Automatic Algorithm-Aided Segmentation of Retinal Nerve Fibers Using Fundus Photographs

This work presents an image processing algorithm for the segmentation of the personalized mapping of retinal nerve fiber layer (RNFL) bundle trajectories in the human retina. To segment RNFL bundles, preprocessing steps were used for noise reduction and illumination correction. Blood vessels were re...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Journal of imaging Ročník 11; číslo 9; s. 294
Hlavný autor: Lujan Villarreal, Diego
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Switzerland MDPI AG 28.08.2025
Predmet:
Algorithms
Artificial intelligence
Blood vessels
Comparative analysis
course of axons
Curves
Customization
Deep learning
Diabetic retinopathy
Fovea
Fundus oculi
Glaucoma
Ground truth
Identification and classification
Image processing
image processing segmentation
Image segmentation
Mapping
Mean square errors
Methods
Nerves
Noise control
Observations
Optic nerve
optic pathways
Orientation
Physiological aspects
retinal nerve fiber layer
retinal nerve fiber trajectory
Visual fields
visual pathways
Japan
Mexico
retinal nerve fiber trajectory
image processing segmentation
visual pathways
optic pathways
retinal nerve fiber layer
course of axons
ISSN:2313-433X, 2313-433X
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:This work presents an image processing algorithm for the segmentation of the personalized mapping of retinal nerve fiber layer (RNFL) bundle trajectories in the human retina. To segment RNFL bundles, preprocessing steps were used for noise reduction and illumination correction. Blood vessels were removed. The image was fed to a maximum–minimum modulation algorithm to isolate retinal nerve fiber (RNF) segments. A modified Garway-Heath map categorizes RNF orientation, assuming designated sets of orientation angles for aligning RNFs direction. Bezier curves fit RNFs from the center of the optic disk (OD) to their corresponding end. Fundus images from five different databases (n = 300) were tested, with 277 healthy normal subjects and 33 classified as diabetic without any sign of diabetic retinopathy. The algorithm successfully traced fiber trajectories per fundus across all regions identified by the Garway-Heath map. The resulting trace images were compared to the Jansonius map, reaching an average efficiency of 97.44% and working well with those of low resolution. The average mean difference in orientation angles of the included images was 11.01 ± 1.25 and the average RMSE was 13.82 ± 1.55. A 24-2 visual field (VF) grid pattern was overlaid onto the fundus to relate the VF test points to the intersection of RNFL bundles and their entry angles into the OD. The mean standard deviation (95% limit) obtained 13.5° (median 14.01°), ranging from less than 1° to 28.4° for 50 out of 52 VF locations. The influence of optic parameters was explored using multiple linear regression. Average angle trajectories in the papillomacular region were significantly influenced (p < 0.00001) by the latitudinal optic disk position and disk–fovea angle. Given the basic biometric ground truth data (only fovea and OD centers) that is publicly accessible, the algorithm can be customized to individual eyes and distinguish fibers with accuracy by considering unique anatomical features.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
content type line 23
ISSN:2313-433X
2313-433X
DOI:10.3390/jimaging11090294