Learning Cell Nuclei Segmentation Using Labels Generated With Classical Image Analysis Methods

Creating manual annotations in a large number of images is tedious bottleneck that limits deep learning use many applications. Here, we present study which used the output classical image analysis pipelineas labels when training convolutional neural network(CNN). This may not only reduce time experts spend annotating but it also lead to an improvement results compared from pipeline training. Inour application, i.e.,cell nuclei segmentation,we generated using CellProfiler(a tool for developing pipelines biomedical applications)and trained on them U-Net-based CNN model. The best model achieved 0.96 dice-coefficient segmented Nuclei and 0.84 object-wise Jaccard indexwhich was better than method generating by 0.02and 0.34, respectively. Our experimental show this such feasiblebut thatthe segmentationsare clear pipelineused annotations.