Methodology
How MIRU reads, and how well.
No accuracy theatre. Here is exactly what the read does, how it was validated, and where it is weakest, in plain language.
The read, step by step
Your capture is graded first on your device. Light, sharpness, contrast. A weak photo is flagged before it can become a confident-sounding answer.
The photo is read once by our analysis service against six signals, then deleted. Deletion is architectural, not a policy promise.
Every signal is labelled Observed, Inferred, or Unknown. MIRU tells you which, every time, on every reading.
Six signals become one focus. Ranking many concerns at once multiplies error. Naming the strongest one keeps the reading honest and the action clear.
Validation, plainly
Validation in progress. Here is the protocol.
- Agreement with dermatologist grading, focus signal
- To be published
- Graded images in the validation set
- To be published
- Fitzpatrick types, reported separately
- To be publishedI, II, III, IV, V, VI
- Independent clinical reviewer
- To be published
No number appears here until the study behind it exists. Publishing the protocol before the results is the honest order.
Every skin tone, read on its own terms
Redness, texture and shine sit differently on different skin tones. MIRU grades your reading against your own tone group, never against a blended average, and the validation above is reported for each Fitzpatrick type separately for the same reason. When lighting or framing makes a tone call less certain, the reading says so before it says anything else.
Where MIRU is weakest
- Dim or warm indoor light exaggerates redness. The quality gate exists because of this.
- One photo is a moment, not a trend. Two weeks apart, same light, is the honest unit of change.
- MIRU reads appearance, not pathology. Moles, lesions and anything worrying belong with a dermatologist, not an app.
Personal guidance only · For clinical concerns, see a dermatologist.