Study On Toxic Chip System Of C.elegans Based On Microfluidics And Computer Vision

The representative model of Caenorhabditis elegans(C.elegans)has the characteristics of small size,short life cycle,simple structure and high gene conservation.It is an important in-vivo research tool in the field of environmental exposure and toxicology research.The microfluidic system is matched with the size of the C.elegans.Compared with the C.elegans experiment on the agar plate,it has the advantages of small reaction system,high throughput,automation and flexible operation.Therefore,the microfluidic chipbased C.elegans research platform provides a new means for high-throughput,large-scale environmental chemical assessment.However,there is still no integrated microfluidic analysis platform that integrates automated opration and image analysis.In response to the above problems,a hardware system platform based on microfluidic chip is set up firstly,including the design and fabrication of two microfluidic chips and the optimization of the speed of video acquisition.These two chips provide a precise microenvironment for the exposure of C.elegans to a given environmental toxin.In the toxicological experiment based on C.elegans,the body length and swing frequency are often used as important physiological indexes to evaluate toxicity.However,by using computer vision to extract the features of C.elegans,the following difficulties exist:a)Since the C.elegans is transparent and the microfluidic chip made by PDMS is also transparent,the foreground contour segmentation of the C.elegans is a difficult point in the whole system.b)When multiple C.elegans are present in one chamber,multiple contours of C.elegans may be entangled,resulting in loss of tracking due to the inability to recognize the contours of individual C.elegans.c)Because C.elegans are non-rigid and their morphological changes are diverse,it is also a difficult point to track the C.elegans in multi-frame images.To alleviate the problems mentioned above,this paper proposes the following solutions: a)Aiming at the shortcomings of the traditional image segmentation algorithm in the foreground segmentation task of the worm(such as poor robustness,dependence on hyperparameter selection and unsatisfactory segmentation),this paper proposes a segmentation algorithm based on deep convolutional networks and conditional random fields.Compared with other segmentation algorithms,the proposed algorithm can significantly improve the performance of foreground segmentation of C.elegans.On the segmentation dataset of the C.elegans we have labeled,the segmentation algorithm proposed in this paper achieves a segmentation accuracy of 0.11% on the pixel error index.b)In order to solve the entanglement between the contours of multi-worms,this paper proposes a encode-decode architecture based on convolutional neural network,which effectively solves the problem of parsing the contour of a single C.elegans.c)In order to achieve multi-worms tracking,this paper proposes a simple but effective tracking strategy to track the C.elegans by nearest neighbor search.The experiment found that the tracking method can achieve robust tracking of the C.elegans under the condition that the foreground segmentation of the C.elegans is complete.Using the traced contour of C.elegans,physiological characteristics such as body length and swing frequency of the C.elegans can be calculated.Finally,the effect of linear concentration gradient hydrogen peroxide on C.elegans activity is studied by this automatic platform,and the experimental results are consistent with the traditional experimental results.The application prospects of microfluidic platform and automated image analysis system in toxicology research are presented.