65 results about "Cell image segmentation" patented technology

An artificial neural network system for image classification, formed of multiple independent individual convolutional neural networks (CNNs), each CNN being configured to process an input image patch to calculate a classification for the center pixel of the patch. The multiple CNNs have different receptive field of views for processing image patches of different sizes centered at the same pixel. A final classification for the center pixel is calculated by combining the classification results from the multiple CNNs. An image patch generator is provided to generate the multiple input image patches of different sizes by cropping them from the original input image. The multiple CNNs have similar configurations, and when training the artificial neural network system, one CNN is trained first, and the learned parameters are transferred to another CNN as initial parameters and the other CNN is further trained. The classification includes three classes, namely background, foreground, and edge.

The invention relates to a cell image segmentation method based on automatic feature learning. As a method for learning features of cell images is very good in feature learning capacity, the cell segmentation accuracy can be greatly improved, and meanwhile, a random forest classifier does not need to select the features, so that the method is capable of well solving the confronted problems of feature extraction and selection in a recognition process. The cell image segmentation method based on the automatic feature learning comprises the following steps: 1, preprocessing: preprocessing initial cell images in a training set and a test set; (2) training a feature extractor; (3) performing recognition by utilizing the random forest classifier; and (4) postprocessing.

The invention discloses a method for segmenting a cervix uteri liquid base cell image. The method comprises the following steps of: 1, cell segmentation: performing illumination correction on an image, filtering noises and enhancing a cell boundary; and dividing the image into three types of cytoplasm, nucleus and background to eliminate a background class area; 2, nucleus segmentation: detecting centers of each of nucleuses in the cell area, estimating the approximate shapes of the nucleuses close to the centers of the nucleuses and correcting the shapes of the nucleuses to acquire an accurate boundary. The method is applied to cervix uteri cells and cell masses in a wide vision range, so that the phenomena of changed illumination, non-uniform dyeing and noise influence to a certain extent are avoided, and the cytoplasm, the single nucleus and adhesion nucleus can be segmented reliably and accurately.

An artificial neural network system for image classification, including multiple independent individual convolutional neural networks (CNNs) connected in multiple stages, each CNN configured to process an input image to calculate a pixelwise classification. The output of an earlier stage CNN, which is a class score image having identical height and width as its input image and a depth of N representing the probabilities of each pixel of the input image belonging to each of N classes, is input into the next stage CNN as input image. When training the network system, the first stage CNN is trained using first training images and corresponding label data; then second training images are forward propagated by the trained first stage CNN to generate corresponding class score images, which are used along with label data corresponding to the second training images to train the second stage CNN.

The present invention discloses a thymic cell image segmentation method based on an improved U-Net network, which comprises the following steps: carrying out image preprocessing on a UCSB breast imagedata set; in a U-Net network, adding a void residual module and an attention module; training the U-NET network according to the set training strategy; setting up the evaluation indexes including F1score, object-level Dice coefficient and Hausdorff distance. The optimal model is obtained by optimizing the network through the evaluation indexes. The cell image to be segmented is inputted into theoptimal model, and the segmentation mask is obtained by feature extraction and feature upsampling. By improving a basic segmentation network and creating a new cell image segmentation method, the invention solves the problem of low precision in the automatic segmentation process of the thymus image, and improves the accuracy and efficiency of the segmentation.

The invention discloses a cervical cell image segmentation method based on an antagonistic generation network, comprising the following steps: a cell image is coarsely segmented, wherein for the cellimage coarse segmentation, a threshold method and a watershed algorithm are used for coarse segmentation of an original image to form guiding factors, and the original image is cut into small images;a virtual body segmentation image is generated, wherein the generated virtual body segmentation image is generated by using an antagonistic generation network designed in combination with a self-encoder, taking a clipped small image as an input, and using the guiding factors to help the neural network to locate a region of interest; a solid cell image is extracted, wherein the solid cell image extraction refers to that a real cell image is extracted from the clipped small image according to the virtual body segmentation image. The cervical cell image segmentation method based on the antagonistic generation network provided by the invention is the first time to use the antagonistic generation network to solve such problems, provides a novel automatic cell image segmentation method, and simultaneously solves the component loss in the traditional overlapped cell segmentation method.

