CARS Preview 2018

Project by RSIP Vision 27 CARS 2018 Preview DC Segmentation: within IVCM images, dendritic cells are identified and distinguished from the background as bright individual dendritiform structures with cell bodies. Training was improved by manual verification and correction of all DCs found by the CNN. See image below, where results from Image J can be compared to results from CNN: Interclass correlation was calculated between the semi-automated analyses (Image J) vs. deep learning (CNN). Shows intermethod agreement values very close to 1. Conclusion: Laser IVCM enables non-invasive visualization of corneal layers. Images of corneal DCs provide precious information about the severity of inflammation. Our results demonstrate that our Deep Learning algorithm allows a standard, objective, accurate and fast analysis of IVCM images, thus providing an effective tool for evaluating pathologies, while at the same time it saves the tedious manual work. As a result, it contributes to improve diagnostic and treatment. Our results demonstrate that Deep Learning allows a standard, objective, accurate and fast analysis of IVCM images, thus providing an effective tool for evaluating pathologies. Image J CNN Image J CNN 120.5 ± 161.6 DENSITY (cells/mm 2 ) 118.2 ± 155.8 1692.6 ± 2641.0 AREA (µm 2 ) 1736.7 ± 2458.7 1040.0 ± 2458.7 PERIMETER (µm) 862.1 ± 1135.0