🐍  A Beijing military hospital has a newly granted Chinese patent for software that works out which kind of venom is in a snakebite, even when the rapid test says there is none.

🌡️  It blends three clues: where you were bitten, what the test strip shows, and how your wound swells and cools on a paired regular and thermal camera over several hours.

💰  China sees 250,000 to 280,000 venomous bites a year, has one licensed antivenom maker, and covers only about 12% of the need.

⚠️  The headline feature lets the algorithm overrule a negative test, and the filing publishes no accuracy figures from real patients.

The negative that wasn't

It's a Saturday afternoon and you're cutting through long grass toward a lookout everyone swore was worth it. Something stings your ankle. You look down and see two small marks, with no snake anywhere, so you decide it was a bug and keep walking.

Forty minutes later your sock feels tight. By the time a friend drives you to the nearest emergency department, the skin around the marks has turned shiny and red.

A nurse dabs fluid from the wound onto a plastic strip about the size of a stick of gum. Ten minutes later there's a single line, which means no venom detected, and you feel your shoulders drop.

Then a doctor wheels over a camera on a stand and photographs your ankle, then again twenty minutes later, then again. On the screen, a thermal image shows a cold patch at the center of the bite, and it's growing. The software flags a conflict. The strip says you're fine, your leg's reaction says otherwise, and the system is built to side with your leg.

That logic comes from a patent recently granted to a military hospital in Beijing. It hands emergency medicine an odd new question. When a lab test and an algorithm disagree about your body, who gets the final word?

The patent belongs to the Fourth Medical Center of the Chinese PLA General Hospital in Beijing, and it targets the question emergency doctors face with every venomous bite. Which antivenom does this person need? One of its four inventors, Liu Hongsheng, heads the center's emergency department and has become a public voice on snakebite first aid in China.

The simple version is a system that reads a bite the way a weather service reads a storm, combining a forecast with what the radar shows right now.

Think about how a good rural doctor already works. They know which snakes live nearby, they look at any test result, and they watch what the wound does over the next few hours. The patent turns each of those habits into a variable.

First, the system takes the GPS coordinates of the bite. It pulls altitude and vegetation type, checks a map of where each venomous species has historically turned up, and produces starting odds for each venom type. A bite in a southern bamboo grove starts with different odds from one on a northern hillside.

Next comes the test strip. A camera reads the strip's color, and the system adjusts that reading for how long ago the bite happened. Venom drains from the wound into the body over time, so a faint line two hours in may mean more than it looks.

The location odds and the corrected strip reading feed a statistical model that picks a leading suspect. That suspect then sets up the camera.

Blood-damaging venoms tend to cause fast-spreading swelling. Tissue-destroying venoms tend to kill tissue near the bite, which shows up as an unusually cold patch on a thermal image. The system tunes how often it shoots and how sensitive the thermal sensor is to whichever pattern it expects.

Then it watches. Paired regular and thermal photos are taken at set intervals. Software tracks how fast the edge of the swelling moves outward and how fast the cold center spreads.

After that, it compares the trajectory against reference curves for each venom type, stretching or squeezing the timeline to allow for patients who deteriorate faster or slower than average.

If the strip reads negative but the wound is worsening past a high-risk threshold, the system cuts the strip's vote sharply and hands the decision to the images. The filing calls this forced correction. In that conflict, the camera decides.

If it works as described, a negative strip would trigger a closer look where it once brought relief, and the choice of antivenom would rest on how the bite behaves over hours.

Antivenom works best when it's matched to the venom. China's antivenoms come in four types, made against the pit viper, five-pacer viper, many-banded krait and cobra (National Business Daily), so picking the wrong one, or waiting too long to pick, costs tissue and sometimes lives.

Most of the time nobody knows which snake did it. Globally, the snake is never identified in nearly half of bites (Frontiers in Artificial Intelligence).

That pushes doctors toward rapid tests, which have a known blind spot. When venom is very concentrated, the antibodies on a strip get saturated and the line can fail to appear, a quirk chemists call the hook effect (Taylor & Francis). Venom also drains out of the wound with time, so a late swab can come back clean.

