124 results about "Manual segmentation" patented technology

Segmenting the prostate boundary is essential in determining the dose plan needed for a successful bracytherapy procedure—an effective and commonly used treatment for prostate cancer. However, manual segmentation is time consuming and can introduce inter and intra-operator variability. This present invention describes an algorithm for segmenting the prostate from two dimensional ultrasound (2D US) images, which can be full-automatic, with some assumptions of image acquisition. Segmentation begins with the user assuming the center of the prostate to be at the center of the image for the fully-automatic version. The image is then filtered to identify prostate edge candidates. The next step removes most of the false edges and keeps as many true edges as possible. Then, domain knowledge is used to remove any prostate boundary candidates that are probably false edge pixels. The image is then scanned along radial lines and only the first-detected boundary candidates are kept the final step includes the removal of some remaining false edge pixels by fitting a polynomial to the image points and removing the point with the maximum distance from the fit. The resulting candidate edges form an initial model that is then deformed using the Discrete Dynamic Contour (DDC) model to obtain a closed contour of the prostate boundary.

The invention discloses a nasopharyngeal-carcinoma (NPC) lesion automatic-segmentation method and nasopharyngeal-carcinoma lesion automatic-segmentation systems based on deep learning. The method comprises: carrying out registration on a PET (Positron Emission Tomography) image and a CT (Computed Tomography) image of nasopharyngeal carcinoma to obtain a PET image and a CT image after registration;and inputting the PET image and the CT image after registration into a convolutional neural network to carry out feature representation and scores map reconstruction to obtain a nasopharyngeal-carcinoma lesion segmentation result graph. The method carries out registration on the PET image and the CT image of the nasopharyngeal carcinoma, obtains a nasopharyngeal-carcinoma lesion by automatic segmentation through the convolutional neural network, and is more objective and accurate as compared with manual segmentation manners of doctors; and the convolutional neural network in deep learning isadopted, consistency is better, feature learning ability is higher, the problems of dimension disasters, easy falling into a local optimum and the like are solved, lesion segmentation can be carried out on multi-modal images of the PET-CT images, and an application range is wider. The method can be widely applied to the field of medical image processing.

An automated method for determining the location of the left ventricle at user-selected end diastole (ED) and end systole (ES) frames in a contrast-enhanced left ventriculogram. Locations of a small number of anatomic landmarks are specified in the ED and ES frames. A set of feature images is computed from the raw ventriculogram gray-level images and the anatomic landmarks. Variations in image intensity caused by the imaging device used to produce the images are eliminated by de-flickering the image frames of interest. Boosted decision-tree classifiers, trained on manually segmented ventriculograms, are used to determine the pixels that are inside the ventricle in the ED and ES frames. Border pixels are then determined by applying dilation and erosion to the classifier output. Smooth curves are fit to the border pixels. Display of the resulting contours of each image frame enables a physician to more readily diagnose physiological defects of the heart.

The invention relates to the medical apparatus field, aims to propose an improved full convolutional neural network to realize machine segmentation of a brain tumor nuclear magnetic resonance image, avoids low efficient and instability existing in manual segmentation, provides the rapid and reliable brain tumor segmentation result and provides accurate bases for diagnosis, treatment and operationof brain tumors. The method comprises steps that 1), an image is selected; 2), a full convolutional neural network FCN model is constructed; 3), the segmentation result is tested, when the FCN model is trained, the trained model is utilized to predict the tumor position and the boundary size of any brain tumor image, corresponding evaluation indexes are utilized to evaluate the segmentation result, and the FCN model can be better ameliorated. The method is advantaged in that the method is mainly applied to NMR image processing occasions.

The invention discloses an image omics based lesion tissue auxiliary prognosis system and method. The method comprises the steps of extracting image data of lesion parts with an automatic or manual segmentation method from a big-data-volume patient image database; according to segmentation results of lesion part images, extracting image phenotypic characteristics of each lesion part, and finishing feature extraction of image data of all lesion parts in the patient image database; and based on characteristic data and clinical information data of each lesion part, performing training data set and test data set classification on the data in the patient image database, performing pathologic analysis, clinical stage analysis, gene mutation prediction and survival time prediction of the lesion parts in the training data set with a computer automatic identification method, and performing verification in the test data set. The method is capable of performing qualitative and quantitative prediction analysis on a specific individual and providing credible prediction and analysis results.

