Communication Dans Un Congrès Année : 2014

Automatic detection of the retinal neovascularization in diabetic patients.

Résumé

Purpose: Diabetes is a major pathology of which the prevalence is increasing worldwide. The retinal neovascularization related to diabetes is a major cause of blindness. Automatic evaluation of the extent of retinal neovascularization on ultra wide field imaging is actually not accurate. This study aimed to develop an automatically detection of retinal neovascularization including quantitative analyses during angiographic phases (surface and number of new vessels).
Methods: A multimodal imaging platform allowing an active eye tracking and ultra-wide field imaging of 105 degrees was used. Ten patients with retinal neovascularization observed during the angiographic phases were included in the analysis. The images have been automatically processed to detect the retinal neovascularization using an algorithm that included anisotropic smoothing, luminance analysis, region fusion, opening and automatic reframing between different phases of angiograms. Retinal neovascularization were computed and compared between images acquired during the early and the late phases of acquisition.
Results: This algorithm has been tested on 9 angiographic sequences from  different patients with proliferative diabetic retinopathy on the Spectralis platform of Heidelberg Engineering GmbH. When compared to manual analysis,  of the 50 neo vessels manually detected have been automatically detected (89.91%). 48.425 mm2 of the 51.616 mm2 (total surface of the neo vessels manually detected) have been automatically detected (92.54%). A mean of less than 1 false alarm (0.88) have been detected and for the total segmented surface (56. mm2).
Conclusions: This software provides surface and number of retinal new vessels analyses and will be able to calculate a diffusion index of the retinal new vessels. Development convivial version of this software will provide ophthalmologists quantitative data to follow diabetic patients with proliferative diabetic retinopathy.

arvos-2014-Beouche.pdf (49.75 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-00949305 , version 1 (17-04-2025)

Identifiants

  • HAL Id : hal-00949305 , version 1

Citer

Benjamin Béouche-Hélias, David Helbert, Bruno Mercier, Philippe Carré, Christine Fernandez-Maloigne, et al.. Automatic detection of the retinal neovascularization in diabetic patients.. ARVO 2014 Annual Meeting : Leading Eye and Vision Research, Association for Research in Vision and Ophthalmology (ARVO), May 2014, Orlando, United States. pp.207 - B0015. ⟨hal-00949305⟩
234 Consultations
13 Téléchargements

Partager

  • More