Intracellular Calcium Signaling Analysis Based On Deep Learning

Ca²⁺is involved in almost all important cellular life activities,and intracellular calcium signaling transduction(intracellular calcium response)is the way of Ca²⁺information transmitting.Cells are sensitive to the changes that occur around them,and make intracellular calcium responses to the changes;at the same time,intracellular calcium signaling can regulate cell life activities;in addition,the characteristics of intracellular calcium signaling curves can reflect the physiological activities of cells.Therefore,in related researchs of cell and tissue engineering,intracellular calcium signaling detection and analysis is a common experimental method.However,in the process of traditional calcium signaling analysis at the tissue level,the thick tissue and dynamic changeable cell contour increase the difficulty of automatic cell identification.The irregularity of calcium signaling curves increases the difficulty of automatic acquisition of calcium response peaks,so manual identification of cells and calcium response peaks are needed.Because of the need to analyze and identify each cell and each calcium response peak during calcium signaling analysis,the amount of data is large and the process is cumbersome.Therefore,the traditional calcium signaling analysis method is inefficient and the accuracy is low.The advantage of deep learning is to learn images or signals recognized by human,extract human recognition experience and use it to complete the human tasks.Therefore,we used Fully Convolutional Neural Networks(FCN)to optimize cell identification methods in process of calcium imaging,and constructed a long short-term memory(LSTM)recurrent neural networks(RNN)"LSTM-F"to identify calcium response peaks,based on deep learning theory and data characteristics of calcium signaling.In addition,in order to automate the calcium signaling analysis process,a method for automatically acquiring calcium signaling curves is constructed based on the connected domain algorithm.The improved calcium signaling analysis method advances the whole automatically process of calcium signaling analysis in cell and tissue engineering research methods,simplifies the processing steps of calcium signaling analysis,improves the efficiency of calcium signaling analysis,and reduces the threshold for calcium signaling analysis.Based on the above,the main contents and results of this study are as follows:(1)Acquisition of calcium response dataCalcium response data acquisition method:We used in situ cellular calcium signaling in pig cartilage tissue as representative data for in situ cellular calcium signaling in tissue.The porcine in situ chondrocytes were labeled with a calcium ion fluorescent probe,and the calcium imaging video recorded within 10 minutes was recorded by a common fluorescence microscope.Then the cells in the video were manually identified,and the calcium curves was obtained.The calcium response peaks in the curve are labeled.The results showed that calcium imaging images are clearly and available,and calcium curves with characteristics of intracellular spontaneous calcium response were obtained as the study data.(2)Construction of cell recognition algorithm based on deep learningCell recognition was performed with a FCN model.The results show that the method of training the FCN model with four times is feasible;using the FCN model to recognize the cells does not require pre-processing and post-processing;It is only necessary to input any size images into the model to obtain the recognition result in a very short time,the FCN model saves 99.93%of the time compared to the method of manually identifying cells.(3)Construction of automatic acquisition algorithm for calcium response curvesThe calcium imaging pictures segmented by the FCN model are overlapped as one,and the traditional calcium curve acquisition method is improved by using the connected domain algorithm according to the overlapped picture and the calcium imaging pictures.The results show that the cell contours in the overlapped images are clear and complete;The Pearson correlation coefficient of the calcium curves automatically acquired by the automatic acquisition algorithm and human is maintained at about 0.985,which indicates that the calcium curves acquired by automatic acquisition algorithm are very close to calcium curves acquired by human.Therefore,the calcium curves automatic acquisition algorithm can be used in the analysis of subsequent study of calcium signaling.(4)Construction of calcium response peak recognition algorithm based on deep learningUsing the long short-term memory(LSTM)recurrent neural networks(RNN),constructed the LSTM-F network based on the characteristics of the calcium curves,to recognize the calcium response peak.The results show that the LSTM-F model we trained can be used to recognize normal calcium peaks.Compared with manual recognition,the LSTM-F model can recognize calcium response peaks in a very short time and saves 99.97%of the time.