Influence of Histogram Equalization on Multi-Classification of Retinal Diseases in OCT B-scans

• Published in: Signal Processing Algorithms, Architectures, Arrangements, and Applications Conference proceedings pp. 102 - 108
• Main Authors: Marciniak, Tomasz, Stankiewicz, Agnieszka
• Format: Conference Proceeding
• Language: English
• Published: Division of Signal Processing and Electronic Systems, Poznan University of Technology (DSPES PUT) 25.09.2024
• Subjects: Adaptive equalizers; Artificial neural networks; classification; Diseases; histogram equalization; Histograms; image processing; machine learning; Optical coherence tomography; optical coherence tomography (OCT); Pathology; Performance evaluation; Retina; retinal diseases; Signal processing algorithms; VGG
• ISSN: 2326-0319
• DOI: 10.23919/SPA61993.2024.10715595

The aim of this article is to analyze the impact of adaptive histogram equalization of the human eye OCT B-scans on the effectiveness of the classification of pathological changes. Tests were performed on two datasets. The first dataset contains 4 class B-scans (CNV, DME, DRUSEN, and NORMAL) taken with the use of Spectralis Heidelberg device, while the second dataset contains images from two devices (Spectralis Heidelberg OCT and Avanti Optovue OCT) and additionally contains the VMT (i.e., vitreomacular traction) class. The images were subjected to standard histogram equalization and contrast limited adaptive histogram equalization (CLAHE) with various parameters. The classification process was carried out using VGG16 artificial neural network. It can be observed that histogram equalization has no significant impact on improving classification performance.