fidoptDuke intermediate AMD SD-OCT, retina and RPEDC
The B-scan through the fovea of a macular SD-OCT volume of an eye with intermediate age-related macular degeneration or of a control, with the retina and the retinal pigment epithelium and drusen complex marked (Farsiu et al. 2014, Duke).
Read before comparing (7)
- Pool. The page's 384 subjects, 269 with intermediate age-related macular degeneration and 115 controls, one Bioptigen SD-OCT volume of the macula each (100 B-scans of 1000 A-scans).
- Cases. The B-scan through the fovea: the one with the most A-scans marked, B-scan 48 to 52 of 100 on the test cases. 1000 by 512 pixels, 8-bit, as delivered; the files give no pixel size, so distances are in pixels.
- Truth. Graded semi-automatic: the authors' software traced the three boundaries and certified graders corrected them. The two layers between them are filled by prepare.py, each boundary rounded to the nearest row. Only the central 5 mm is marked: the A-scans outside it (23 to 29% of a B-scan) are ignored.
- Strictness. The RPEDC is about 10 pixels thick at the median: moved down by 2 pixels it keeps a Dice of 0.806 (0.729 to 0.971), the retina 0.976.
- Start. Labelling the whole marked region as each layer scores a Dice of 0.277 (retina) and 0.039 (RPEDC) at the median of the test cases, an Otsu mask 0.398 and 0.087. TotalSegmentator has no OCT model.
- Overlap. No other dataset of the corpus holds an OCT image; Duke's other OCT sets (DME, Chiu 2015) are other patients and are not in it.
- Licence. No licence is named: research and educational use only, no commercialisation or redistribution of the images, citation asked. Research board only, never uploaded.
No method has a result on every case of this dataset yet.
Case by case
Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- No split is published. 30 test subjects drawn with seed 2026 among the 384, AMD and controls in proportion (21 and 9), and 30 development subjects; one B-scan a subject.
- Truth
- contours by a program, verified by people the layers between the deposit's three boundaries on the B-scan: retina from the inner limiting membrane to the inner aspect of the RPEDC, RPEDC from there to the outer aspect of Bruch's membrane; the boundaries were segmented by the authors' software, then corrected by certified graders of the Duke Advanced Research in SD-OCT Imaging laboratory, within 5 mm of the fovea
- Source
- people.duke.edu/~sf59/RPEDC_Ophth_2013_dataset.htm
duke.box.com/s/cm7esb9y5oat8an1n7hpusrq05bqygza
duke.box.com/s/8jmt4wburj56s8y3ipt5urx65u8jhlqn
doi.org/10.1016/j.ophtha.2013.07.013 - Licence
- none named: the dataset page allows use in research and educational settings, prohibits commercialisation and redistribution of the images, and asks for citation
30 cases for tuning, never tested
AMD_1069, Control_1074, AMD_1254, AMD_1225, AMD_1097, Control_1066, AMD_1176, AMD_1179, Control_1005, AMD_1060, AMD_1041, Control_1038, AMD_1257, AMD_1258, AMD_1086, Control_1003, AMD_1049, AMD_1126, Control_1008, AMD_1244, AMD_1227, Control_1097, AMD_1189, AMD_1208, AMD_1138, Control_1095, AMD_1048, AMD_1079, Control_1067, AMD_1065