The invention discloses an SVM (support vector machine)-based medical image blood vessel recognition method. The method includes the following steps that: since an SVM classifier is obtained after cross validation and optimization are performed on an SVM model trained by samples automatically selected by an FCM by using an SVM model which is trained by samples obtained through the manual segmentation of experts, segmentation results are more accurate; the SVM is adopted to segment a blood vessel, namely, pixels are divided into a foreground (blood vessels) and a background (other parts), and the blood vessel part is extracted out; morphological processing and thresholding are performed, and therefore, a blood vessel network can be enhanced, the branches and intersections of the blood vessels are reserved; and an obtained image is converted into a binary image, so that blood vessel distribution can be reflected more directly. According to the method of the invention, the FCM, the SVM and the morphological image processing are combined together, so that a recognition effect is better.

The invention discloses an amygdaloid nucleus spectral clustering segmentation method based on resting state function connection, and the method is used for carrying out the automatic high-efficiency segmentation of an encephalic region based on a spectral clustering algorithm according to the similarity of internal voxel functions of an amygdaloid nucleus. The method comprises the steps: firstly carrying out the preprocessing of resting state magnetic resonance data; secondly carrying out the extraction of the encephalic region of the amygdaloid nucleus; thirdly carrying out the connection calculation of internal voxel whole-brain functions of the amygdaloid nucleus; and finally carrying out the spectral clustering segmentation of a function connection matrix. The automatic segmentation algorithm proposed by the invention and an amygdaloid nucleus clinic dissection result are enabled to be greatly consistent with each other, and the stability and noise interference resistance are enabled to be more satisfying. Compared with a conventional manual segmentation method, the method is simpler and more convenient and efficient, and is high in repeatability.

The utility model relates to a segmentation method of active web page, and is characterized in that the method first receives the content streams of web page and then builds a DOM tree, then makes the nodes of the DOM tree to feature codings, next compares the corresponding nodes of the DOM trees to build common blocks and customization blocks. The utility model can understand and identify the common parts (common blocks) sharing among multi-pages and the special parts of different variation rules (customization blocks) according to the dynamic characteristic and the structural characteristic of the web pages, and dynamically divide the web pages without human interference. The utility model provides a solution with a good expandability, lowers the labour costs for manual segmentation, and can be widely used in technology field of active web page.

The invention relates to a level set retinal vessel image segmentation method with shape prior being fused. The method comprises the following steps: 1) enhancing a retinal vessel image by utilizing a morphological operator and through Gaussian convolution; 2) roughly segmenting the retinal vessel image through anisotropic property of a Hessian matrix and an improved vessel response function, and serving the images as shape constrains and initial information; and 3) constructing a retinal vessel segmentation level set model comprising a local area energy fitting item, a shape constraint item, a level set function regularity maintenance item, a length punishment item and a weight area restraint item by utilizing shape prior and retinal image data. The segmentation method is high in segmentation result accuracy, can replace manual segmentation, and can play an important helping role in diagnosis and treatment of clinical related eye diseases, and has a higher clinical application value.

The invention designs a lung parenchyma CT image segmentation method based on a deep convolutional neural network. The method comprises the steps that 1, using The VGG-16 network improved by cavity convolution to extract the convolution characteristics of lung parenchyma in lung CT images;; 2, constructing a multi-scale feature supercolumn descriptor based on multi-layer extraction; And 3, learning a nonlinear predictor for classifying pixels. Results show that the method can realize accurate lung parenchyma image segmentation, compared with the prior art, the method has the advantages that the segmentation accuracy and speed are improved, and the problems of low manual segmentation efficiency, low accuracy, excessive consumption of manpower resources and the like of the lung parenchyma image are successfully solved.

The invention belongs to the technical field of image segmentation, and specifically relates to a 3D full-automatic glioma segmentation method combining a Volume of Interest and a GrowCut algorithm. The method first expands a Bounding box algorithm to 3D, utilizes the algorithm to extract a Volume of Interest (VOI) containing glioma, then utilizes a reflection symmetry algorithm to estimate the VOI and overcomes the defects of Bounding box when the glioma is detected to cross over a median sagittal plane, and finally marks pixel points in an image based on the accurate VOI, and thus a semi-automatic 2D GrowCut algorithm is optimized to a full-automatic 3D segmentation method. While accurately segmenting the glioma, the method is more rapid in theory and reality compared with the 2D segmentation algorithm of the same principle, and is better in convenience and feasibility compared with a manual segmentation method. As an image segmentation method, the 3D full-automatic glioma segmentation method provided by the invention can serve as a powerful auxiliary tool for clinical diagnosis of glioma.

