Plate counter and analysis
Counts colonies on photographed agar plates, turns counts into CFU per mL, and applies the rules from the test plan. Everything runs in this page; no images leave your computer.
agar_plate_protocol.md.Load a photo of a plate (taken through the lid, straight on, even light, dark background). Click a marker to remove a wrong one; click empty agar to add a missed colony.
Dilution factor: 1 for undiluted, 100 for 1 in 100, 1000 for 1 in 1,000. The count saved is the corrected count (auto count, minus removed, plus added).
Rows are saved in this browser automatically. Export a CSV regularly as a backup; it uses the same columns as plate_log_template.csv, plus auto_count_day7.
Findings
Counts by round
Chart
Counter calibration
How to use it
- Photograph each sealed plate through the lid, straight on, evenly lit, on a dark background, with the plate filling most of the frame.
- Load the photo, press Auto-detect plate, and adjust the circle so it sits just inside the agar edge.
- Check the markers. Click a wrong marker to remove it, click a missed colony to add it. Orange markers (×n) are touching groups counted by size.
- Fill in the sample details and press Add to log. Do this for counter 1 and counter 2 (two people counting independently), and for days 3, 5 and 7.
- Open Analysis to see CFU per mL and the findings.
How the numbers are worked out
CFU per mL = colonies ÷ volume plated (mL) × dilution factor. A plate with 30 to 300 colonies is used as a proper count. With 1 to 29 colonies the value is shown as an estimate (~). Above 300 it is shown as a lower bound (>). With none it is shown as below the detection limit (<).
How accurate is the counter?
Tested on computer-made plates with known colony numbers (tests/counter.test.js), including glare, uneven lighting, noise, dark or light colonies and a smaller image:
- 30 to 280 colonies: average error about 3% to 7%, worst case about 14%.
- Fewer than 30: usually within 1 to 4 colonies.
- When about a quarter of colonies touch, it undercounts by about 14% on average. It flags touching groups so you can check them.
- It finds the plate edge to within about 2% of the radius.
Real photos will be harder than synthetic ones (condensation, bubbles, reflections, overlapping colonies, similar colours). Before trusting it, count 10 or more plates by hand as well and look at the Counter calibration section of the Analysis tab. The automatic count is stored separately from the manual counts, so you can compare them.
What it can't do
It doesn't identify organisms, can't tell yeast, mould and bacteria apart, and can't judge whether anything is harmful. Counts show exposure, not harm. Confirm anything important with an accredited laboratory.