32results about How to "Automatic segmentation" patented technology

The invention discloses an ultrasonic image segmentation method of cascaded total convolution neural network for thyroid nodule, which comprises the following steps of: constructing an ultrasonic image segmentation method based on U-NET, the region of interest (ROI) is segmented from the ultrasound images of thyroid gland based on the simplified total convolution neural network (TCNN); using the VGG19-FCN network as the down-sampling layer to extract the deep features of ROI, so as to realize the automatic semantic segmentation of thyroid nodules. The simplified total convolution neural network comprises a fifth convolution layer for downsampling and a fifth upsampling layer for upsampling. The first five convolutions conv are composed of two 3x3 convolution layers and one pooling layer, each convolution layer uses ReLU as the activation function, and the last five upsampling layers are deconvolution layers. The invention provides a high-precision nodule image for identifying benign and malignant thyroid nodules, thereby playing a better auxiliary role in medical diagnosis.

The invention relates to the technical field of pelvis and pelvis artery blood vessel tree segmentation, in particular to a deep learning-based pelvis and pelvis artery blood vessel multi-level segmentation method, which is used for solving the problem that an abdomen pelvis and a pelvis artery blood vessel tree cannot be automatically, efficiently and accurately segmented in a multi-resolution CTimage in the prior art. The method comprises the following steps: step 1, preparing and labeling data; 2, preprocessing the data; 3, constructing a first-level segmentation model of the 3D convolutional neural network based on multi-level segmentation; 4, constructing a second-stage segmentation model; 5, training a first-stage segmentation model and a second-stage segmentation model by using thecalibrated data and the synthesized loss function; and 6, performing abdomen information segmentation on the input three-dimensional CT image by using the first-stage segmentation model and the second-stage segmentation model trained in the step 5. According to the method, the abdominal pelvis and the pelvis vascular tree can be automatically, efficiently and accurately segmented in the multi-resolution CT image.

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.

The invention relates to a dual-source CT coronary artery automatic extraction method. The method comprises the following steps of S1, calculating eigenvalues lambda 1, lambda 2 and lambda 3 of a Hessian matrix for each pixel point in dual-source CT data, and building a new blood vessel model equation; effectively enhancing a blood vessel structure by utilizing the new blood vessel model; enablinga blood vessel region to be relatively bright in grayscale, and enabling a background region to be relatively dark; performing Gaussian modeling on the blood vessel region and the background region by using a statistical model; S2, by utilizing a Gaussian mixture model, classifying enhanced data into two types, namely, the background region and the blood vessel region; and S3, in the blood vesselregion, restraining a neighborhood relation through the growth of a ball model, growing coronary arteries, and detecting and removing pseudo-branch blood vessels in the grown coronary arteries by utilizing hierarchical clustering-based branch detection.

The invention discloses a splitting method of a CD image of a solid engine. The method comprises the following steps: a CD faultage image data of the solid engine is read in, an artifact which is formed outside the solid engine owning to the air agitation formed in the CD detection is eliminated; the image, the artifact of which is eliminated, filters and eliminates background noise by median filter, meanwhile, the edges of each component of the solid engine are sharpened in the CD image and a casing, a propelling agent and an astropyle of the solid engine is cut in the CD image; a defect is cut from the propelling agent. By adopting the method, an outer layer air ring artifact, the casing, the propelling agent, the astropyle and various defects of the solid engine in the CD image are cut automatically and accurately, thus solving the problem of low efficiency and poor accuracy in the CD detection image of the solid engine by the manual interpretation as well as being not limited by the eigenvalue of image information.

The invention relates to an echocardiographic ventricle segmentation method and device based on deep learning and a deformation model, and aims at ironing out a defect that a conventional manual boundary marking method exerts a big negative impact on the calculation of the related indexes of a ventricle because the conventional manual boundary marking method usually consumes a large amount of manpower and material resources and the marking results of different persons are different. The method comprises the steps: carrying out the training of a ventricle coarse segmentation model through manually marked training data, and obtaining a coarse segmentation training result; calculating the central point of the coarse segmentation training result on each section, carrying out the fitting of oneline through all central points, and calculating the mean value of the distances between the central points in a direction perpendicular to the line and an outer edge of the coarse segmentation training result as the radius; carrying out the resampling according to the calculated central points and the radius, and reconstructing a three-dimensional initialization model based on a sampling result;and carrying out the fine segmentation of the ventricle coarse segmentation result through the deformation model. The method is suitable for the ventricle image processing.

