Classification Of Carcinogenesis Based On Cellular Mechanical Properties

Cancer is always threatening people’s life safety,so the diagnosis and treatment of cancer is a hot topic in the medical field.The origin of cancer is cancer cells.Previous studies on cancer cells observed and studied from the aspects of cell structure and biochemical characteristics,without drawing conclusions through precise data analysis,while quantitative studies on cell mechanical characteristics can accurately characterize the changes in the process of cell canceration.Due to its nanoscale resolution,AFM has great advantages in measuring cell surface characteristics.In this paper,AFM is used to measure the mechanical characteristics of cells in experiments,and neural network model is used to identify the four categories of carcinogenicity.Young’s modulus,adhesion,work of adhesion and membrane tension are the main mechanical properties of cells studied in this paper.Based on Hertz contact model,generalized JKR model and surface tension model,the mechanical characteristics of cells were theoretically studied and experimentally designed.The original forcedisplacement curve was measured by AFM experimental system,which provided the original curve for the subsequent fitting of mechanical characteristics of cells.The processing of original force-displacement curve and data analysis are the premise of using neural network model.The curve was corrected and young’s modulus,adhesion and work of adhesion were extracted with AFM system software.According to the analysis of actual functional requirements,MATLAB 2017 A was used to write graphical user interface(GUI)to extract membrane tension,and it was proved that cancer grade could not be identified based on the single mechanical characteristics and double mechanical characteristics of cells.The neural network model was applied to recognize the four classification of cell canceration grade,and the recognition rate reached 91.25%.In this paper,the BP neural network model was built.Through the input of four mechanical features and the recognition of the output of four cancer grades,the final recognition rate reached91.25%.The generalization of the model was evaluated by the confusion matrix,ROC curve and AUC value,and it was proved that the recognition rate and generalization of the model improved with the increase of the number of input features.Cytoskeleton visualization experiments confirmed that cytoskeleton confounding degree increased with the increase of cancer grade.The obtained images were transformed into fluorescence color map,gray scale map and binary map to extract the fractal dimension value of the images,so as to quantify the degree of cytoskeleton confounding and prove that the degree of cell confounding is consistent with the gradation law of cell canceration grade.