CAVRI-H5: A Homogenous OCT Dataset for Retinal Classification Featuring VMA-VMT Pathologies
Description
This database contains a total of 609 2D optical coherence tomography (OCT) images (B-scans through the macula), including 96 images in the CNV category, 99 in DME, 137 in DRUSEN, 135 in VMA-VMT, and 142 in NORMAL. The images were collected using the Heidelberg Spectralis device from patients in the Department of Ophthalmology, University Hospital in Poznań, Poznań University of Medical Sciences.
Files
Steps to reproduce
Images were included if they met the following criteria: (i) the B-scan was centred on the fovea and captured the central macular cross-section; (ii) image quality was rated as acceptable or higher by a retinal specialist, defined as a signal-to-noise ratio sufficient for clear delineation of the internal limiting membrane (ILM) and retinal pigment epithelium (RPE); and (iii) the dominant pathology could be unambiguously assigned to one of the five target classes. Images were excluded if they presented with poor signal quality (excessive noise or motion artefacts), if anatomical landmarks were obscured, or if two or more concurrent pathologies of equal clinical prominence were present, preventing single-class assignment. The process of acquisition, selection, and classification lasted 2 years. The study was conducted in accordance with the guidelines of the Declaration of Helsinki and was approved by the Medical Ethics Committee of Poznan University of Medical Sciences (resolution no. 422/14, dated May 8, 2014). The approved protocol covers the anonymised acquisition, storage, and public dissemination of OCT images for research purposes. All patients provided written informed consent prior to imaging. A total of 609 images were selected, including 96 images in the CNV category, 99 in DME, 137 in DRUSEN, 135 in VMA-VMT, and 142 in NORMAL. These images were obtained using a Heidelberg Spectralis OCT device (Heidelberg Engineering Inc.) from 609 unique patients aged 23 to 98 years, with a male-to-female ratio of 1.5:1. The Spectralis device provided raw images with varying resolutions of 505 x 490, 761 x 490, and 1529 x 490 pixels. The images were anonymized and extracted in TIFF format. The team of specialist ophthalmologists selected and categorized the subjects into appropriate classes based on the pathology. Image labelling was performed independently by three certified retinal specialists, each with five to twelve years of experience in OCT interpretation. Disagreements between graders were adjudicated by a senior vitreoretinal surgeon (co-author M.S.).
Institutions
- Poznan University of Medical SciencesGreater Poland, Poznan
- Poznań University of TechnologyGreater Poland, Poznan