Exploring the quantitative analysis of fundus features in FEVR using Deep Neural Networks
Objective Deep neural networks(DNNs)were used to quantitative analysis the abnormal features exhibited in posterior pole reti-na images in newborns with mild familial exudative retinopathy(FEVR).Methods Using a total of 15,370 posterior polar retinal images from 7685 full-term infants,we constructed a dataset of posterior polar retinal images for full-term infant.Based on the datasets,our system was trained,which also extracted and quantified the vessel density,vessel tortuosity,diameter ratio,and the ratio of the disc-to-macula(DM)distance to the disc diame-ter(DD).Results In the comparison set,significant anatomic differences between Mild FEVR and Normal were identified.In Mild FEVR,vessel density was larger[12.9666±0.9637(%)vs.12.0053±0.9246(%)],vessel tortuosity was smaller[(1.9098±0.2647(×104/cm3)vs.3.3767±0.2345(×104/cm3)],the maximum diameter ratio was higher(1.8796±0.1789 vs.1.5075±0.1644),and the DM/DD ratio(3.2676±0.2638 vs.2.8199±0.3286).All values were statistically different(P<0.005).Conclusion Quantitative analysis of the abnormal fundus features with mild FEVR is of high significance in clinical.It could help ophthalmologists provide more clues in the diagnosis and prediction of the disease.
Deep Neural NetworksMild Familial Exudative VitreoretinopathyFull Term Infants
万方数据
维普
