How disease detection works¶

You increasingly ask us how disease detection in Apisense actually works and how the app "knows" that something is wrong with a colony. This page explains it in plain language — no formulas, no statistical tables.

If you are interested in how our models are built step by step (from the idea, through the laboratory, to deployment in apiaries), see How our models are built.

In a nutshell¶

The VitalSensor, placed between the frames, continuously "smells" the air inside the hive. Every colony has its own natural "scent" — a chemical fingerprint made up of many compounds. When a disease develops, this fingerprint changes in a characteristic way, usually long before you can see any symptoms with the naked eye.

Our artificial intelligence (AI) has learned to recognise these patterns and, based on them, assigns the colony a health status and an infection level — and you see the result in the Apisense Pro AI app.

A complement, not a replacement

Detecting disease from hive air supports your observations and inspections — it does not replace them. Treat alerts as an early "take a closer look at this colony" signal, not as a final diagnosis.

What the sensor "senses"¶

The VitalSensor measures more than a dozen parameters inside the hive. The most important are:

The chemical composition of the air — volatile organic compounds (VOCs), i.e. the colony's chemical "scent",
Temperature and humidity — the microclimate of the nest,
Other environmental parameters — including carbon dioxide level and acoustic activity (the sounds of the colony).

You can compare it to a doctor who recognises an illness from a patient's breath or smell — except here the "nose" is a sensor and the "memory of experience" is artificial intelligence.

How the AI knows what is sick¶

A sensor reading on its own would tell us nothing without a reference to reality. That is why our models learn from thousands of labelled samples, in which sensor readings are paired with the actual test result for a given colony — e.g. the number of Nosema spores under a microscope or the Varroa infestation level.

We collect this data both in the laboratory (colonies in cages with sensors) and in real apiaries. Each disease leaves its own recognisable "signature" in the data, and the model learns to tell it apart from a healthy state.

We describe this whole process — from idea to global rollout — on the How our models are built page.

Infection levels — what they mean¶

For some diseases the app shows not just "present / absent" but also a level: low, moderate or high. The values below are approximate thresholds we use to label our data — they are aligned with the standard tests beekeepers perform every day, so they are easy to relate to your own apiary.

Varroosis (Varroa destructor)¶

Level	Flotation (alcohol wash)	Daily mite drop	What it means
Low	up to ~2% infestation	2–3 mites per day	Acceptable and common — Varroa is present but does not yet threaten the colony.
Moderate	~3–4% infestation	4–10 mites per day	A disease — worth planning a treatment.
High	above ~5% infestation	above 11 mites per day	Advanced varroosis — the colony needs intervention.

Flotation: alcohol wash vs. sugar roll

The flotation thresholds above refer to the alcohol wash — the reference method (the most accurate, but it destroys the sample). For everyday monitoring we also describe the gentler, non-invasive sugar roll, after which the bees go back to the hive. Both methods report infestation as a percentage, but the sugar roll usually recovers slightly fewer mites.

Nosemosis (Nosema / Vairimorpha spp.)¶

Level	Spore count (per 10 bees)	What it means
Low	1–3 spores	Acceptable and common — nosemosis is present but does not yet pose a threat.
Moderate	4–40 spores	A disease.
High	above 40 spores	Advanced disease.

Where this spore count comes from

The spore count is determined under a microscope — we describe it in the Nosema/Vairimorpha microscopy procedure. The values in the table are the average number of spores per 10 bees.

American foulbrood and chalkbrood¶

For these diseases the model detects presence, not a level:

American foulbrood (AFB) — a "foulbrood risk has appeared" signal. This is a notifiable disease, which is why we collect its data in a special way — described on the How our models are built page.
Chalkbrood — we distinguish a healthy state, mere exposure (contact without visible symptoms) and clear disease symptoms.

More diseases on the way¶

We are constantly developing the system. We are working on detecting further threats, including amoebiasis, the greater wax moth and viruses transmitted by Varroa (e.g. deformed wing virus, DWV). Once the models pass our validation process, we will add them to the app.