Detection And Multi-object Tracking Of Cancer Cells Based On Phase Contrast Microscopic Sequence Images

In biomedicine domain,it is important to replace manual methods with automatic methods f or monitoring and analysing the characteristics of cancer cells.Thus,it can help to develop anti-c ancer drugs efficiently.In recent years,with the development of deep learning,it can be widely a pplied for target classification,detection and tracking,etc.In this paper,object detection and mul ti-object tracking methods based on deep learning are proposed.These methods have a good perf ormance of detection and tracking.The main achievements and contributions of this paper are as follows:1.In this work,we adopt an object detection strategy for classifying the proposed region by extracting region proposals from the image.Firstly,we made manual labeling of cancer cell data.Then,based on the Faster RCNN detection framework,we employ a deep convolution neural network model to train the region proposal network.Finally,we adopot a classifiter to complete the classification of the proposed regions,and the detection task of cancer cells is achieved.2.In order to improve the detection effect of the model in the cell adhesion region,we design a CSA algorithm,which is based on threshold segmentation and scanning circle to optimize the cell adhesion region.Meanwhile,in order to overcome the problem of less training data,we initialize the network model parameters by adopting some auxiliary data to pre-train network parameters.The results show that our detection algorithm can achieve a good detection effect on the scattered and adherent parts of cancer cells.3.Based on the above detection results,we propose a multi-object tracking method by combination of deep learning and traditional methods of cancer cells.The results indicate that the proposed tracking approach can achieve good tracking effect in the dispersed and adherent regions of cells.4.When matching the target in the tracking algorithm,we designed a comprehensive descriptor for cancer cell targets based on the morphological characteristics and the data we used.We use the convolution feature in the comprehensive characterization descriptor,making our descriptor more efficient.At the same time,we have achieved an efficient multi-target tracking system for cancer cells.