Hypertensive Retinopathy
Conditions
Keywords
convolutional neural network, hypertensive retinopathy
Brief summary
The current study is aimed at estimating the diagnostic effectiveness of a developed neural network "RetinAIcheck" in grading the severity of hypertensive retinopathy in patients of the Russian population. The training data set was obtained from an open source and relabeled by seven independent retina specialists, the sample size was 30,000 fundus photographs. The test sample included 755 patients (1374 eyes). Among the 1.374 eyes, 94 were without HR (class 0), 330 had class 1, 660 had class 2, 280 had class 3, and 10 had class 4 HR.The reference standard was the result of independent grading of HR stage by two ophthalmologists, controversial clinical cases were evaluated with the involvement of a third ophthalmologist.
Interventions
DIAGNOSTIC_TESTConvolutional neural network "RetinAIcheck"
A convolutional neural network is a medical decision support system that processes digital fundus photographs obtained during mydriasis and determines the probability of the presence/absence of hypertensive retinopathy and it's grading due to Keith Wagener Barker's classification.
Sponsors
I.M. Sechenov First Moscow State Medical University
Study design
Observational model
OTHER
Time perspective
CROSS_SECTIONAL
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
\- Patients with and without a diagnosis of arterial hypertension, based on medical records
Exclusion criteria
* anophthalmia, * optic nerve atrophy, * eyeball injuries, * age-related macular degeneration, * central serous chorioretinopathy, * central serous chorioretinitis, * clouding of the optical media of the eye, which affects the quality of the image.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Accuracy | The ability to correctly identify the presence or absence of condition | The ability of a test to correctly identify the proportion of true positive cases |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity | February 2026 | The ability of a test to correctly identify the proportion of true positive cases |
| Specificity | February 2026 | The ability of a test to correctly identify the proportion of true negative cases |
| Positive predictive value | February 2026 | The probability that a person who tests positive for the condition actually has that one |
| Negative predictive value | February 2026 | The probability that a person who tests negative for the condition truly does not have it |
| AUROC, area under the ROC curve (one-versus-rest) | February 2026 | An average metric used to evaluate multi-class classification models by computing the Area Under the ROC Curve for each class separately against all other classes and then averaging the results |
| Quadratically weighted kappa | February 2026 | A statistical measure that evaluates the level of agreement between two raters or outcomes on an ordinal scale, penalizing errors based on the squared distance between categories |
Countries
Russia
Contacts
PRINCIPAL_INVESTIGATORPhilipp Yu Kopylov, Prof.
Sechenov First Moscow State Medical University (Sechenov University)
Outcome results
None listed