The invention relates to a blood leukocyte image segmentation method based on UNet++ and ResNet. The method comprises the following steps: firstly, extracting multi-scale image shallow features by using an encoder with a convolution block and a residual block; extracting deep features of the image by using a decoder with convolution and deconvolution, and fusing shallow features and deep featuresby using mixed jump connection so as to reduce a semantic gap between the shallow features and the deep features; finally, designing a loss function based on the cross entropy and the Tversky index, guiding the model to learn effective image features by calculating a loss function value of each layer, and solving the problem of low training efficiency caused by imbalance of sample categories in aconventional segmentation loss function.

The invention relates to a cell separation method based on morphology, which comprises the following steps that: (1) region labeling: a binary image of a result is cut for a cell image, and the label values of all adhesion areas are obtained by labeling based on norphological labeling, and (2) a monomer cell is extracted; and 2. (1) a process of region extraction, (2) first, a label map with adhesion areas are found out from a regional label map, a circular structure with the structural element of 1 is selected and expanded until the boundary of the label map exceeds the outer boundary of an adhesion map, and the size of the label map is the whole of one monomer cell in adhesion cells, and (3) a reconstructed new binary image set is traversed, and the attribution of all monomers are determined according to coordinate and area information, and an image set is established. The cell separation method based on morphology can effectively separate adhered cells and prepares for accurately calculating the morphological parameters of the cells.

The invention discloses a cell image segmentation method based on a Res2-UNeXt network structure, and the method comprises the steps: designing a network structure, and designing a proper and effective network according to the features of a cell image; adding a residual structure and a multi-scale convolution method into a U-Net network. The segmentation process comprises the following steps: obtaining a weight graph required for calculating Loss by utilizing a label graph of a training image; and inputting the original training data set into the Res2-NeXt network, and updating the parametersof the network according to the calculated loss; carrying out iterative training continuously until the precision of network prediction can reach a stable level; using a trained network to predict andinput new data to obtain a cell segmentation map. The invention provides a multi-scale network structure Res2-UNeXt, coarse-grained and fine-grained information can be better obtained, so that the segmentation performance of the method is improved.

The invention discloses a single-cell image segmentation method. The method includes the following steps that: 1) image preprocessing is performed: an image is converted into a grayscale image, noisesare removed, and contrast enhancement is performed; 2) block threshold segmentation is performed so as to segment the image into A*A small blocks, the optimal threshold of each block is calculated byusing the OSTU, so that the foreground and the background of each block can be separated from each other; 3) whether nucleus state features obtained by the previous segmentation step are normal or not is judged, if the nucleus state features are normal, it is proved that a segmentation result is relatively good, and the result is outputted; 4) if the segmentation result does not conform to the nucleus state features, the result is an inaccurate segmentation result, a next step of image processing is performed; and 5) adaptive threshold segmentation is performed, the result of the adaptive threshold segmentation is outputted together with other normal segmented images. With the single-cell image segmentation method of the present invention adopted, the problems of inaccurate nucleus segmentation and slow segmentation speed can be solved. The single-cell image segmentation method combines the advantage of high speed of the block segmentation threshold segmentation and the advantages ofhigh accuracy and low workload of the adaptive threshold segmentation; and since the advantage of the block segmentation threshold segmentation and the advantages of the adaptive threshold segmentation are complementary, the quality of an output image can be improved.

The invention discloses a cell image segmentation method based on a particle swarm neural network. The method comprises following steps of inputting 3*3 windows to replace a traditional single pixel channel; adopting a method based on information entropy to determine the quantity of hidden layer neural elements; selecting activation functions of each layer; simplifying weight training to be an optimization problem; using an improved particle swarm to optimize and solve the problem; carrying out a network training; and analyzing test results. According to the invention, the method is improved based on the traditional BP neural network, so cell images can be well segmented; problems about how to determine the network structure and how to ensure that the network is converged to be overall optimal are solved; segmentation effects of the cell images are improved; and time loss in the cell segmentation process is relatively reduced.

The present invention relates to an image segmentation method and system, and a cell image segmentation method and system based on a corrected gradient image and a watershed algorithm. Since a gradient image of a to-be-segmented image is corrected in advance by a foreground mark obtained by distance transformation and a background mark obtained by watershed transformation and then the corrected gradient image is subjected to watershed transformation to obtain an image segmentation result, not only are the advantages of performing watershed transformation on the gradient image, effectively locating the edges of a target object and segmenting out a complete outline of the target object reserved, but also targets with no obvious boundaries in adhesion areas can be distinguished by the foreground mark and the background mark, the phenomena of under-segmentation and over-segmentation do not occur, and the accuracy of image segmentation is improved. Therefore, the methods and the systems are especially suitable for the field of adhesive and overlapped cell image segmentation.