The invention relates to a pulmonary nodule segmentation method for a lung CT image, and the method comprises the following steps: obtaining a lung CT image data set, and carrying out the data enhancement of the lung CT image data set; dividing the lung CT image data set after the data enhancement operation into a training set and a test set; constructing a semantic segmentation model based on the enhanced UNet structure, and training and testing the semantic segmentation model by using the training set and the test set to obtain a pulmonary nodule segmentation model; and inputting the actual lung CT image to be measured into the pulmonary nodule segmentation model, and outputting to obtain a corresponding pulmonary nodule segmentation image. Compared with the prior art, the pulmonary nodule segmentation method has the advantages that deep learning and a semantic segmentation technology are combined, an end-to-end semantic segmentation model is provided, and the lung and pulmonary nodule can be automatically, quickly and accurately segmented from the lung CT image by utilizing the ResNet101 backbone network, the context information extraction module and the spatial information extraction module in the semantic segmentation model.

An integral graph algorithm-based fabric flaw detection method is disclosed. An integral graph algorithm is used for rapidly extracting statistical characteristics of gradient energy, and the statistical characteristics of gradient energy are used for flaw detection; image learning operation on a flawless template is performed, statistics are run on characteristic distribution of the gradient energy, distribution peak values are extracted, threshold parameters are calculated in a self-adaptive manner and used for distinguishing subsequent flaws, gradient energy of a window where each pixel point of an image to be detected is positioned can be calculated via the integral graph algorithm, whether a current pixel point is a defect point can be determined based on the threshold parameters, and whether the current image is a flawed fabric can be determined after statistics are run on a total quantity of flaw points of the whole image. According to the method disclosed in the invention, based on principles of accelerated operation of integral graphs, the characteristic distribution of the gradient energy of the fabric image can be rapidly extracted, real time detection of fabric flaws can be realized, the peak values of the distribution are calculated, the self-adaptive threshold parameters for flaw determination are obtained, and accurate segmentation of fabric flaws can be realized. Via the method, real-time property and high accuracy can be ensured.

The invention relates to a probability graphic model image segmentation method based on flag fusion. The method comprises the steps of (1) acquiring a target image and multiple training images, conducting registration on the target image and the training images, and obtaining a corresponding deformation field; (2) acquiring the manual segmentation flag image of each training image subjected to registration, and obtaining a corresponding candidate flag according to deformation field reflection; (3) establishing a probability graphic model according to the training images and target image subjected to registration; (4) calculating the flag value of each pixel point in the target image by means of the local weighted voting flag fusion algorithm and the maximum posterior probability according to the probability graphic model, so that segmentation of the target image is achieved. Compared with the prior art, the method has the advantages that segmentation can be achieved automatically and segmentation precision is high.

The invention belongs to the field of computer-aided diagnosis, medical image processing, ultrasonic tomography and medical image analysis, and particularly discloses an anti-artifact feature extraction method based on an ultrasonic tomography reflection image convex target, which comprises the following steps of: (1) drawing an initial convex hull aiming at the convex target in a mode of selecting nodes; (2) calculating a geometric gravity center O1 of a target representation pixel; (3) updating and iterating the nodes; and (4) fitting and shaping the nodes, wherein a region correspondingly surrounded by a contour line obtained by connecting the shaped new nodes is a corresponding image of the extracted convex target. According to the method, details such as the overall process design and iteration conditions of the method are improved, so that the anti-artifact feature extraction method based on the ultrasonic tomography reflection image convex target can be accurate and effective in feature extraction for the reflection image of medical ultrasonic tomography equipment, and the extraction efficiency is improved. And meanwhile, the method has good robustness for artifacts and noise.

The invention relates to a mammary gland X-ray auxiliary diagnosis technology, and aims to provide a method for automatically segmenting mammary gland calcification points based on deep learning. Themethod comprises the steps of making a data set; pre-fetching the data; constructing a deep convolutional neural network; training a deep convolutional neural network by using the normalized data set;and reasoning a to-be-detected image by using the trained deep convolutional neural network. According to the method, by introducing related technologies of deep convolutional neural network learningand training, all calcification points on the mammary gland X-ray image can be rapidly and automatically segmented. Based on the application of the invention, the accuracy of judging canceration by adoctor can be improved.