Deep Learning-Based Classification of Inherited Retinal Diseases Using Fundus Autofluorescence.

Background. In recent years, deep learning has been increasingly applied to a vast array of ophthalmological diseases. Inherited retinal diseases (IRD) are rare genetic conditions with a distinctive phenotype on fundus autofluorescence imaging (FAF). Our purpose was to automatically classify different IRDs by means of FAF images using a deep learning algorithm. Methods. In this study, FAF images of patients with retinitis pigmentosa (RP), Best disease (BD), Stargardt disease (STGD), as well as a healthy comparable group were used to train a multilayer deep convolutional neural network (CNN) to differentiate FAF images between each type of IRD and normal FAF. The CNN was trained and validated with 389 FAF images. Established augmentation techniques were used. An Adam optimizer was used for training. For subsequent testing, the built classifiers were then tested with 94 untrained FAF images. Results. For the inherited retinal disease classifiers, global accuracy was 0.95. The precision-recall area under the curve (PRC-AUC) averaged 0.988 for BD, 0.999 for RP, 0.996 for STGD, and 0.989 for healthy controls. Conclusions. This study describes the use of a deep learning-based algorithm to automatically detect and classify inherited retinal disease in FAF. Hereby, the created classifiers showed excellent results. With further developments, this model may be a diagnostic tool and may give relevant information for future therapeutic approaches.

Automated quantification of choroidal neovascularization on Optical Coherence Tomography Angiography images.

OBJECTIVES: To report the design of an automated quantification algorithm for choroidal neovascularization (CNV) in the context of neovascular age-related macular degeneration (AMD), based on Optical Coherence Tomography Angiography (OCTA) images. MATERIAL AND METHODS: In this study, 54 patients (mean age 75.80 ± 14.29 years) with neovascular AMD (type 1 and type 2 CNV) were included retrospectively and separated into two groups (Group 1-24 images; Group 2-30 images), according to the lesion topology. All patients underwent a 3 × 3 mm OCTA examination (AngioVue, Optovue, Freemont, California). The proposed algorithm is based on segmentation and enhancement methods including Frangi filter, Gabor wavelets and Fuzzy-C-Means Classification. Our results were compared to the manual quantifications given by the embedded quantification software "AngioAnalytics". RESULTS: Automated CNV segmentation and quantification of three neovascular AMD biomarkers: the total vascular area (TVA), the total area (TA) and the vascular density (VD) were possible in all cases. Automated versus manual quantification comparison revealed a statistically significant difference for TVA and VD measurements for both groups (p = 0.00036 for Group 1 TVA, p < 0.0001 for Group 1 VD and Group 2 TVA and VD). The difference in TA measurements was not significant in Group 2 (p = 0.143). Bland-Altman analysis revealed low inter-method bias for TA measurements and higher bias for TVA and VD. CONCLUSION: This paper presents a method for segmenting and quantifying CNV that constitutes a valid option for clinicians. Complementary validations have to be carried out to compare our method's accuracy to "AngioAnalytics".