DR. ARVIND KUMAR MORYA

, Dr.RAJENDRA PRASAD, Dr. SIDDHARAM JANTI, Dr. PRIYA SISODIYA

Abstract

Purpose: Deep Learning (DL) and Artificial Intelligence (AI)now widespread due to advanced technologies&data. Supervised learning algorithm developed for an annotation tool with a customizable feature set. Evaluated viability of smartphone annotation tool.Methods:Smartphone-based grading system for grading multiple retinal fundi.Designed flow of user interface (UI) from expert feedback.Quantitative&qualitative analysis done. The dataset size200,000 images with adjudicated labels by 3 doctors.AI model validation on public datasets.

Results: DL model analysed for a binary referrable task, whether a retinal image has Referrable DR or not.32 doctors graded2000images.Analytics shown portability&flexibility. Mean89.11% in agreement.

Conclusions: Our aim was to make Annotation & Screening of Retinal imaging easier&to minimize time taken for annotations without quality degradation. Incorporation of brightness and contrast variations, green channels &zooming add-ons for best results.