Automatic Nuclei Segmentation Method using Median Filter for Denoising

Nuclei deliver vast information about the microenvironment. The workload of pathologists can be reduced and precise observations of the microenvironment for clinical research can be made with an automatic nucleus segmentation approach. Although the present deep-learning (DL) based nuclei segmentation techniques generally outperform the conventional ones, this task is still challenging particularly when the nuclei present are cohesive and overlap with one another. In the present work, we have proposed a new nuclei segmentation technique in which we are using a median filter for the pre-processing of the source image along with the U-Net convolution neural network (CNN) model. The median filter can remove the noise present in the source image keeping the edge information preserved. The pre-processing step used enhances the efficacy of the proposed segmentation technique. The performance of the proposed method is compared with the existing techniques for nuclei segmentation on a similar dataset. The proposed technique attained IOU(JI), Accuracy, Precision, and F1 score values as 0.8024, 96.23%, 0.8895, and 0.8704 respectively. The empirical findings indicate that the performance of our proposed technique is better as compared to the other existing methods.