Where a commercial bite-site kit does exist, the record is patchy. Australia's decade-long national snakebite study found its kit returned positives for 36% of people who were never envenomed (Medical Journal of Australia), so roughly one in three people with nothing to treat still saw a warning.

The real bottleneck is a fast test doctors can trust when it says no.

The field splits three ways.

1.  People building better tests
2.  People identifying the snake, and
3.  People watching the wound.

This patent tries to do all three at once.

The closest neighbors are other Chinese filings. The patent's own citations include a neural-network venom identification method from Ocean University of China and an intelligent snake venom poisoning diagnosis method from Shantou University, filed in 2025. China's patent office is seeing a cluster of software-led snakebite ideas, and none has yet shown up as a product in hospitals.

The incumbent to watch is Syren Biotech, the only company in China licensed to make antivenom. It is developing rapid diagnostic reagents for cobra, pit viper and five-pacer venom, and already runs a WeChat mini-program that helps the public find hospitals stocking serum (National Business Daily). A company that controls the drug and the map of where it sits could fold a decision tool into its sales channel with little effort.

Outside China, Denmark's VenomAid Diagnostics, co-founded by antibody researcher Andreas Laustsen, is building an affordable lateral-flow test for low-resource clinics, with the Novo Nordisk Foundation and Good Ventures among its backers (Caplight). It's the better-strip approach, the exact layer this patent assumes will sometimes fail.

On identification, a University of Geneva team trained an AI model on 386,006 snake photos and reported 96% species-level accuracy (PLoS Neglected Tropical Diseases). MSF has piloted it around two hospitals in South Sudan (MSF). Its weakness is plain enough. It needs a picture of the snake, and most bite victims never get one.

Wound watching is the least developed. A 2025 study in India followed 39 envenomed patients with serial thermal imaging and found the thermal changes often reached further than the swelling doctors could detect by hand (Transactions of the RSTMH).

The story starts with antivenom, the one product in this space with real revenue. IMARC puts the global market at US$1.31 billion in 2025, while Persistence Market Research says US$2.6 billion. A spread of nearly double tells you the category has fuzzy edges (scorpion and spider products, veterinary use, donated stock).

Syren Biotech booked RMB 168 million (roughly US$23 million) in antivenom revenue in 2025, selling about 150,000 vials a year against 250,000 to 280,000 venomous bites, which works out to around 12% coverage (National Business Daily).

The same report explains why. A vial costs RMB 1,000 to 2,500, needs strict storage and expires. One Yunnan county said its main hospital sees six or seven bite patients a year, so stock often expires unused, and smaller clinics don't carry any.

That's the structure a diagnostic tool would sit inside. Hospitals under-stock because they can't predict demand or pick the right type with confidence. A tool that tells a county hospital which of four antivenoms it needs could let a region stock centrally and route patients faster. Saving a hospital from tossing vials is an easier pitch to a budget holder than better diagnosis.

A 2024 paper in Scientific Reports noted only one commercial multiplex venom test on sale, and only in Australia. That leaves no incumbent product to displace and no proven buyer to sell to.

The likeliest early buyer is a military. The US Department of Defense put US$13.8 million into Ophirex's oral snakebite drug for troops operating far from hospitals (Stars and Stripes). A PLA hospital filing this patent fits the same logic, with soldiers on remote postings bitten by snakes nobody saw.

From there the frame widens to every emergency department that handles bites without a toxicologist on call. So will a tool priced for a military budget ever reach the county hospital that sees seven bites a year and needs the help most?

Snakebite money arrives in small, lumpy checks, and most of it goes to treatments, with diagnosis a distant second.

The treatment flagship is Ophirex, a California public benefit corporation developing varespladib, an oral pill that blocks a venom enzyme common to most snake venoms. It won a US$13.8 million US Army contract in 2022 and raised a US$37 million round the following year (Stars and Stripes). Wellcome also paid to manufacture its clinical trial supply (Business Wire).

