Pathological Diagnosis Model Of Cardiomyocytes Based On Deep Learning

With the development of industrial society,people’s quality of life continues to improve,but cardiovascular disease is still a major problem threatening human health.As of 2021,cardiovascular disease has ranked first in the global mortality ranking for many years.The risk of cardiovascular disease is mainly manifested in the microscopic view of the abnormal function of myocardial cells regulated by calcium ions.Calcium ions are widely distributed in organisms,and many physiological activities in organisms are filled with the shadow of calcium ions.One of the main functions of calcium ions is to participate in the transmission of calcium signals.This transmission function is related to the excitation and contraction of cardiomyocytes in many aspects.Cardiovascular diseases such as arrhythmia and heart failure are often accompanied by calcium leakage events on the sarcoplasmic network of cardiomyocytes.The current detection of calcium leakage events mainly focuses on calcium sparks,calcium waves and calcium release events in cardiomyocytes.And so on.Calcium spark is the most basic manifestation of calcium signal,and its essence is a "diffusion-reaction" process,that is,the process of calcium ion diffusion from the release end to the surrounding,constantly interacting with various buffer molecules.Observed in cardiomyocytes of various organisms,calcium sparks can be observed in more than 95% of cardiomyocytes.This paper uses deep learning methods to study calcium spark recognition,analysis and pathological classification.The specific research content and results are as follows:First,the deep learning method is used to identify and process the fluorescence images of cardiomyocytes scanned by the laser confocal microscope,and the improved Faster-R CNN network framework and the deep residual network Resnet50 structure combination are used to generate the calcium spark recognition and analysis model CSDAM（Calcium Spark Detection and Analysis Model）.CSDAM can identify and extract 95.8% of calcium sparks in the picture.After completing the recognition of the calcium spark,extract the pixels of the recognition area to fit the data in time and space,and obtain the five spatiotemporal parameters of the calcium spark: space half-width（FWHM）,time half-width（FDHM）,rise phase time（TRise）,fall half time（T50）and amplitude（Amplitude）.Finally,classify the temporal and spatial parameters of the calcium spark obtained by the CSDAM model,and design the calcium spark pathological judgment model CSPJM（Calcium Spark Pathological Judgment Model）using the logistic regression method and the similarity scoring method based on Euclidean distance.CSPJM is divided into two types: Pathology-healthy model and stimulus induced pathological changes-healthy model.The two models have accuracy rates of 83.4%and 84.2%,respectively.Using the model,the correct pathological results of 70%calcium sparks in each cardiomyocyte picture can be obtained.Based on this result,the pathological type of the cardiomyocyte can be judged.The pathological judgment accuracy rate of the final cardiomyocyte image was 92.3%.