Chief & Presenting Author: Dr.Markandeya Singh

Co Author(s): Dr. Deepanjan Ghosh

Abstract

Study design: Analytical
Purpose: Create a deep learning model to help identify (DR)
Methods: We derived an algorithm by developing a deep learning model using fundus images of diabetic patients from an open source dataset. The model was trained to look for changes associated with DR and classify the images into those with DR and those without. This was then compared with some previously developed models for validation of the results.
Results: The deep learning model was able to classify the images shown to it with great accuracy, sensitivity and specificity and the results were comparable to models developed earlier and diagnosed cases from other databases.
Conclusion: The model showed great efficacy in classifying the images presented to it into those with and without DR. Current diagnosis involves examination by clinicians by funduscopy. This modality could help in comparable accurate diagnosis in lesser time. Similar studies have been done elsewhere but few have been done in India