Frame one dish
Keep the full rim visible with a little space around it. Photograph straight down, with the camera parallel to the dish. Keep your framing consistent across time points.
Guides / Petri dishes
Your first Petri
dish review.
Turn a batch of plate photos into inspectable proposals and a reviewed record. Follow these five steps with an experiment you already know.
Give the image a fair starting point.
Start with an existing dish that is ready for observation. Keep it closed and follow your lab’s handling and containment procedures throughout photography.
Use even light
Choose diffuse, even lighting and a plain background. Check for bright reflections, shadows, condensation, and writing that hides the area you need to inspect.
Check focus & scale
Inspect the image at full size before uploading. Small structures should be distinguishable. Keep camera distance consistent when comparing images.
Pixels need context.
DishFlow’s colony size summary is in pixels. A ruler in the photograph does not automatically calibrate it to millimetres. If physical measurements matter, retain an appropriate scale reference in your original record and use your lab’s validated measurement method.
- One complete, in-focus dish per image.
- A readable plate ID recorded without obscuring the area being reviewed.
- The actual observation time recorded, especially if uploading later.
- An independent copy of the original image kept in your lab records.
Put the batch in one experiment.
Open DishFlow and enter an experiment name for your first batch. If you already have experiments, select the one this batch belongs to or choose a new experiment. Add the photos in the same form.
Use JPEG, PNG, or WebP files, each no larger than 8 MB and 16 megapixels; automated analysis needs at least 128 pixels on each side. The uploader shows your current batch allowance.
The defaults are Colonies and Round dish / plate. To change them, open Measurement type and experiment context and choose the profile that matches your image. The coverage-only profile intentionally does not report a discrete object count.
Match the IDs before you start.
Open Plate IDs and observation time. In Plate IDs, enter exactly one non-empty line per selected photo, in the displayed order. Use the same ID for the same physical plate at later observation times.
Plate-A-01
Plate-A-02
Plate-A-03
If you leave IDs blank, a single readable barcode may supply the ID; otherwise the filename is used. Check every saved ID, particularly when the photo contains multiple labels.
| Batch note | A short reminder about this set of observations. Avoid using the note as the only record of a plate’s identity. |
|---|---|
| Condition & replicate | Optional shared labels. Enter values that apply to the entire batch; you can edit these per plate later. |
| Observed at | The actual observation time, entered in your local time. An empty field uses the upload time. Stored timestamps are UTC. Use separate batches for different observation times. |
| Count context | Dilution multiplier and plated volume are optional and must be supplied together. Leave them blank if you only need an image count. |
Check your choices, then select Analyse this batch. The batch form takes images and these fields; it does not import a metadata CSV.
Inspect the image behind the number.
Start with the Review queue to find unreviewed results and images flagged for quality, dense growth, or no eligible objects. Inspect every plate: a result with no warning still needs your review.
- Open the plate and confirm its ID, observation time, and measurement profile.
- Check that the detected dish area sits inside the actual dish.
- Compare the automated boxes against the photograph. Look for missed objects, merged regions, reflections, labels, and rim artifacts.
- Read the quality warnings alongside the overlay. The quality score is an image heuristic, not a probability that the count is correct.
Dense or overlapping growth needs a decision.
The detector can merge adjacent structures or omit large regions. Its dense warning is not a universal “too numerous to count” decision. Apply your own protocol’s countability rules; keep unresolved or TNTC decisions with your lab record instead of forcing an exact count from an ambiguous image.
Coverage describes segmented image area. It does not establish viable biomass, organism identity, or a biological outcome. See the current evidence and limits.
Correct the marks, then save your review.
Select Start review from these suggestions to copy the automated centres into the editable human layer. This gives you a starting point for checking the image.
- Tap an unmarked object to add a mark.
- Tap near a mark to remove it when it represents a reflection, label, duplicate, or another ineligible feature.
- Inspect the whole dish again, including the edge of the review area.
- Confirm I reviewed the marks shown on this image, then select Save human-reviewed count.
Loading suggestions alone does not save a review. A saved human review has its own revision in the history, separate from the automated result.
Working through a batch? Choose Save and review next to save this plate and continue to the next unreviewed observation.
Need a count, or a concentration?
The count is the number of saved human marks. CFU/mL appears only when a human-reviewed count and both dilution multiplier and plated volume are present. The calculation still depends on valid metadata and a protocol that supports interpreting this plate that way.
Practice adding and removing marks before using your own image →
Take the record back to your lab.
From the experiment’s DishFlow page, choose Download experiment. The ZIP brings together the results CSV, observation metadata, review images, available overlays, and the experiment’s trend graphic.
For one plate, choose Export this observation on its observation page. That ZIP includes the stored image, available overlay, counts CSV, and a JSON record with automated results, human-review history, and observation context.
Keep your export with the original images. AssayPilot stores metadata-stripped JPEGs with a long edge of at most 1,600 pixels; these are review copies, not archival originals.
For a later time point, add a new photo to the same experiment with the same physical plate ID and its new observation time. The notebook can connect those observations. A change in framing, light, or focus can also change an image measurement, so inspect the photos together.
When the first attempt is imperfect
Photo troubleshooting.
The dish boundary is wrong or clipped.
Retake the closed-dish photo with the entire rim visible, a small margin, and the camera centred above it. Use the round-dish geometry for a full dish. A centred fallback means the rim was not confidently detected; inspect that area before accepting any count.
The count is unexpectedly low or zero.
Check focus, glare, the chosen profile, and whether objects overlap. A zero means no eligible objects passed the detector filters, not that the dish is biologically empty. If the original image is readable, use the human marking layer; otherwise retain the image as an unresolved observation under your protocol.
There are marks on glare, writing, or the rim.
The engine attempts to exclude strong glare, decoded barcode areas, and long marker strokes, but that masking is imperfect. Remove false marks during review. If a reflection or label hides the underlying image, a manual click cannot recover the missing information.
The batch IDs or times do not match my notes.
Before uploading, match one ID to each selected photo in order. The batch shares one observation time; create separate batches when those times differ. Inspect the saved IDs and times before using the trend view or exporting results.
Why is CFU/mL blank?
Save a human-reviewed count first, then provide both the dilution multiplier and plated volume on that plate’s count context form. Leave the fields empty when your experiment does not support that calculation.
My phone image will not upload.
Export a still JPEG, PNG, or WebP. HEIC files, animations, files over 8 MB, and images over 16 megapixels are not supported by this uploader. Keep your original, and create a supported copy with enough detail to review.
Does this identify the organism or interpret susceptibility?
No. This workflow proposes image objects and lets you review their marks. It does not identify species, establish viability, diagnose contamination, or make an antimicrobial-susceptibility decision.
Ready for a first batch?
Try the generated example, or start with a small experiment whose images and expected observations you already understand.