample Preparation
The microfluidic component of point-of-care hematology devices does a surprising amount of work in a very small space. The basic idea is to move a tiny volume of blood through channels narrower than a human hair, where physical forces separate cells, mix reagents, and position everything for imaging. Research into microfluidic blood analysis has demonstrated that on-chip devices can now replicate a wide range of routine laboratory blood tests, combining hydrodynamic, optical, electromagnetic, and acoustic methods to characterize blood cells with precision.
8PubMed Central. Microfluidic Systems for Blood and Blood Cell CharacterizationOne particularly clever application involves using geometric constrictions in the microchannel to separate red blood cells from plasma. When blood flows through a narrowing section of the channel, the deformable red blood cells are pushed toward the center of the flow by hydrodynamic forces, leaving a cell-free layer of plasma near the walls. This effect can be tuned by adjusting the constriction geometry, flow rate, and fluid viscosity.
9Biorheology: The Official Journal of the International Society of Biorheology. Geometrical focusing of cells in a microfluidic device: An approach to separate blood plasmaStaining is another critical step. In a traditional lab, a technician smears blood on a glass slide and applies chemical stains that color different cell components. Microfluidic systems adapt this by incubating suspended cells with staining solutions directly in the channel. One approach uses eosin to highlight red blood cells and white blood cell cytoplasm, and methylene blue to stain white blood cell nuclei, with concentrations adjusted to optimize clarity for each cell type.
10PubMed Central. Microfluidics-based cell recognition through optimizing suspended cell staining techniques and artificial intelligence The entire process happens inside a disposable capsule that the user simply inserts into the analyzer, which is a far cry from the manual preparation required in a conventional hematology lab.
Capillary Versus Venous Blood
A recurring concern with fingerstick-based devices is whether capillary blood gives the same answers as the venous blood used in standard laboratory testing. Capillary samples are inherently more variable. They are affected by how hard the finger is squeezed (which can dilute the sample with tissue fluid), how warm the hand is, and how quickly the sample is collected. These factors introduce noise that does not exist with a clean venous draw.
Research comparing complete blood count parameters between capillary and venous samples has found strong linear correlations and acceptable variation between the two collection methods.
11PubMed. Stability and comparison of complete blood count parameters between capillary and venous blood samples The agreement is good enough for most clinical decisions, though individual parameters like hematocrit and platelet counts can show more scatter with capillary samples. For Hilab’s use case, the convenience of a fingerstick is a deliberate trade-off: losing a small amount of analytical precision to eliminate the need for phlebotomy, a tourniquet, and a blood collection tube makes the test accessible to settings that simply could not offer venous draws on demand.
Known Limitations and Interference
No point-of-care device is immune to the factors that confound blood analysis. Known interferences for portable hematology analyzers include cold agglutination (where red blood cells clump at low temperatures), very small red blood cells, high bilirubin levels, the presence of nucleated red blood cells, and elevated lipid content. All of these arise from the inherent limitations of the measurement method used.
12The Journal of Applied Laboratory Medicine. A Novel Approach to Hematology Testing at the Point of CareSome of these interferences affect point-of-care devices differently than they affect large lab analyzers. For instance, hemolysis, the rupture of red blood cells that releases hemoglobin into the surrounding fluid, typically inflates hemoglobin readings on conventional analyzers. Some portable devices measure mean corpuscular hemoglobin directly rather than calculating it, which can make them more robust to hemolysis up to a point. The HemoScreen, for example, can detect erythrocyte membranes and fragments to estimate the degree of hemolysis and will suppress results if hemolysis exceeds a threshold where accuracy cannot be maintained.
12The Journal of Applied Laboratory Medicine. A Novel Approach to Hematology Testing at the Point of CareHemoglobin measurement specifically is a weak spot for several point-of-care instruments. A study comparing three different POC devices against a reference hematology analyzer found clinically discordant hemoglobin results in the critical transfusion-decision range of 70 to 100 grams per liter in roughly 9% to 35% of cases, depending on the device.
13PubMed. Hemoglobin determination with point-of-care testing, performance evaluation compared to central laboratory analyzers in transfusion decision, an in vitro and retrospective study That is a range where an inaccurate hemoglobin value could lead a clinician to transfuse a patient who does not need blood, or to withhold a transfusion from one who does. Hilab’s dual-verification model, where a human specialist reviews every result, is partly a response to this kind of risk.
Why Turnaround Time Matters
The clinical value of point-of-care testing is not just accuracy. It is speed. In a conventional emergency department workflow, a blood sample goes from the patient to a phlebotomist to a transport bag to a pneumatic tube or runner to a central lab, where it queues behind other samples before being loaded onto an analyzer. That chain can take 45 minutes to well over an hour, depending on the facility. In rural emergency departments, where the lab might be at a distant hospital, the delays are even longer and directly affect patient care.
14PubMed Central. Point of Care Testing (POCT) Hematology devices in emergency departments: a mixed methods studyA POC hematology device that returns a CBC in 10 to 20 minutes at the bedside collapses that entire chain. Clinicians can make disposition decisions faster, patients spend less time waiting, and in time-sensitive conditions like sepsis or acute hemorrhage, the faster information can change outcomes. Competing multi-biomarker platforms are pushing the envelope even further: one system in development reports a full panel spanning hematology, clinical chemistry, and immunoassays in about 20 minutes from less than half a milliliter of whole blood.
