Diabetic Retinopathy via Hypertensive Modeling

This interdisciplinary study develops a novel prognostic for Diabetic Retinopathy (DR) by modeling hypertensive changes at arterial-venous (AV) crossings

Summary of Research

This interdisciplinary study develops a novel prognostic for Diabetic Retinopathy (DR) by modeling hypertensive changes at arterial-venous (AV) crossings, ultimately discovering new symptoms (venous compression, deflection, and tortuosity) to diagnose and monitor the condition earlier.

In unison, the research develops a low-cost, portable device for rapid and automated diabetic retinopathy stage classification through Random Forest (RF) learning ensembles and computer vision. Annually, 39 million people are blinded by the late identification of diabetic-retinopathy. The developed system can help reduce these blindness rates through low-cost, rapid, automated screens. By exploring correlations with Hypertensive Retinopathy the paper goes on to propose a novel Blood Pressure Quantification (BPQ) metric capable of modeling AV interactions for early detection of retinopathy, with state-of-the-art 78% prognostic ML model accuracies. The developed PCA-derived lesion classifier and feature generation mechanisms seed a stochastic RF module for stage classification of DR images, receiving 81% cross-validation accuracy (trained from 3,000 retinal images) comparable to a human gold standard of 90%. Fundus biomicroscopy with an aspheric condensing lens attached to a smartphone camera interfaces with the computer-vision approach to image retinas with 40-degree FOV’s and algorithmically conduct prognostics. Full-scale results are returned within minutes on an app compared to the current 6-week screening time.

Leave a Reply

Diabetic Retinopathy via Hypertensive Modeling

This interdisciplinary study develops a novel prognostic for Diabetic Retinopathy (DR) by modeling hypertensive changes at arterial-venous (AV) crossings

Scroll to top