Application of neural networks for improving the methods of assessment of corneal nerve fibers (preliminary report)

Processing large datasets using artificial intelligence is a promising approach in disease diagnosis and monitoring that focuses on improving research algorithms for existing technologies. Interest in studying corneal nerve fibers (CNFs) arises not only from the need to understand the pathogenesis a...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Vestnik oftal'mologii Ročník 141; číslo 2; s. 117
Hlavní autori: Avetisov, S E, Surnina, Z V, Georgiev, S
Médium: Journal Article
Jazyk:Russian
Vydavateľské údaje: Russia (Federation) 2025
Predmet:
Adult
Algorithms
Cornea - innervation
Female
Humans
Image Processing, Computer-Assisted - methods
Male
Microscopy, Confocal - methods
Middle Aged
Nerve Fibers - pathology
Neural Networks, Computer
corneal nerves
neural networks
corneal confocal microscopy
ISSN:0042-465X
On-line prístup:Zistit podrobnosti o prístupe
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Processing large datasets using artificial intelligence is a promising approach in disease diagnosis and monitoring that focuses on improving research algorithms for existing technologies. Interest in studying corneal nerve fibers (CNFs) arises not only from the need to understand the pathogenesis and progression of various ocular diseases but also from the potential for thin, unmyelinated corneal nerves to be used as biomarkers for systemic polyneuropathies. This study evaluates the preliminary results of using a neural network-based algorithm for analysis of confocal images of CNFs. The comparative study of CNFs was conducted in a group of 50 healthy volunteers (100 eyes) aged 25 to 55 years without concomitant ocular or systemic diseases. Confocal microscopy of the central cornea was performed to assess the state of CNFs. Image analysis and nerve recognition were carried out using special software (Liner calculate, Liner 1.2S) and a newly developed neural network-based algorithm. The study considered suitable encoders for image processing, including ResNet_50, VGG_16, and InceptionResNetV2. Among these, images processed with the VGG_16 encoder in Imagenet mode demonstrated the highest quality. Quantitative CNF parameters (length and density of the main trunks, macrophage count, anisotropy and symmetry coefficients) were comparable between the regular software and the neural network-based algorithm. The results indicate the potential of using neural networks, particularly the VGG_16 encoder family, for structural assessment of the CNFs. Key advantages of the proposed algorithm include improved quality of image interpretation and reduced time required for analysis.
Bibliografia:ObjectType-Article-2
SourceType-Scholarly Journals-1
ObjectType-Feature-1
content type line 23
ISSN:0042-465X
DOI:10.17116/oftalma2025141021117