Dr. LUBHANI JAIN

Dr.Kulin J Kothari, Dr. SAUMIL K. KOTHARI, Dr. SONIA S KOTHARI

Abstract

Aim: Assess the diagnostic accuracy of an offline smartphone based artificial intelligence (AI) algorithm to detect Referable Diabetic Retinopathy (RDR) compared to grading by ophthalmologists. Methods: Observational study in which type 2 diabetics underwent 3 field fundus imaging using Remidio NM FOP 10. Images were diagnosed with diabetic retinopathy (DR) or other retinal disorders by 3 ophthalmologists. Patient-wise diagnosis was based on a consensus grading of the worse eye. Medios, an Offline AI system was used to detect RDR (moderate NPDR and higher). Results: Of 254 patients imaged, 225 met requirements for grading which were included. A 100% sensitivity & 69.19% specificity of RDR & 70.72% sensitivity & 100% specificity for no DR was noted. On excluding the 53 non-DR cases falsely identified as RDR by AI, 100% sensitivity & 96.97% specificity of RDR & 100% sensitivity & specificity for no DR was noted. Conclusions: The study highlights the usefulness of AI in DR telescreening.

Full Text