Diabetic CF-FA registration Keypoints Ground truth

Citation Author(s):
Jiacheng
Wang
Vanderbilt University
Submitted by:
Jiacheng Wang
Last updated:
Tue, 11/19/2024 - 05:42
DOI:
10.21227/4hrd-p580
Research Article Link:
Retinal IPA: Iterative KeyPoints Alignment for Multimodal Retinal Imaging
License:
Creative Commons Attribution
19 Views
Categories:
Machine Learning
Keywords:
retinal fundus images; fluorescein angiography; registration;
0
0 ratings - Please login to submit your rating.
ACCESS DATASET CITE SHARE/EMBED

Abstract 

We present a publicly available dataset of keypoints for multimodal retinal image registration using Color Fundus (CF) and Fractional Anisotropy (FA) images. This dataset is derived from the diabetic retinal image dataset [1], consisting of 59 subjects (29 healthy controls and 30 diabetic patients). The CF images are captured in RGB, while FA images, highlighting microvascular structures, are provided in grayscale, both at a resolution of 720 × 576 pixels. For each subject, we manually annotated 6-8 corresponding keypoint pairs between the modalities, enabling precise alignment of retinal structures. Annotations were independently performed by two experts from Vanderbilt University Medical Center, ensuring high-quality landmark identification.

[1]. Hajeb Mohammad Alipour, S., Rabbani, H., & Akhlaghi, M. R. (2012). Diabetic retinopathy grading by digital curvelet transform. Computational and mathematical methods in medicine, 2012(1), 761901.

Instructions: 

The dataset is organized into two folders: abnormal_gt and normal_gt, corresponding to CF-FA image pairs from diabetic patients and healthy controls, respectively. Each folder contains files for individual subjects, where the keypoint annotations are stored in plain text format.

Each annotation file includes the following structure:

  • The first two columns represent the coordinates of the keypoints in the FA image.
  • The last two columns represent the corresponding coordinates of the keypoints in the CF image.
  • Each row represents one keypoint pair, aligning a point in the FA image with its corresponding point in the CF image.

This format follows the structure of the FIRE dataset [1], ensuring compatibility with common registration evaluation workflows. Detailed instructions on parsing the dataset and using the annotations for registration tasks are provided in the accompanying documentation.

Please reference the FIRE dataset documentation for examples of how to work with this format and ensure proper preprocessing when applying the dataset in your research.

[1]. Hernandez-Matas, C., Zabulis, X., Triantafyllou, A., Anyfanti, P., Douma, S., & Argyros, A. A. (2017). FIRE: fundus image registration dataset. Modeling and Artificial Intelligence in Ophthalmology, 1(4), 16-28.

Comments

This datset is served as Multimodal retinal image landmarks (keypoints) for medical image analysis registration research. 

Submitted by Jiacheng Wang on Tue, 11/19/2024 - 05:45

Dataset Files

LOGIN TO ACCESS DATASET FILES

 

QUESTIONS?