A Computer-Aided Diagnosis System of Fetal Nucleated Red Blood Cells With Convolutional Neural Network

Abstract

Context.— The rapid recognition of fetal nucleated red blood cells (fNRBCs) presents considerable challenges. Objective.— To establish a computer-aided diagnosis system for rapid recognition of fNRBCs by convolutional neural network. Design.— We adopted density gradient centrifugation and magnetic-activated cell sorting to extract fNRBCs from umbilical cord blood samples. The cell-block method was used to embed fNRBCs for routine formalin-fixed paraffin sectioning and hematoxylin-eosin staining. Then, we proposed a convolutional neural network–based, computer-aided diagnosis system to automatically discriminate features and recognize fNRBCs. Extracting methods of interested region were used to automatically segment individual cells in cell slices. The discriminant information from cellular-level regions of interest was encoded into a feature vector. Pathologic diagnoses were also provided by the network. Results.— In total, 4760 pictures of fNRBCs from 260 cell-slides of 4 umbilical cord blood samples were collected. On the premise of 100% accuracy in the training set (3720 pictures), the sensitivity, specificity, and accuracy of cellular intelligent recognition were 96.5%, 100%, and 98.5%, respectively, in the test set (1040 pictures). Conclusions.— We established a computer-aided diagnosis system for effective and accurate fNRBC recognition based on a convolutional neural network.