Research On Segmentation Method Of Medical Microscopic Cell Image

Medical microscopic cell image processing covers many fields,such as artificial intelligence,image processing,computer vision,biomedicine,and so on.It plays an important role in the clinical diagnosis process.Quantifying cell behavior by segmenting medical microscopic cell images,which has a great value to biomedical research such as tissue cell classification,cell mutation,DNA damage detection,hematology,cancer research,etc.In the thesis,the image database of human oral mucosal cells built by our team was used as the cell image dataset.This thesis studies the design of fluorescence microscopy imaging system based on structured light illumination and cell image acquisition,medical microscopy cell image enhancement method,and medical microscopy cell image segmentation methods.The main lines as follows.1.A fluorescence microscopy imaging system based on structured light illumination is designed,and cell image is obtained.By analyzing the problem of diffraction limit resolution in the optical imaging process,combining the structured light illumination fluorescence microscopy imaging process and the image reconstruction algorithm,thus a structured light illumination generation scheme is established.A two-dimensional wide-field fluorescence microscopic optical imaging system based on DMD and LED is proposed.Then,the structured light illumination fluorescence microscopic imaging system is used to perform super-resolution imaging of oral mucosal cells,and the imaging system is calibrated to achieve SIM super-resolution imaging.It is the premise and key of the medical microscopy cell image processing.2.A robust medical microscopic cell image enhancement method based on dual-tree complex wavelet transform and morphology is proposed.To begin with,we utilize the pre-processing method to the captured medical microscopic cell images.Then,the dual-tree complex wavelet transform(DTCWT)is applied to decompose the cell image to obtain high-pass subbands and low-pass subbands.Then,a Contourlet-based denoising method is applied to the high-pass subbands.Combined Bayesian shrink principle to obtain the threshold,and then we improve the denoising method,considering the local correlation of the neighborhood,and the adaptive optimal threshold is used to achieve denoising.For the low-pass subbands enhancement,we improve the morphology top-hat transform by adding dynamic multi-scale parameters to achieve an equivalent percentage enhancement and at the same time achieve multi-scale transforms in multiple directions.Finally,we develop the inverse DTCWT method to obtain the medical microscopic cell image after processing the low-frequency subimages and high-frequency subimages.3.A new method for adhesion cell images segmentation based on weighted curvature and gray-scale distance transformation is proposed.The cell and nucleus thresholds are calculated by the double threshold iterative method,and the watershed transform is used to realize the three class cell image initial segmentation.On this basis,through the curvature calculation of contour points,the segmentation line is established.We propose a distance transform method of correlated gray image,and obtain the marked image by the threshold method.Then,we use a marker controlled watershed transform for secondary segmentation to achieve adhesion cell segmentation.Experimental results show that our proposed method has a good segmentation effect for complex adhesive cells.4.A multi-class medical microscopic cell image segmentation method based on graph model is proposed.Based on graph model and convolutional multi-scale fusion FCN network,a polytree model for over-segmented cell image is built,and a segmentation method based on graph model is proposed.Then,combined with the prior information of polytree model,the closed form solution for the posterior of the polytree model is solved,and the cell image segmentation network framework is built,and the image quality evaluation function is established.The experimental results show that the cell image segmentation method based on graph model can extract a smooth cell boundary,and can extract binucleate cells continuously.At the same time,it can segment nucleus and background information well,which is very close to the ground truth.5.A medical microscopic cell image segmentation method based on Neural Ordinary Differential Equations is proposed.Based on U-Net network model,the specific position of adding the ODE-block to the U-Net network is determined through comparative experiments,and a new cell image segmentation network model based on the neural ordinary differential equations and U-Net(named as NODEs-Unet)is proposed.Then,by adjusting the error tolerance of ODE-block to increase the network depth,a binary segmentation network based on NODEs-Unet(named as 2NODEs-Unet)is proposed.Experimental results show that our proposed method can increase the depth of the network model without increasing parameters of the network model,and has low computational complexity.It can successfully segment background,cell,and nucleus with clear boundary and complete details,which is very close to the ground truth.