The invention discloses a method of a cervix uteri single cell image segmentation algorithm, and includes the steps of: 1) performing preprocessing on a cervix uteri single cell image to enhance a cell boundary; 2) removing a background of the cervix uteri cell image; 3) determining a gray scale gradient of a cell nucleus and cytoplasm along a ray direction; 4) using a stack gray scale difference compensation algorithm to optimize the ray gray scale gradient in the Step 3), and determining an edge map AREM of the cytoplasm and the cell nucleus of the cervix uteri single cell according to the noise-removed cervix uteri single cell image; and 5) applying gradient vector field GVF Snake model evolution to the determined edge map AREM of the cervix uteri single cell image to accurately locate profiles of the cytoplasm and the cell nucleus of the cervix uteri cell. The method simplifies a complex process of a traditional segmentation algorithm, solves the problems that in the traditional segmentation method cells overlap and adhere, dyed color is inconsistent and a background contains impurities, and improves the segmentation accuracy degree and segmentation efficiency of a normal cervix uteri single cell and a cancerous cervix uteri single cell.

The invention provides a cell classification method and device, and relates to the technical field of biological information, and the method comprises the steps: obtaining a to-be-identified cell image, wherein the cell image comprises at least two types of cells; dividing the cell image into N sub-images, and determining a contour information graph corresponding to the sub-images, the contour information graph comprising contour shapes of cells to be identified in the sub-images, N being greater than or equal to 2; inputting each contour information graph into a classification model to obtaina classification result of the to-be-identified cells in the sub-image, the classification model being determined according to the contour shape of the cells of which the classification result has been determined; and determining a classification result of the cell image according to the classification result of the to-be-identified cell in the sub-image. The method provided by the embodiment ofthe invention can be used for carrying out cell classification on the whole to-be-identified cell image, is high in cell classification accuracy, is beneficial to carrying out quantity statistics on the classified cells, and improves the diagnosis precision.

The utility model relates to an acquisition and processing system of the cellular morphology image, which aims to construct the image processing platform required by the space cellular embarkation experimental device, comprising a video input processing module, a global logic control module, an image data processing module and an image data transmission module. The video input processing module can convert the analog video signal to digital signal and output the abstracted synchronous signal and pixel clock simultaneously; the global logic control module can realize the logic control of the video input processing and the buffer control of the image data; the image data processing module can complete the acquisition of the cellular morphology image and the cellular image segmentation method based on the edge detection; the image data transmission module can realize the rapid data transmission between the system and the host computer. The utility model has the advantages of integral function, simple structure and flexible programming; in addition, the system can realize the real time online acquisition of the cellular morphology image with high resolution and realize a novel cellular image segmentation method.

The invention relates to a cell image segmentation method and device, computer equipment and a storage medium. The method comprises the following steps: reading a cell grayscale image; carrying out histogram equalization operation on the cell grayscale image to obtain an equalized image; performing morphological operation on the equalized image to obtain a morphological image, wherein the morphological operation comprises top cap operation and gradient operation; detecting cell edges in the morphological image through an edge detection algorithm to obtain a cell edge image; carrying out binarization processing on the cell edge image to obtain a binarized cell image; and carrying out cell segmentation according to the binarized cell image to obtain a cell segmentation image. By adopting themethod, the accuracy of cell image segmentation can be improved.

A cell image segmentation method includes receiving a cell image, performing a nuclei initialization step to find an internal marker and an external marker to obtain a potential nuclei and a potential cell boundary, calculating a gradient map of the received cell image, performing a filtering step on the gradient map to generate a filtered gradient map, performing a nuclei detection step to obtain a segmented nuclei, and performing a nuclei validation step to obtain a valid nuclei. The nuclei initialization step includes performing a blob detection step to obtain a nuclei candidate, an outlier removal step to obtain the internal marker, a distance transform step to obtain a distance map, and a cell boundary initialization step to obtain the external marker. In another embodiment, a nuclear-to-cytoplasmic ratio evaluation method using the above cell image segmentation method is proposed.

