Speaker
Description
Deep learning algorithms became more and more popular for solving image processing tasks in the biomedical field. One of such tasks is cell detection and segmentation in differential interference contrast microscopy (DIC) brain tissue images. These algorithms require hundreds and thousands of images with ground truth segmentations for training to be highly accurate and we lack similar publicly available datasets for this domain. Obtaining a proper amount of DIC brain tissue image data is an expensive process, thus a creation of artificial images with ground truth segmentations could possibly address this problem. In this work, we try different approaches - including generative adversarial neural networks - for generation of DIC brain tissue images with ground truth segmentations to acquire deep learning based segmentation methods. As addition, we evaluate the performance of these methods on different artificially created datasets. The models can be further integrated into the patch-clamping system to improve the success rate by accurately targeting the cell centroid.
