Open Access

ARTICLE

Context Encoding Deep Neural Network Driven Spectral Domain 3D-Optical Coherence Tomography Imaging in Purtscher Retinopathy Diagnosis

Anand Deva Durai Chelladurai¹, Theena Jemima Jebaseeli², Omar Alqahtani¹, Prasanalakshmi Balaji^1,*, Jeniffer John Simon Christopher³

1 Department of Computer Science, College of Computer Science, King Khalid University, Abha, 61421, Saudi Arabia
2 Division of Artificial Intelligence and Machine Learning, Karunya Institute of Technology and Sciences, Coimbatore, 641114, India
3 Department of Computer Science and Engineering, Anna University, Chennai, 600025, India

* Corresponding Author: Prasanalakshmi Balaji. Email:

(This article belongs to the Special Issue: Medical Imaging Based Disease Diagnosis Using AI)

Computers, Materials & Continua 2025, 84(1), 1101-1122. https://doi.org/10.32604/cmc.2025.062278

Received 14 December 2024; Accepted 02 April 2025; Issue published 09 June 2025

Abstract

Optical Coherence Tomography (OCT) provides cross-sectional and three-dimensional reconstructions of the target tissue, allowing precise imaging and quantitative analysis of individual retinal layers. These images, based on optical inhomogeneities, reveal intricate cellular structures and are vital for tasks like retinal segmentation. The proposed study uses OCT images to identify significant differences in peripapillary retinal nerve fiber layer thickness. Incorporating spectral-domain analysis of OCT images significantly enhances the evaluation of Purtcher Retinopathy. To streamline this process, the study introduces a Context Encoding Deep Neural Network (CEDNN), which eliminates the time-consuming manual segmentation process while improving the accuracy of retinal layer thickness measurements. Despite the excellent performance of deep learning-based Convolutional Neural Networks (CNNs) in multiclass ocular fluid segmentation and lesion identification, certain challenges remain. Specifically, segmentation accuracy declines in regions with very tiny patches of subretinal fluid, often due to limited training data. The proposed CEDNN addresses these limitations by reducing processing time and enhancing accuracy. The approach incorporates advanced diffusion techniques in the 2D segmentation process using a gradient convergence field that accounts for the anisotropic nature of image features. Experimental results on public datasets and clinical OCT images demonstrate that the CEDNN approach achieves remarkable performance, with an accuracy of 99.3%, sensitivity of 99.4%, and specificity of 99%. Furthermore, the use of 3D representations of surface data outperforms traditional 2D surface estimates, enhancing segmentation quality. The system also incorporates temporal dimension estimation, making it feasible to forecast rapid disease progression. This advanced approach holds significant potential for improving retinal disease detection and analysis, setting a new benchmark in automated OCT-based diagnostics.

---

Keywords

---