Computer Vision News 46 Deep Learning for the Eyes darkest (for pRPE) and brightest (for BrM) point in each A-scan from the initial boundaries were searched for. After smoothing and manually thresholding the binarised enface image, regions with no visible hyporeflective gap were found. With the increasing age of healthy subjects, this gap was found to first become invisible in the center of the retina, the parafovea. The second part was the training of a neural network for depth map prediction. Since previous work on Bruch's membrane segmentation is available, this part is assumed to be given. The novelty of her work lay in automatically detecting the boundary of the posterior Retinal Pigment Epithelium. She trained her depth map regression network (DMR-Net) with 3D OCT volumes as an input and a 2D enface depth map (estimation of the RPE depth) as an output. Due to the BrM layer flattening preprocessing step, the depth map can be interpreted as the thickness map of the hyporeflective gap. For the final evaluation, Wenke generated average thickness maps. Having access to a range of age groups, she successfully showed the declining trend of thickness with increasing age and its spatial dependencies. Future work will set a strong focus on the diseased eyes, hopefully proving an increased hyporeflective gap in patients with age-related macular degeneration.
RkJQdWJsaXNoZXIy NTc3NzU=