Sign in
Tympanic membrane screening support
Use the Google account your administrator registered.
Authorised clinical staff only. Activity is recorded.
Tympanic membrane screening support
Use the Google account your administrator registered.
Authorised clinical staff only. Activity is recorded.
Your account has a temporary password. Choose your own before continuing.
Screening recommendation
Most likely finding
Differential
Screening aid only. Not a clinical diagnosis. Findings require review by a qualified ENT practitioner before any treatment decision.
Select an image or drop it here
JPG, PNG, BMP, TIFF or WebP, up to 15 MB
The image and result are saved to this patient's examination record.
No analysis yet.
Enter the patient details, select an image and run the analysis.
Analysing image
Normalising the field of view and evaluating the image.
Tinted areas contributed most to the finding. Attention on the tympanic membrane or canal is expected; attention concentrated on glare, the image edge or debris unrelated to the finding is a reason to distrust it.
| Date | Record | Patient | Ear | Model finding | Recommendation | Review | Clinician |
|---|
Measured on this clinic's patients, taking each reviewing surgeon's verdict as the true diagnosis. This is the evidence that matters for local use, and it becomes more precise with every review.
| Measure | All reviewed cases | Recent cases | As given at the time |
|---|
| Run | Model | Cases | Surgeon agreed | Sensitivity | Sensitivity, recent | Result |
|---|
| Examined | Surgeon's diagnosis | Model's finding | Abnormal probability |
|---|
Every day the model in service is run again on every case a surgeon has reviewed, and its answers are compared with the surgeons' verdicts. Cases a model was retrained on are left out of its figures. The backtest never changes the model; it shows whether performance is holding as patients, devices and models change.
| Condition | Reviewed cases | Recognised | Correct when predicted |
|---|
| Source | Reviewed cases | Surgeon agreed | Referral sensitivity |
|---|
Accuracy can differ between devices. If one camera or endoscope does markedly worse, check its focus, lighting and resolution before relying on it.
They sign in with this email address and must choose their own password at first sign in.
| Name | Email address | Role | Status | Last sign in |
|---|
Reviewed cases are added to the original training data and a candidate model is trained. It goes live only if it is at least as good as the current model on the frozen test set and on held out local cases, and only when an administrator approves it.
Current run
To remove a single examination, for example one recorded against the wrong patient, open it from History and use Delete this examination.
Permanently deletes every patient, examination, review and stored image, and the backtest results that describe them. Use it to remove test data before the system is used for real. It cannot be undone.
| Time | User | Action | Item | From |
|---|
Permanently deletes activity entries. The log is the record of who did what, so this cannot be undone.