The invention relates to a cell image semantic segmentation method fusing image segmentation and classification. The method comprises: preprocessing cell image data and then processing the cell imagedata through a bilinear fine-grained classification neural network, an OSTU algorithm and a filling algorithm, and then respectively obtaining a cell classification model and a cell segmentation map;and predicting a foreground connected region of the cell segmentation map by the cell classification model, assigning a prediction result to the connected region so as to obtain a region-by-region classification result, and finally obtaining a semantic segmentation result of the cell test image in combination with a background region obtained by segmentation. A traditional threshold method and a deep learning method are fused to achieve accurate semantic segmentation of the cell image. Compared with a traditional cell image segmentation method, the cell image semantic segmentation method is advantageous in that semantic information of cells can be obtained, the semantic information is of a pixel-by-pixel semantic category, and the method can be applied to identification and isolation of cell contamination.

The embodiment of the invention provides a cell image segmentation method and device. The method comprises the steps of obtaining a cell image to be segmented; inputting the cell image to be segmented into a cell segmentation model to obtain a cell segmentation result outputted by the cell segmentation model, wherein the cell segmentation model is obtained by training an enhanced U-Net network based on a sample cell image and a real label image corresponding to the sample cell image, the enhanced U-Net network is a neural network in which the number of network layers is increased in an initial U-Net network, and a BN layer is added to the network. According to the method and the device provided by the embodiment of the invention, the segmentation effect of the overlapping cells and the adhesion cells is improved based on the initial U-Net network, the cell segmentation accuracy is optimized, while the model training is accelerated based on the BN algorithm, and the model performance is optimized. By inputting the to-be-segmented cell image into the cell segmentation model obtained by training, the rapid, accurate, simple and convenient cell segmentation can be realized.

The invention relates to a particle swarm and fuzzy means clustering based cell image segmentation method. On the basis of an intuitive fuzzy means clustering theory, an intuitive fuzzy membership degree form is improved aiming at cell images and combined with fractional order particle swarms to realize alternate optimization, and a morphological method is adopted for result improvement to realize accurate segmentation of the cell images. The method includes steps: firstly, selecting appropriate parameters, initial particle swarms and relevant data; secondly, starting alternate iterative optimization of the fractional order particle swarms and intuitive fuzzy clustering; thirdly, comparing results with standard fuzzy means clustering results, and selecting a morphological processing method through differential images; finally, subjecting result images to range conversion and watershed segmentation to obtain a final result. The particle swarm and fuzzy means clustering based cell image segmentation method can be widely applied to various cell image based application systems and has a promising market prospect and a high application value.

The invention discloses a synechia cell image segmenting method based on a polyphase mutual exclusion level set. The circle centers of all cells are acquired through carrying out Hough transformation circle detection to quasi-circle cells in a microscopical cell image; a small circle is automatically set according to the circle center of every cell; each small circle is taken as an initial curve; independent evolution is carried out to a synechia cell area under the effect of mutual exclusion energy according to the guidance of image gradient information and the mode of simultaneously evoluting multi-level functions; mutual exclusion among different closed curves can be ensured; therefore, the synechia cell image is segmented; the problem that the traditional segmenting method cannot segment synechia crowded cell group areas is solved; and the method of the invention is advantaged by that the processing steps are few, the segmented cell outlines are natural and the method is applicable for segmenting and counting the quasi-circle cell image, has high measurement precision and strong adaptive capacity.

An artificial neural network system for image classification, formed of multiple independent individual convolutional neural networks (CNNs), each CNN being configured to process an input image patch to calculate a classification for the center pixel of the patch. The multiple CNNs have different receptive field of views for processing image patches of different sizes centered at the same pixel. A final classification for the center pixel is calculated by combining the classification results from the multiple CNNs. An image patch generator is provided to generate the multiple input image patches of different sizes by cropping them from the original input image. The multiple CNNs have similar configurations, and when training the artificial neural network system, one CNN is trained first, and the learned parameters are transferred to another CNN as initial parameters and the other CNN is further trained. The classification includes three classes, namely background, foreground, and edge.

The invention relates to an adhesion hemocyte image segmentation method based on the improved fractional differential and graph theory.The method comprises the steps of pretreating a hemocyte image by combining morphological denoising with the improved quasi-circular mask operator fractional differential algorithm aiming at the phenomenon that the hemocyte image is fuzzy and low in contrast ratio, wherein by the adoption of the improved fractional differential algorithm, cell edge details are reserved while staining contamination and grain noise of the hemocyte image are filtered out; then conducting preliminary segmentation on the pretreated image with the watershed algorithm, and mapping an over-segmentation area into a node; finally conducting resegmentation on the cell image obtained in the second step with the improved graph theory minimum spanning tree (MST) algorithm.By the adoption of the method, segmentation precision of adherent cells in the cell image can be improved.