Systems, computer-readable media, and methods for assessing morphometric measures of spinal vertebrae and inter-vertebral discs in a quantitative manner using imaging data and human body weight measures, for identification and prediction of structural spine pain, are described. The systems, media, and methods of the present disclosure utilize simple measurements of axial areas of vertebrae using endplate data of routinely acquired digital imaging pixels and body weight. More specifically, they provide a calculation pressure value based on one or more ratios from body weight and measurements of spinal structures and regions, which are readily determinable, e.g., through manual segmenting or automated programming of image analysis software.

The invention discloses a multi-objective organ-at-risk automatic segmentation method, device and system based on deep learning. The method comprises the steps of receiving an input image of a patient; carrying out format conversion on the input image of the patient, and converting the input image into JPEG format data; inputting the JPEG format data into an Overfeat positioning detection networktrained according to a physician manual segmentation result, and automatically selecting a region of interest containing multi-objective organ-at-risk; inputting the automatically selected region of interest into an FCN initialization segmentation network, and carrying out contour inference; carrying out the coordinate processing of the initial boundary contour obtained through the contour inference and the received artificial mark boundary, mapping the initial boundary contour and the received artificial mark boundary to an input image, extracting a DAISY feature, and obtaining a DAISY feature image; and inputting the DAISY feature image into a deep belief network trained according to a physician manual segmentation result to obtain an accurate segmentation boundary of organ-at-risk, namely a segmentation result.

The invention discloses a server thermal fault monitoring and diagnosing method based on infrared images. The server thermal fault monitoring and diagnosing method comprises acquiring infrared images and visible-light images of a server in different thermal fault states; carrying out tilt correction processing on the images in each thermal fault state by means of standardization method based on image registration, and determining the region-of-interest including the server in each infrared image by means of a manual segmentation method; converting the image of the determined region-of-interest into a corresponding gray scale image, and extracting an image entropy feature from the gray scale image; carrying out dimension reduction processing on the obtained global entropy and the local entropy row-column mean value feature by means of a PCA method; wherein the feature main component obtained after dimension reduction is used for training a support vector machine classifier; acquiring the infrared image and the visible-light image of the server to be detected, and diagnosing the thermal fault state type of the server by means of the trained SVM model. According to the invention, the management efficiency of management personnel in the data center is greatly improved, and engineering debugging and system maintenance are facilitated.

The invention discloses a high-resolution remote sensing image house extraction method based on morphological house indexes. The morphological house indexes are constructed on the basis of the morphological algorithm according to the characteristics of high brightness, isotropy and similar rectangle degree of a house on a high-resolution remote sensing image, and a remote sensing image house is automatically extracted according to the morphological house indexes. On this basis, morphological shadow indexes are derived from the morphological house indexes according to similar spatial characteristics and opposite optical characteristics of a shadow and the house, house extraction is restrained through the morphological shadow indexes, and consequently the accuracy degree of house extraction is further optimized. According to the high-resolution remote sensing image house extraction method based on the morphological house indexes, manual segmentation and artificial training are not needed, full-automatic extraction of the remote sensing image house can be achieved; the morphological shadow indexes are added for restraint, the accuracy degree and the similar rectangle degree of house extraction can be obviously improved.

The embodiment of the invention provides a CT image abdominal multi-organ segmentation method and device. The method comprises the steps of obtaining a first image sequence and a second image sequenceaccording to an abdominal CT image sequence to be segmented; inputting the first image sequence and the second image sequence into a multi-organ segmentation model, and outputting a multi-organ segmentation result of each abdominal CT image in the abdominal CT image sequence to be segmented; wherein the scale of the image in the first image sequence is greater than that of the image in the secondimage sequence; wherein the multi-organ segmentation model is obtained after training based on abdominal CT image sequence sample data and a predetermined manual segmentation marking result. According to the CT image abdominal multi-organ segmentation method and device provided by the embodiment of the invention, the multi-organ segmentation result of each abdominal CT image in the abdominal CT image sequence to be segmented is obtained according to the context information of the organ levels obtained by the two different scales, and a more accurate and stable segmentation result can be obtained.

The invention discloses a CT image-based rib segmentation method, a CT image-based rib segmentation device, a detection device, a computer readable storage medium and electronic equipment. According to the method, the position information of each pixel in a CT image is acquired; the position information and the CT image are inputted into a neural network model; the image of each rib and the corresponding serial number of the image are automatically generated by utilizing the neural network model; and therefore, the workload of medical personnel in the radiology department is greatly reduced, the efficiency of rib segmentation is improved, the situation of wrong manual segmentation number can be avoided, and the accuracy of the rib segmentation is improved.

