Diabetic Retinopathy, Glaucoma
Conditions
Keywords
Artificial Intelligence, Validation, Screening
Brief summary
Artificial Intelligence (AI) algorithms require validation in a variety of populations to ensure widespread clinical applicability. In Ophthalmology, AI algorithms are reaching maturity in diagnosis such as diabetic retinopathy and glaucoma. Higher-at-risk subjects of African descent are nevertheless usually under-represented in training datasets and therefore unclear about representativity. A small scale validation study in consecutive patients in a large Eyesore unit in Mozambique will be performed to determine the diagnostic ability of these AI softwares in this population
Detailed description
Artificial Intelligence (AI) algorithm's are the next frontier in medical management, usually meant to improve diagnostic capabilities and to optimize the existing resources. They are particularly relevant in settings where there is a lack of specialised Human Resources such as physicians. Ensuring these algorithms can be used in a wide population is therefore crucial to clinical implementation. Validation studies in specific segments of populations are needed to ensure all patients are represented and the results are therefore reliable. Higher-at-risk subjects of African descent are nevertheless usually under-represented in training datasets and therefore unclear about representativity. A pilot study for validation of an AI algorithm for Glaucoma and Diabetic Retinopathy will be done for the MONA G-RISK® and diabetic retinopathy. Consecutive patients from a large Eye Unit in Mozambique's capital will be screened using these AI algorithms and validated using clinical standard as ground truth.
Interventions
DIAGNOSTIC_TESTFundus Picture AI testing
G-Risk AI algorithm will assess the optic disc centered fundus picture and determine whether or not there is a need for referrable based on a pre-determined threshold (\>=0.73)
Sponsors
Dr. Agarwal's Eye Hospital
Centro Hospitalar Universitário Lisboa Norte
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
DIAGNOSTIC
Masking
NONE
Intervention model description
Single arm. Consecutive patients in a general ophthalmology consultation will be offered intervention
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
Yes
Inclusion criteria
* subjects age above 18 years old presenting at the Eye Unit * willingness to sign an informed consent for the screening process
Exclusion criteria
* none * Poor quality in screening image will be included in the intention to treat analysis, but excluded from the diagnostic comparator outcome. * Patients with a known glaucoma diagnosis will not be excluded from the screening
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic agreement between referring decision and reading center decision | Duration of the study - 3 weeks | Level of agreement will be done between referring decision and the ground truth as assessed by the reading center (normal, glaucoma suspect; definitive glaucoma). All subjects from both centers (referred and non-referred) will be reviewed. For a primary outcome analysis, the middle category (glaucoma suspect) will be pooled together with the normal diagnosis |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Level of agreement (in %) between AI-risk score and human-based assessment of disease severity | After the study - 6 months | Reading center risk score of disease severity (ranked from 0 to 100) will be compared to the AI-based disease score. This will be done separately in each of the 3 categories (normal; glaucoma suspect; glaucoma). Analysis of this score would help refine clinical risk (high risk vs low risk patients) of each category. Exploratory analysis will be made to determine the added value of including this risk score in refining AI-based referral |
Contacts
Primary ContactLuis Abegao Pinto, MD, PhD
Backup ContactQuirina Tavares Ferreira, PhD
Outcome results
None listed