Research On Application Of Deep Learning In Cell Image Segmentation

Automatic cell segmentation is an essential step in the pipeline of computer-aided diagnosis(CAD).At present,the diagnosis and grading of many cancers depend on the number and morphology of cells.Accurate segmentation of cells can not only assist the pathologists to make a more precise diagnosis,but also save much time and labor.However,this task suffers from stain variation,cell inhomogeneous intensities,background clutters and cells from different tissues.To address these issues,we have made some improvements to the convolutional neural network.The contributions can be summarized as:1.We propose an Attention Enforced Network(AENet),which is built on spatial attention module and channel attention module,to integrate local features with global dependencies and weight effective channels adaptively.Besides,we introduce a feature fusion branch to bridge high-level and low-level features.Finally,the marker controlled watershed algorithm is applied to post-process the predicted segmentation maps for reducing the fragmented regions.2.Due to the limited number of cell images in the training set,some data augmentation strategies are implemented on training set,including flip,rotation,zoom and random crop.These operations can ensure the richness of the data and reduce the impact of over-fitting.3.Because the images in the test sets are taken from different hospitals,the staining difference is large,which seriously affects the final segmentation effect.We present an individual color normalization(ICN)method to deal with the stain variation problem by normalizing the color of pathology images individually.Besides,in the test stage,we employ patch-based and multi-scale strategy to improve the accuracy of segmentation.The proposed method can segment cell images effectively.We evaluate this model on the MoNuSeg dataset.The quantitative comparisons against several prior methods demonstrate the priority of our approach.