Abstract
Purpose :
The aim of our study was to evaluate sensitivity and specificity of OCTA en face images for detection of CNV using software-supplied preset slabs compared to manually modified slabs as well as additional use of OCTA B-Scans with flow information.
Methods :
In this prospective observational study, 74 eyes of 53 patients with subretinal hyperreflective material (SHRM) and/ or pigment epithelial detachment (PED) on optical coherence tomography (OCT) in at least one eye underwent OCTA imaging (Spectralis). Fluorescein angiography (FA) was performed at the same day.
Automatically by the software provided avascular complex slabs were exported. Furthermore, custom slab 1 (2) was created by manual positioning of the automatically provided Bruch’s membrane boundary (transverse lines) anterior as well as posterior to any SHRM and/ or PED. For custom slab 3, manual delineation of segmentation lines in case of errors was performed.
Presence of CNV on OCTA en face images was evaluated by two independent graders masked to all other images of the patient. After this, a combination with structural OCTA B-Scans with flow information was evaluated.
Sensitivity and specificity for CNV detection on OCTA images were calculated using FA as reference. Additionally, intergrader reproducibility was assessed.
Results :
A total of 74 eyes of 53 consecutive patients were included with age-related macular degeneration (AMD) (n=29) and other diseases (n=41). 4 eyes were healthy. CNV was diagnosed on FA in 41% of eyes. SHRM or PED was present on SD-OCT in 78%.
Sensitivity (specificity) was 60% (93%) for the automatically generated avascular complex slab, 63% (93%) for custom slab 1, 63% (95%) for custom slab 2 and 60% (91%) for custom slab 3. Sensitivity was highest with 87% (80%) when combining the avascular complex en face slab with structural OCTA B-scan and flow information.
Intergrader reproducibility was >0.8 for all slabs. Concordance with FA as reference was lowest with 0.53 for the avascular complex slab and highest with 0.65 for the combination of OCTA en face images with structural OCTA B-Scans.
Conclusions :
A combination of OCTA en face images generated using manually correction of the position of segmentation lines as well as OCTA B-Scans with flow information is recommended for identification of CNV on OCTA.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.