Abstract
Abstract
Purpose : Optical Coherence Tomography Angiography (OCTA) is a novel imaging technique that to study retinal vasculature patterns in detail. OCTA presents open challenges in the interpretation of some angiographic patterns. Over the past decade, Deep learning has been used successfully for pattern recognition, signal processing, and statistical analysis. Additionally, there has been increased interest in deep learning applied to medical imaging based on Convolutional Neural Networks (CNN). The purpose of this study was to determine the accuracy of CNN to calculate nine capillary patterns on OCTA images
Methods : A cross-sectional study with OCTA image was designed for this research. A novel end-to-end Deep Learning model based on CNN was used for automatical analysis of nine parameters related to OCTA images. The proposed model used as an input OCTA images (superficial retinal circulation 3 x 3 mm scan size). The total experimental dataset contained 100 OCTA images. Two random split datasets were chosen as follows, 70 OCTA images for training and 30 OCTA images for test set. Whole image capillarity, foveal capillarity (1 mm central circle), parafoveal capillarity (3 mm central circle), upper half capillarity, lower half capillarity, temporal parafoveal capillarity, superior parafoveal capillarity, nasal parafoveal capillarity and inferior parafoveal capillarity patterns were automatically analyzed and estimated by the CNN
Results : The experimental results showed that the proposed model is able to estimate the whole image capillarity, foveal capillarity, parafoveal capillarity, upper half capillarity, lower half capillarity, temporal parafoveal capillarity, superior parafoveal capillarity, nasal parafoveal capillarity and inferior parafoveal capillarity with a mean absolute percentage error of 2.6%, 19%, 3.7%, 5%, 4.9%, 3.8%, 7%, 6% and 5.5% respectively
Conclusions : CNN showed an outstanding performance measuring capillarity patterns in OCTA images. Deep learning methods have the ability to find and interpret associated features for measuring capillarity in parts of the OCTA image. Future studies with a major number of OCTA images, different retinal conditions and parameters are required to validate the model in the parameter prediction task that could be used clinically
Purpose : Optical Coherence Tomography Angiography (OCTA) is a novel imaging technique that to study retinal vasculature patterns in detail. OCTA presents open challenges in the interpretation of some angiographic patterns. Over the past decade, Deep learning has been used successfully for pattern recognition, signal processing, and statistical analysis. Additionally, there has been increased interest in deep learning applied to medical imaging based on Convolutional Neural Networks (CNN). The purpose of this study was to determine the accuracy of CNN to calculate nine capillary patterns on OCTA images
Methods : A cross-sectional study with OCTA image was designed for this research. A novel end-to-end Deep Learning model based on CNN was used for automatical analysis of nine parameters related to OCTA images. The proposed model used as an input OCTA images (superficial retinal circulation 3 x 3 mm scan size). The total experimental dataset contained 100 OCTA images. Two random split datasets were chosen as follows, 70 OCTA images for training and 30 OCTA images for test set. Whole image capillarity, foveal capillarity (1 mm central circle), parafoveal capillarity (3 mm central circle), upper half capillarity, lower half capillarity, temporal parafoveal capillarity, superior parafoveal capillarity, nasal parafoveal capillarity and inferior parafoveal capillarity patterns were automatically analyzed and estimated by the CNN
Results : The experimental results showed that the proposed model is able to estimate the whole image capillarity, foveal capillarity, parafoveal capillarity, upper half capillarity, lower half capillarity, temporal parafoveal capillarity, superior parafoveal capillarity, nasal parafoveal capillarity and inferior parafoveal capillarity with a mean absolute percentage error of 2.6%, 19%, 3.7%, 5%, 4.9%, 3.8%, 7%, 6% and 5.5% respectively
Conclusions : CNN showed an outstanding performance measuring capillarity patterns in OCTA images. Deep learning methods have the ability to find and interpret associated features for measuring capillarity in parts of the OCTA image. Future studies with a major number of OCTA images, different retinal conditions and parameters are required to validate the model in the parameter prediction task that could be used clinically
| Original language | English (US) |
|---|---|
| Journal | Investigative Ophthalmology & Visual Science |
| Volume | 59 |
| Issue number | 9 |
| State | Published - Jul 2018 |