15Royal Society of Chemistry. VitalOne™: a point-of-care platform for rapid, comprehensive, central-lab quality blood testingCost and Health Economics
Point-of-care testing is almost always more expensive per test than running the same assay in a high-throughput central lab. The reagent cartridges cost more, the instruments are less efficient at processing large batches, and the quality-control overhead per result is higher. But the per-test cost is only part of the equation. A systematic review of health-economic evidence for POC testing found that while upfront costs often increase, the downstream savings from shorter hospital stays, fewer unnecessary specialist referrals, and reduced inappropriate antibiotic prescribing can offset or exceed those costs.
16PubMed Central. Health Economic Evidence of Point-of-Care Testing: A Systematic ReviewIn the emergency department specifically, a cost-effectiveness analysis of different POC testing combinations found that adding a point-of-care CBC was one of the most cost-effective permutations, primarily because the time saved in the ED translated into throughput gains and reduced staffing costs. The analysis concluded that higher staffing costs made POC testing even more economical, since each minute of ED time saved is worth more at facilities where labor is expensive.
17PubMed Central. The cost-effectiveness of upfront point-of-care testing in the emergency department: a secondary analysis of a randomised, controlled trialFor a system like Hilab, the economic calculation includes the cost of the cloud infrastructure and the salaries of the remote analysts who review every result. That ongoing operational expense is unusual for a POC device and adds a layer of cost that purely automated analyzers do not carry. Whether the improved reliability from human oversight justifies that expense depends heavily on the clinical setting. In a rural clinic where the alternative is no hematology testing at all, the value proposition is clear. In a large urban hospital with a 24-hour stat lab down the hall, it is harder to justify.
Usability Outside the Lab
A device that produces accurate results but is too complicated for frontline staff to operate reliably is not going to change clinical practice. Usability has been a persistent barrier to POC adoption. When community paramedics evaluated two commercially available POC devices, there was a clear preference for the one with simpler operation: the preferred device scored 84 out of 100 on a standardized usability scale, compared to roughly 60 for the other.
18PubMed Central. Community paramedic point of care testing: validity and usability of two commercially available devicesHilab’s design addresses this partly by offloading the analytical complexity to the cloud. The operator’s job at the bedside is limited to collecting the fingerstick sample, loading the capsule, and inserting it into the device. The staining, imaging, and analysis happen either automatically within the device or remotely in the cloud. A QR code on each capsule tracks sample identity through the system, reducing the chance of mix-ups.
2medRxiv. Analytical Accuracy and Clinical Agreement of a Novel Internet of Things and AI-based Point-of-Care Testing Laboratory The trade-off is dependence on internet connectivity. In settings with unreliable connections, a cloud-dependent system introduces a failure mode that a standalone analyzer does not have.
AI-Driven Detection of Blood Cancers
One of the more promising frontiers for AI-assisted blood analysis is the early detection of hematological malignancies. Acute lymphoblastic leukemia, the most common childhood cancer, is diagnosed by identifying abnormal blast cells in blood smears. An automated detection system using the YOLOv4 object-detection algorithm achieved a mean average precision above 96% on one standard dataset and nearly 99% on another when classifying cells as either leukemic blasts or healthy.
19Biomedical Signal Processing and Control. Automated blast cell detection for Acute Lymphoblastic Leukemia diagnosisThis is not the same as diagnosing leukemia in a clinical setting. Research datasets are curated, well-stained, and pre-labeled, which makes the AI’s job considerably easier than analyzing a real-world sample that might have poor staining, debris, or unusual cell morphologies. Still, the direction is compelling. A point-of-care device that could flag a suspicious blood smear and route it for immediate specialist review could shorten the time between a child’s first concerning blood count and a definitive diagnosis. For systems like Hilab that already have a remote pathology specialist in the loop, adding a blast-cell detection module to the AI pipeline is a natural extension of the existing architecture.
Testing in Resource-Limited Settings
The settings that stand to benefit most from portable hematology are precisely the ones where validation is hardest. A study of hemoglobin point-of-care testing in rural Gambia found that while one tested device performed well, operational limitations around measurement duration made it less suitable for high-throughput screening in a busy field clinic.
20PLOS ONE. Hemoglobin point-of-care testing in rural Gambia: Comparing accuracy of HemoCue and Aptus with an automated hematology analyzer Heat, humidity, dust, inconsistent power supply, and limited cold-chain storage for reagents all affect device performance in ways that controlled validation studies cannot fully capture.
Hilab’s cloud-dependent model introduces an additional constraint in these environments: you need a stable enough internet connection to transmit high-resolution blood smear images. In urban clinics across Latin America, that is generally feasible. In a rural sub-Saharan African health post, it may not be. Competing devices that perform all analysis on-board, without a cloud connection, have an advantage in truly remote settings, even if they sacrifice the human-review safety net. The ideal technology for a given location depends less on which device has the best published correlation coefficients and more on whether it can actually function reliably under local conditions, day after day, operated by the staff who are available.