Methodology for comprehensive cell-level analysis of wound healing experiments using deep learning in MATLAB

摘要

Workflow of deep learning MATLAB application for valid usage of iCD network. Live cell images from cell culture experiment were labelled manually supported by the implemented semi-automatic labeling module, which is based on the threshold method. Segmented image and the corresponding raw image form an image pair. By using rotating, scaling and skewing the image pair is augmented which creates a multiplied dataset for training. iCD Network is trained by using the raw image and the manually (semi-automatic) labelled image as input. Validation is performed by using only raw images only as input to the iCD Network and comparing the results with manually (semi-automatic) labelled image. After successful iCD training and validation, the network can be applied to novel live cell images to analyze cell motion at cell scale and population scale