Study On Morphological And Active Mapping Model Of Bone Marrow Mesenchymal Stem Cells Based On Three-Dimension Features

Bone marrow mesenchymal stem cells(BM-MSCs)are stem cells that support hematopoiesis and regulate microenvironment.They not only have self-replication ability,but also have the potential of multi-directional differentiation,which plays an important role in clinical treatment.The experiment found that the bone marrow microenvironment of patients with acute myeloid leukemia changed,the patient's bone marrow mesenchymal stem cells function weakened,and the proportion of aging cells increased.The morphological characteristics and activity of bone marrow mesenchymal stem cells are correlated,and the structural changes of stem cell function subpopulations are analyzed and identified.This paper provides a new idea for the quality control of clinical bone marrow mesenchymal stem cell sorting and enrichment process.Following three aspects of work are done in this paper.The data samples obtained by the rotary fluorescence confocal microscope will be affected by the optical system to produce image degradation and blur.The iterative blind deconvolution algorithm is used for image restoration.Based on the restored image,the adaptive threshold algorithm is applied to realize image binarization.Three-dimensional reconstruction of 60 single-cell sections is performed by marching cube algorithm to obtain three-dimensional structure of cells,and threedimensional morphological features are calculated to describe cell size,shape and cellular transport capacity.Among them,30 cases of normal samples and aging samples are used.The principle component analysis and T-distribution stochastic neighbor embedding algorithm are implemented for feature dimension reduction.Support vector machine and multilayer perceptron are used for classification and identification.The highest accuracy rate is 95.83%.To understand the relationship between cell structure and function from a new perspective,a dedicated convolutional neural network is designed for small sample data sets,and the maximum area slice of single cell as the feature plane was input to classify bone marrow mesenchymal stem cells in patients with acute myeloid leukemia/normal control group.176 samples are data augmented,and the neural network is trained by 8-fold cross validation method.The average accuracy of classification is 98.88%.By transforring the trained network and using the principal component analysis-ridge regression model to analyze the cell activity,the quantitative evaluation of bone marrow mesenchymal stem cells is carried out.The model determination coefficient is 0.8736,verifying the validity of the model.The three-dimensional features proposed in this paper describe the cells in terms of volume and shape from the morphological point of view,and provide an effective basis for cell classification.The proposed model achieves quantitative evaluation of bone marrow mesenchymal stem cells,and provides a basis for sorting and enriching therapeutically active stem cells from an image perspective.It can provide important reference for clinical diagnosis.