The invention discloses a fundus image lesion segmentation method based on deep network aggregation, and the method comprises the following steps: 1), obtaining a plurality of fundus lesion images, carrying out the manual segmentation of the lesion contour in each fundus lesion image, obtaining a true value label, and constructing a training set and a test set; 2) adding a deep aggregation networkmodule into a backbone network of the U-Net model; 3) migrating the U-Net model obtained in the step 2) to focus segmentation of the fundus image, training the U-Net model, and taking the trained U-Net model as a fundus image focus segmentation model; and 4) segmenting the fundus image to be segmented by using the fundus image lesion segmentation model. The method can effectively solve the problem of poor fundus image lesion segmentation effect based on the deep convolutional neural network in the prior art.

The invention discloses an abnormal sound monitoring method based on large-scale farm mammals, belongs to the field of sound recognition, and particularly relates to an unsupervised sound recognitionmethod. The method mainly comprises the following parts: 1. spectrogram analysis: analyzing collected voice frequencies to determine feasibility of a voice recognition scheme; 2. voice frequency noisereduction: performing noise reduction processing on the voice frequencies to improve accuracy of voice recognition; 3. unsupervised voice frequency segmentation: simplifying a voice frequency processing process without manual segmentation to obtain voice frequency segments containing required sound events; 4. voice frequency feature extraction: adopting a Mel frequency cepstrum coefficient as a feature extraction technology; and 5. unsupervised classification: adopting an unsupervised classification method as a K-means algorithm. According to the method provided by the invention, unsupervisedsound recognition of large-scale farm animals is realized by adopting an unsupervised voice frequency segmentation technology and a K-means classification method, combining a spectrum and time spectrum analysis technology, a voice frequency noise reduction technology and a Mel frequency cepstrum coefficient feature extraction technology.

The invention relates to an aortic valve ultrasonic image segmentation method based on probability distribution and continuous maximum flow. The method comprises the following steps: 1, acquiring medical ultrasonic image data of a human body aortic valve short axis, and extracting a five-frame prior image at equal intervals; 2, segmenting the five-frame prior image; 3, constructing a two-dimensional gray-distance histogram; 4, calculating to obtain a comprehensive probability estimation function through the two-dimensional gray-distance histogram; 5, respectively calculating a respective independent probability estimation function; 6, respectively calculating the pixel gray values which can respectively represent the foreground and background for the five-frame prior image; 7, solving an independent probability estimation map for the current image to be segmented; 8, respectively measuring the similarity for the foreground area and the manual segmentation result of the five-frame prior image; and 9, obtaining the segmentation result. Compared with the prior art, the aortic valve ultrasonic image segmentation method is stable, reliable, convenient to implement and suitable for actual clinical application.

The invention relates to a prostate transrectal ultrasound (TRUS) image segmentation method based on a convolutional neural network. The method comprises the following steps: 1) extracting multi-scaleconvolution features in a prostate TRUS image by using an extended spatial pyramid pooling (DSPP) module with global feature coding semantic information; 2) in order to process the missing boundary in the segmented prostate shadow region, proposing a super-hybrid feature (SHF) capable of mixing low-level features and high-level features; and 3) a novel network composed of an encoder and a decoderis provided. The result shows that the method can achieve the precise segmentation of the prostate TRUS image, improves the segmentation accuracy and speed compared with the prior art, and successfully solves problems that the manual segmentation efficiency of the prostate TRUS image is low, the accuracy is low, and excessive manpower resources are consumed.

The invention is suitable for the technical field of image processing, and particularly relates to an ultrasound image segmentation method and system. The method includes the following steps that coarse registration is conducted on a statistical shape model and collected three-dimensional ultrasound data of a specified organ to obtain initialized coordinate transformation parameters; according to the initialized coordinate transformation parameters, an image segmentation algorithm based on particle filter is used for conducting iteration segmentation on the three-dimensional ultrasound data, wherein the statistical shape model is a combination of an average value obtained by training manual segmentation results of a plurality of high-definition three-dimensional data and a group of feature vectors of a representation change mode. Therefore, the problem that more artificial participation is needed in manual segmentation and semi-automatic segmentation is solved. Compared with an existing full-automatic segmentation method, the ultrasound image segmentation method and system solve the problems of low image resolution and segmentation accuracy under the image fuzzy condition.

