Maculopathy
Conditions
Keywords
macular diseases, screening, OCT, Deep learning, AI
Brief summary
The investigators expect to develop an algorithm that can interpret OCT images and automated determine whether the macula is normal or not by using OCT image-based deep learning techniques. And investigators wish to develop software applications that will help better screen and diagnose macular diseases in resource-limited areas.
Detailed description
The investigators will apply deep learning convolutional neural network by using ImageNet for an automated detection of multiple retinal diseases with OCT horizontal B-scans with a high-quality labeled database. Datasets, including training dataset, testing dataset and validation datasets, will be built by ophthalmologists of the First affiliated hospital of Nanjing Medical University according to the standardized annotation guidelines.
Interventions
None listed
Sponsors
The First Affiliated Hospital with Nanjing Medical University
Study design
Observational model
OTHER
Time perspective
CROSS_SECTIONAL
Eligibility
Sex/Gender
ALL
Inclusion criteria
* All patients attending the Ophthalmology Department of the First Affiliated Hospital of Nanjing Medical University within 5 years and who received known, clear diagnoses with digital retinal imaging (including OCT, fundus digital photographs and fundus fluorescein angiography, at least with OCT images) as part of their routine clinical care, will be eligible for inclusion in this study.
Exclusion criteria
* Hardcopy examinations (i.e., photos of paper reports of OCT imaging performed at other hospitals) will be ineligible. * Data from patients who have previously manually requested that their data should not be shared, even for research purposes in anonymised form, and have informed the Ophthalmology Department of the First Affiliated Hospital of Nanjing Medical University of this desire (even in previously conducted studies or other on-going studies in this hospital), will be excluded, and their data will not be upload to the cloud platform before research begins. * Data from eyes tamponed with silicone oil or gas (i.e., C3F8) will be ineligible. * Data with poor image quality, such as incomplete images, inverted images, blurred or cracked images and images with a very weak signal (i.e., vitreous haemorrhage), will be ineligible.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| receiver operating characteristic(ROC) curve of the algorithm | approximately 1 year | It is also called sensitivity curve. The ROC curve shows how sensitive the algorithm model is to automatically detect the desired output. |
| Area under the ROC curve(AUC) | approximately 1 year | It shows the operating value of the algorithm model, which can represent the effect of the model. |
Countries
China
Outcome results
None listed