Purchase this article with an account.
Pedro Guimaraes, Pedro M. Rodrigues, Rui Bernardes, Pedro Serranho; Vascular Network of the human macula from high-definition OCT. Invest. Ophthalmol. Vis. Sci. 2012;53(14):4108.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To compute the vascular network of the human macula from high-definition optical coherent tomography (HD-OCT) to the level of detail of that of color fundus photography.
Macular cube protocol scans of 512x128x1024 and 200x200x1024 voxels of 6 eyes from 5 type 2 diabetic patients and 10 eyes from 10 healthy volunteers were collected from the Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA, USA) database. Additionally, color fundus photographs (CFP) and fluorescein angiograms (FA) from the patients' eyes were gathered from the imaging database. OCT reflectivity data was exported by the Cirrus Review Software to be processed and analyzed. Three distinct fundus references were computed from the OCT volumetric data per eye scan after proper filtering, alignment and processing. An additional OCT fundus image (OCTref) is computed as the average of these 3 OCT fundus references. CFP, FA and OCTref images were manually segmented to obtain the visible vascular network for each of the imaging modalities. Finally, a support vector machine (SVM) pattern classification algorithm was used to classify each of the OCTref image pixels into vessel and non-vessel classes from the 3 computed fundus references and from a set of features computed from the OCTref image.
Over 64% of the vascular network manually segmented from the FA was rmanually segmented from the CFP, while this percentage raises to over 71% for OCT, 64.19(8.59)% and 71.61(8.98)%, respectively (average(SD)). In this way, the computed fundus reference image from the HD-OCT (OCTref) allows to compute an extended set of the vascular network as compared to the CFP (109.44(19.92)%). When comparing the automatic classification (SVM) versus the manually segmented OCT vascular network, a specificity of 79.1(4.4)% and a sensitivity of 99.3(0.6)% were obtained. Overall, the accuracy of the automatic classification in detecting the vascular network from HD-OCT data is of 97.4(0.6)%.
The proposed algorithm allows for the segmentation of the vascular network from HD-OCT scans of the ocular fundus to a level similar to that of color fundus photography.
Clinical Trial: :
This PDF is available to Subscribers Only