Research Article Open Access

A Robust Deep Learning Approachto Detect Nuclei in Histopathological Images

Abstract

Automated cell detection in histopathology images is challenging due to large variations in size, density, and batch variations. Nuclei detection provides useful information for evaluating cancer progression and prognosis. The performance of most classical nuclei detection methods relies on appropriate data selection. These methods require experts to create useful features. On the other hand, deep learning can extract feature sets from the data automatically, not requiring the design of feature extractors by experts. In this work, a new enhancement for deep learning approach is proposed to learn a continuous mapping from H&E image patches centered around nucleus centroids to nuclear distance maps. Our approach formulates the problem as a continuous regression problem and builds a fully convolutional regression network. In this method, it handles partial detections and irregular-shaped, neighboring nuclei, and different nuclei sizes and color. We train the network with the colorectal dataset which is publicly available dataset. The work is evaluated with the human bone marrow dataset without re-training and superior results are achieved.

Laith Alzubaidi1, Raja Daami Resan2, Huda Abdul_hussain3, Haider A. Al-Wzwazy4, Hayder Albehadili5

To read the full article Download Full Article