Hybrid method and CNN regressor for OCT images to predict visual acuity of retinal occlusion patients after 12 months of treatment

RVO is the second most common retinal vascular disorder. The use of anti-vascular endothelial growth factor (Anti-VEGF) has been established as the standard of care in cases of RVO; however, many studies show that not all visual gains are maintained beyond the first year and patient responses to anti-VEGF vary. Additionally, the injection carries a risk of sight loss and is very expensive. Although it is yet to be investigated, there would be substantial clinical utility in a means to predict precisely which patients will benefit from injections. This work proposes a novel CNN architecture for prognosis prediction that can assist clinicians in making decisions regarding RVO treatment. A hybrid method using HOG, BLBPs, and GLCM to extract features from OCT images are examined. The regression system is studied by employing separate local feature representations of OCT images. These features are then combined to improve the recognition rate. The hybrid feature provides a good representation of OCT images to complement EMRs and is input to our CNN regressor which foresees VA after one year of treatment. The performance of the system was evaluated by comparison with 11 different specialist ophthalmology registrars, and 1 specialist retina consultant. The mean absolute error using the proposed method suggested equivocal performance with a trend towards model superiority.