The invention discloses a cancer cell recognition and diagnosis system, which comprises a cell image preprocessing module for selecting proper structural elements by adopting an improved mathematicalmorphology opening and closing filtering algorithm to remove background impurities and small normal discrete cells of a cell image; a cell image segmentation module used for extracting a cell nucleusregion by adopting an improved region growth algorithm, extracting a non-laminated cell body contour by adopting a watershed algorithm based on price tag control, and extracting a laminated cell contour by adopting a segmentation algorithm based on a snappe model; a cancer cell image feature extraction module used for calculating the nuclear-to-mass ratio of each region after segmentation, whetherthe cell nucleus size is uniform, whether nuclear staining is abnormal and whether the nucleus spacing is uniform. and a cell image classification and recognition module which recognizes cancer cellsthrough a neural network recognition technology. The system can be used for efficiently and accurately carrying out quantitative analysis and detection identification on cell images.

The invention discloses a cell image segmentation method based on a Mask contour. The method comprises the following steps of 1, making a data set; 2, cell feature extraction, which comprises the following steps: 2.1, constructing a feature extraction network; 2.2, performing feature multi-scale fusion; 3, establishinga multi-task branch network, mainly establishing a classification branch network, a segmentation branch network and a Center branch network, and respectively sending the fused features into the multi-task branch network for further operation; 4, generating a target Mask contour:generating the Mask contour of an initial target through deformable convolution and a Graham algorithm; and 5, carrying out segmentation Mask contour refinement on the cell image. According to the cell image segmentation method based on the Mask contour provided by the invention, the complexity of an image segmentation task is reduced, the image segmentation processing time is shortened, and the performance is improved.

A classifier engine provides cell morphology identification and cell classification in computer-automated systems, methods and diagnostic tools. The classifier engine performs multispectral segmentation of thousands of cellular images acquired by a multispectral imaging flow cytometer. As a function of imaging mode, different ones of the images provide different segmentation masks for cells and subcellular parts. Using the segmentation masks, the classifier engine iteratively optimizes model fitting of different cellular parts. The resulting improved image data has increased accuracy of location of cell parts in an image and enables detection of complex cell morphologies in the image. The classifier engine provides automated ranking and selection of most discriminative shape based features for classifying cell types.

The invention discloses a cell image segmentation model, which takes a U-Net network as a basic framework and comprises an encoder, a decoder and an attention mechanism. A residual block is adopted toreplace the encoder of the U-Net network and a convolutional layer in the decoder.

The invention discloses a cell image segmentation method based on a U-Net network, and relates to the technical field of cell image segmentation. The method comprises the following steps: step 1, an image preprocessing method: 1.1, contrast image enhancement; 1.2, performing Gaussian filtering; 2, a cervical cell segmentation method based on U-Net: 2.1, performing preliminary segmentation by usinga U-Net network; 2.2, adding a hole convolution structure and a residual structure into the U-Net network; 2.3, detecting the edges of the cells by using a clustering detection method; according to the method, a deep learning network is mainly used for segmenting a cervical cell image, and a clustering method is used for auxiliary segmentation; not only can the design difficulty of the classifierbe reduced, but also the accuracy of cancer cell diagnosis can be remarkably improved.

The invention discloses a single-cell hemoglobin determination method and device based on image processing. The method comprises the following steps of: converting a blood sample slide RGB image intoa blood sample slide HSV image, extracting a blood sample slide V channel image, and performing binarization processing on the blood sample slide V channel image according to a set first threshold value to obtain a blood sample slide binarization image; carrying out cell image segmentation processing on the blood sample slide binarization image to extract a single erythrocyte image, and carrying out binarization processing on the single erythrocyte image according to a set second threshold value to remove a central lightly-stained area of erythrocytes; graying the single erythrocyte image of the erythrocyte non-central lightly-stained area to obtain a single erythrocyte gray level image, and calculating the area and hemoglobin content of the whole erythrocyte according to the single erythrocyte gray level image. The method can be adopted to accurately calculate the area of a single erythrocyte and the hemoglobin content.