Platform money is bigger but less committed. Centivax, the South San Francisco biotech behind a broad-spectrum antivenom built from the antibodies of a man who self-immunized for 17 years, raised a US$45 million Series A led by Future Ventures in 2025 (HIT Consultant) and another US$37 million led by Structure Fund in March 2026 (Pulse 2.0). Most of it backs a universal flu vaccine, with antivenom one of four follow-on programs.

Diagnostics get seed-sized checks. VenomAid extended its seed round to US$1.8 million in January 2024 (BioSpace), including a US$990,208 investment recommended by Open Philanthropy (Good Ventures).

In China, the money looks different. Syren Biotech posted 2025 revenue of RMB 196 million (about US$27 million) and paid out 79% of net profit as dividends (NetEase), a payout ratio that reads like a company harvesting a steady earner. Jiangxi Bio, a biologics maker that listed in Hong Kong in June 2026, plans to file its own pit viper antivenom for approval in late 2027 (National Business Daily), which would end Syren's monopoly on paper.

The core harm is a wrong override. The feature that makes this patent interesting, its willingness to set aside a negative test, is also the one most likely to hurt someone.

Start with a farmer who arrives with a hot, swollen foot after a bite in a field. The strip is negative because the bite was dry (no venom injected), but a fast-moving infection makes the swelling curve steep enough to cross the danger threshold, and the system recommends antivenom.

Antivenom carries its own danger. In Australia's national study, 24% of patients given antivenom had a systemic allergic-type reaction and 6% had a severe one (Medical Journal of Australia).

Now flip it. Krait bites are known for causing little pain or swelling at the site while the venom goes after the nerves. A system that leans on swelling and cooling may see a calm wound and give extra weight to the strip's no. The patent acknowledges neurotoxins in its background section, but its camera layer is built for the visible damage that blood and tissue toxins cause.

Then there's feasibility. The comparison step needs reference curves showing how swelling and skin temperature evolve, hour by hour, for each venom type. Building those means serial thermal imaging of many confirmed bites, and the main published study so far covered 39 patients.

The filing explains how its thresholds would be set (the negative cutoff from healthy volunteers' test results, the danger cutoff from confirmed severe cases), but it reports no accuracy figures from real patients. Transfer is a problem too. The hospital's own published snakebite work centers on northern pit vipers (Journal of Clinical Emergency), while in Yunnan two other viper groups account for about 80% of venomous bites (Frontiers in Toxicology via PMC). Curves tuned in Beijing may not hold in the southwest.

For patients, the near-term change could be small and important. A negative strip would prompt a closer look before anyone sends you home. The camera layer could also give doctors something they rarely have now, a time-stamped record of how a bite progressed, which matters for the next treatment decision and for any review afterward.

The obvious first home is the PLA's own hospitals and field units, where bites happen far from specialists. A phone with a clip-on thermal sensor would push it further, toward township clinics.

This week's patent is CN122245716B, titled 'A method and system for auxiliary determination of snake bite venom type by integrating rapid snake venom immunoassay results, bite image and data evolution characteristics', published by the Fourth Medical Center of the Chinese PLA General Hospital.

Hit reply and tell us where you'd draw that line, or come argue it out with us on Instagram and LinkedIn.

•  Thermal imaging versus the naked eye with Transactions of the Royal Society of Tropical Medicine and Hygiene: Read the Indian study that photographed envenomed limbs on a schedule, the closest real-world evidence for this patent's camera layer.

•  Stress-testing a multi-venom strip with Scientific Reports: Explore how researchers pushed a lateral-flow test across six orders of magnitude of venom concentration and where blood and tissue samples tripped it up.

•  An AI model for 772 snake species with PLoS Neglected Tropical Diseases: Learn how the Geneva team built a global photo identifier, and why it pairs naturally with location data the way this patent does.

•  Every snakebite treatment in the pipeline with Wellcome: Discover the commissioned dataset tracking snakebite drugs and biologics since 2015, a useful map of where the money has gone.

•  The original case for AI in snakebite care with The Lancet Digital Health: Zoom out to the 2019 comment that sketched matching clinical signs to local snakes when no photo exists, years before this patent formalized it.