Research And Application Of Automatic Analysis Of Calcium Imaging Data Based On Cell Recognition And Event Detection Algorithm

Neuroscience is a comprehensive research field aiming to understand brain functions.During the last 20 years,neuroscience has experienced rapid developments.It is of great importance to the advances of artificial intelligence and the treatment of various neurological and psychiatric diseases.Computer technology plays an essential role in the development of neuroscience.With the novel neural activity recording technologies,analyzing the data of brain science has become a challenging issue,which needs various computer techniques,e.g,automatic data analysis,for improving the efficiency and accuracy.In the past 10 years,two-photon calcium imaging technique has been widely used in the tracking neuronal population activity,and can be easily combined with cell type specific markers for the analysis of specific types of neurons in the neural circuit.In the analysis of cellular imaging data,the neuronal activity of each cell needs to be extracted.However,it is time-consuming and difficult to unify the criterion of manual analysis.Therefore,it is of great importance to identify the position and profile of individual cells automatically and accurately by using computer algorithms.Base on the segmented cells,the neuronal activities of each cell can be extracted as the relative change of fluorescence,which can be further combined with the behavioral variables to investigate how behavior related information is processed in neural circuits.So the automated segmentation of neuronal cells and the detection of calcium transients are fundamental and crucial issues in the analysis pipeline for the cellular imaging.So,this paper is divided into two parts:(1)This paper proposed a new cell recognition and segmentation method.This method can be mainly considered as 3 steps:(a)using Multi-scale Laplassian of Gaussian filtering(Multi_LoG)to identify the positions of local extrema as the seed points of imaged cells;(b)using Convolutional Neural Network(CNN)to remove false positive results;(c)using the Threshold of Weighted intensity And seed-Normal Gradient dot product image(TWANG)algorithm to segment the contour of the cell,the advantage of the algorithm is that it balances the accuracy of segmentation and computational complexity.In this paper,the method is applied to the image of benchmark data and the two-photon calcium imaging data recorded by the Brain Research Center,Third Military Medical University.(2)After calculating the relative fluorescence change for each imaged neuron,this paper proposed a method for detecting calcium transients: the noise level was estimated by using a sliding window,and then the feature information(amplitude and rise speed)of calcium signal in the following detection window was extracted to compare with the predefined parameters.If the calcium signal is determined as an event,the noise level of the detected transient is estimated for the next round detection.In this paper,the method is applied to the simulation data of calcium event and the calcium fluorescence signals recorded by the Brain Research Center,Third Military Medical University.Based on the above work,this paper evaluated the proposed analysis methods with respect to Recall rate,Precision and F-score.The efficacy and accuracy of the proposed methods were validated by benchmark and simulation data.The evaluation results were compared with the state-of-the-art methods,the comparison results showed that the proposed methods can provide higher accuracy for the recognition of cells in imaging plane and the detection of calcium events.This will benefit the large scale imaging data and play an important role in the celluar imaging.