Source-linked AI summary
OCTID: Optical Coherence Tomography Image Database
Peyman Gholami, Priyanka Roy, Mohana Kuppuswamy Parthasarathy, Vasudevan Lakshminarayanan
TL;DR
The paper addresses limitations in existing OCT databases, including restricted disease coverage and inconsistent image properties. It introduces OCTID, an open-access database of categorized high-resolution retinal OCT images with ground-truth segmentations and a clinician-oriented GUI. The database is intended to support classification, segmentation, and computer-aided retinal disease analysis.
Problem
Existing OCT databases may have limited images, normal-only data, single-disease coverage, or inconsistent image properties that complicate comparisons.
Method
The paper constructs OCTID with categorized high-resolution OCT images, clinician-provided segmentations for 25 normal images, and a GUI supporting manual and semi-automated segmentation.
Results
OCTID contains more than 500 spectral-domain OCT volumetric scans across Normal, Macular Hole, Age-related Macular Degeneration, Central Serous Retinopathy, and Diabetic Retinopathy categories.
Takeaways & Limitations
OCTID provides public access to disease-categorized OCT images and segmentation ground truths for developing and evaluating image-analysis and computer-aided disease-identification methods.
Takeaways & Limitations
The database is still being expanded and categorized, with the authors reporting access to more than 100,000 additional images for future incorporation.
Abstract
from arXiv · showhide
Optical coherence tomography (OCT) is a non-invasive imaging modality which is widely used in clinical ophthalmology. OCT images are capable of visualizing deep retinal layers which is crucial for early diagnosis of retinal diseases. In this paper, we describe a comprehensive open-access database containing more than 500 highresolution images categorized into different pathological conditions. The image classes include Normal (NO), Macular Hole (MH), Age-related Macular Degeneration (AMD), Central Serous Retinopathy (CSR), and Diabetic Retinopathy (DR). The images were obtained from a raster scan protocol with a 2mm scan length and 512x1024 pixel resolution. We have also included 25 normal OCT images with their corresponding ground truth delineations which can be used for an accurate evaluation of OCT image segmentation. In addition, we have provided a user-friendly GUI which can be used by clinicians for manual (and semi-automated) segmentation.
Optical Coherence Tomography (OCT)
OCT provides clinically important cross-sectional views of retinal structures, but existing databases vary in disease coverage and image properties. OCTID addresses this gap with an open-access, high-resolution database spanning retinal diseases and clinician-provided ground-truth segmentations.
- OCT enables cross-sectional visualization of retinal structures relevant to early diagnosis of retinal and optic-nerve-head pathologies.
- Existing SD-OCT databases often have limited image numbers, normal-only data, or focus on a single ocular disease.
- Differences in resolution, image quality, size, retinal field coverage, and other image properties make comparisons across databases challenging.
- A database containing multiple ocular conditions with consistent image characteristics would support more reliable research comparisons.
- OCTID provides an open-access collection of high-resolution OCT images from retinal diseases, including clinician-produced manual segmentations for evaluating image-analysis methods.
Database Images properties
OCTID provides more than 500 high-resolution OCT images across five retinal categories, using consistent raster-scan acquisition and spanning multiple disease severities. It also includes expert delineations for 25 normal images and a GUI supporting manual segmentation.
- More than 500 spectral-domain OCT scans are organized into Normal, Macular Hole, AMD, CSR, and DR categories.
- Images were acquired using a 2 mm raster scan with 512x1024 pixels on a Cirrus HD-OCT system.
- Each disease dataset includes less severe, medium severe, and more severe stages, with severity randomized across images.
- The database includes 25 expert-delineated normal images with ground-truth boundary files for segmentation evaluation.
- A GUI enables clinicians to select retinal boundaries and manually or semi-automatically re-segment them.
- The database provides 102 MH, 55 AMD, 107 DR, and 206 NO retinal images through public DOI-linked datasets.
3. Future Work and Conclusion
The paper presents OCTID as an open-access resource intended to support computer-aided OCT analysis, disease classification, and segmentation research. Future work will expand the database using more than 100,000 additional images under examination and categorization.
- OCTID provides public high-resolution pathological OCT images to support computer-aided analysis, deep-learning classification, and segmentation studies.
- Future work will expand OCTID by examining and categorizing more than 100,000 additional images.