45 datascEYEnce! Computer Vision News Moving on, Wenke did her master’s thesis at the Pattern Recognition Lab at FAU which has a close collaboration with the Biomedical Optical Imaging and Biophotonics Group at MIT and the New England Eye Center in Boston. The collaboration made it possible for Wenke to stay in Boston during her thesis and gave her the perfect opportunity to work with experts in deep learning, optical device engineers as well as clinicians. So, what did Wenke do during her time in Boston? It was everything from getting to know OCT imaging devices and understanding the image acquisition process, experiencing the imaging procedure, as well as reviewing, analyzing and refining data and labels - and obviously lots of coding. Wenke's goal was to analyze a thin feature in OCT B-scans called the hyporeflective gap in the outer retina, a proposed biomarker for the common eye disease “agerelated macular degeneration (AMD)”. Since this gap is very thin, the data needed to be of high resolution and was pre-processed by her supervisor Stefan Ploner. Especially the merging of 6 volumes, motion correction, and layer flattening were important steps for removing inconsistency between the B-scans as well as enhancing the visibility of the hyporeflective gap. In order to analyze the hyporeflective gap, she first needed to derive a precise and reliable ground truth - one boundary for the posterior Retinal Pigment Epithelium (pRPE) and one for the Bruch’s membrane (BrM), which together enclose the gap. A semi- automatic intensity-based label refinement approach was applied using initial labels from a 2D nnUnet approach. In the next step, the “Pictorial representation and vision are important characteristics in life, without which you are strongly impaired!”
RkJQdWJsaXNoZXIy NTc3NzU=