The invention provides a full-automatic CT image kidney segmentation method. Multi-template-based image matching algorithm is used and a multi-resolution two-step algorithm strategy is adopted. A region of interest of the kidney in a low-resolution CT image is firstly positioned, and then kidney tissues in a region of interest in a high-resolution CT image are precisely segmented. The experimental result shows that in view of renal arteriography and urography CT images, accurate segmentation results can be provided, the workload of manual segmentation by a doctor can be greatly reduced, and the efficiency and effects of diagnosis and treatment can be effectively improved.

The invention relates to medical instrument, applies a computer image processing technology to segmentation of a brain tumor magnetic resonance image, avoids the defects of low efficiency and instability existing in manual segmentation, provides a rapid and reliable brain tumor segmentation result by use of a computer algorithm and provides an accurate basis for diagnosis, treatment and surgical guidance of a brain tumor. According to the technical scheme, a brain tumor segmentation method in combination with a convolutional neural network and fuzzy inference comprises the following steps of (1) selecting an image; (2) constructing a CNN (Convolutional Neural Network) model; (3) carrying out nonlinear mapping; and (4) establishing a fuzzy inference system. The brain tumor segmentation method in combination with the convolutional neural network and fuzzy inference is mainly applied to processing of medical images.

The invention discloses a server and a Chinese character segmentation method and device and belongs to the technical field of search engines. The Chinese character segmentation method includes: receiving a word segmentation instruction; acquiring a first Chinese character set; acquiring retrieving information corresponding to each character in the first character set according to preset corresponding relations; acquiring multiple combined characters and retrieving probabilities according to the first character set and the retrieving information corresponding to each character in the first character set; performing path combination according to the characters included in the multiple combined characters; acquiring the retrieving probability of each path; determining the path with the highest probability; performing segmentation on keywords according to the combined characters included in the path with the highest probability. Manual segmentation is omitted, independence on tools of dictionaries and the like is not needed, convenience in operation is achieved, data sources are dynamically updated, wrong segmentation modes can be rapidly corrected, high distinguishing degree is achieved for new characters, and accuracy in segmentation is improved.

The invention relates to the technical field of brain tumor medical image segmentation, in particular to a brain tumor medical image segmentation method based on deep learning, which comprises the steps of training a segmentation model, receiving the to-be-segmented brain tumor medical image data information, performing the segmentation processing on the received to-be-segmented brain tumor medical image data information, and outputting a segmentation result. The brain tumor medical image segmentation system based on deep learning comprises a segmentation model training module, a brain tumor medical image receiving module, a brain tumor medical image segmentation processing module and a segmentation result output module. According to the medical brain tumor image segmentation method basedon deep learning, deep learning training is carried out on the segmentation model, the segmentation model trained through deep learning carries out segmentation processing on the received medical brain tumor image data information to be segmented, the segmentation result is accurate, and the problem that a traditional manual segmentation method is time-consuming and labor-consuming is solved.

The invention discloses a human brain MRI hippocampus detection and segmentation method based on deep learning. The method includes the steps of firstly, conducting data pretreatment on human brain MRI data and labels both of which are obtained by means of various channels and creating a model; secondly, determining a final model and determining super parameters of the final model; thirdly, training and estimating the model; finally, predicating the model, comparing a model predication result with a manual segmentation result, observing the effect, and conducting analysis to obtain a final predicated image. According to the method, historical manual segmentation result image information is fully utilized, not only can detection and segmentation be automatically and efficiently carried out,but also convenience is provided for solving the problems of shortage of doctors in the image department, a poor primary medical capability, a great disparity in the proportion of doctors and patients and the like. When the model is subjected to fitting, L2 regular terms are added for the first time, and regular term super parameters are also added, correspondingly the variance of the model is reduced, and the effect is obviously improved. Through several experiments, the network depth is increased to five layers, and the effect of the model is also improved.

The invention discloses a medical image automatic segmentation method based on deep learning, and the method specifically comprises the steps: (1) obtaining a manual segmentation result of an original medical image of a patient and a target region, and constructing a training data set related to a segmentation task; (2) designing a deep U-shaped convolutional neural network model fusing a residual module and an attention mechanism; (3) constructing a mixed loss function by using Dice and binary cross entropy; (4) performing network training by using the training data set; and (5) segmenting a target area in the test image by using the trained network. The invention relates to an end-to-end medical image full-automatic segmentation method, which can effectively solve the problems of fuzzy segmentation boundary, low small target detection rate, unstable training process, difficulty in convergence and the like in the current deep learning image segmentation field.