Cell nuclei detection and segmentation for computational pathology using deep learning

This work presents a deep learning model and image processing based processing flow to detect and segment nuclei from microscopy images. This work aims at isolating each nuclei by segmenting the boundary and detecting the geometric center of the nuclei. The deep learning model employs a multi-layer convolutional neural network based architecture to extract features from both spatial and color information and to generate a gray scaled image mask. Subsequent image processing steps smooth nuclei boundaries, isolate each individual nuclei and calculate the geometric center of the nuclei. The proposed work has been implemented and tested using H&E stained microscopy images containing seven different tissue samples. Experimental results demonstrated an average precision of 0.799, recall of 0.955, F-score of 0.86, and IoU of 0.835.