Modern convolutional object detectors for nuclei detection on pleural effusion cytology images

Abstract

Nuclei detection is a crucial step in the cell-based analysis of a wide range of pathological images. It is also the basis of many automated methods such as cell counting, segmentation, and tracking. Nevertheless, it is seen as a challenging task because the nuclei display variability of size, shape, orientation and intensity and number of nuclei. In recent years, a variety of Convolutional Neural Network (CNN) based object detectors have been proposed. Although these methods have demonstrated superior success in different object detection problems, they have not yet been used for nuclei detection on pathology images. The main contributions of this work are: 1) We propose the first implementation of the faster region-based convolutional neural network (Faster R-CNN), the region-based fully convolutional network (R-FCN), and the single shot multibox detector (SSD) methods for nuclei detection on pathology images. These methods are viewed as ‘modern convolutional object detectors’. 2) We present a novel database of pleural effusion cytology images. We used proposed object detectors with different ‘feature extractors’ such as Residual Network (ResNet), Inception v2, and MobileNet for performance comparison. Experiments show that all these object detectors using different feature extractors achieve remarkable detection speed and accuracy.