71 results about "Lesion feature" patented technology

A method of calculating a disease assessment by analyzing a medical image, comprising (1) extracting at least one lesion feature value from the medical image; (2) extracting at least one risk feature value from the medical image; and (3) determining the disease assessment based on the at least one lesion feature value and the at least one risk feature value. The method employs lesion characterization for characterizing the lesion, and risk assessment based on the lesion's surroundings, i.e., the environment local and distal to the lesion. Computerized methods both characterize mammographic lesions and assess the breast parenchymal pattern on mammograms, resulting in improved characterization of lesions for specific subpopulations, combining the benefits of both techniques.

An automated method for determining prognosis based on an analysis of abnormality (lesion) features and parenchymal features obtained from medical image data of a patient. The techniques include segmentation of lesions from radiographic images, extraction of lesion features, and a merging of the features (with and without clinical information) to yield as estimate of the prognosis for the specific case. An example is given for the prognosis of breast cancer lesions using mammographic data. A computerized image analysis system for assessing prognosis combines the computerized analysis of medical images of cancerous lesions with the training-based methods of assessing prognosis of a patient, using indicators such as lymph node involvement, presence of metastatic disease, local recurrence, and / or death. It is expected that use of such a system to assess the severity of the disease will aid in improved decision-making regarding treatment options.

An apparatus and method for determining a characteristic of a selected skin lesion. A capture device produces data representing an image of an object. A processing device processes the data to derive a three-dimensional model of the skin lesion, which is stored in a database. A reporting device indicates at least one specific property of the selected skin lesion. The processing device determines a change in at least one specific property of the skin lesion by comparing the three-dimensional model with at least one previously derived three-dimensional model that is stored in the database.

A method and system for in-vivo characterization of lesion feature is disclosed. Using a non-invasive medical imaging apparatus, an image of an interior region of a patient's body is obtained. The interior region may include lesion feature (such as plaques) components from a list of components. The lesion feature components are identified by classifying each point in the image as either corresponding to one of the lesion feature components in the list of components or not, using image intensity information and image morphology information, a first relationship (such as an intensity score) correlating image intensity information with the components in the list of components and a second relationship (such as a morphology score) correlating image morphology information with the components in the list of components. Further, a variety of lesion feature characteristics is derived from the result of the classification.

The invention discloses a premature infant retinal image classification method and device based on an attention mechanism, and the method comprises the steps: carrying out the preprocessing of a to-be-recognized two-dimensional retinal fundus image, and obtaining a preprocessed two-dimensional retinal fundus image; inputting the preprocessed two-dimensional retinal fundus image into a pre-traineddeep attention network model, and outputting the classification result of the image to identify a premature infant retinopathy ROP image, wherein the deep attention network model is formed by respectively adding a complementary residual attention module and a channel attention SE module behind a third residual layer and a fourth residual layer of the original ResNet18 network. According to the method, rich and important global and local information can be obtained, so that the network can learn correct lesion features, the classification network can better solve the problem of great data imbalance between lesions and backgrounds, and the classification performance of the deep attention network model is improved.

The invention discloses a deep learning-based diabetic retina image classification method and system in the technical field of artificial intelligence. The method comprises steps: a fundus image is acquired; the same fundus image is imported to a microvascular tumor-like lesion recognition model, a hemorrhagic lesion recognition model and an exudative lesion recognition model for recognition; andaccording to recognition results, lesion feature information is extracted, a trained SVM classifier is then adopted to classify the extracted lesion feature information, and a classification result isacquired, wherein the microvascular tumor-like lesion recognition model is obtained when a microvascular tumor-like lesion candidate area in the fundus image is extracted and is then inputted to a CNN model for training, and the hemorrhagic lesion recognition model and the exudative lesion recognition model are obtained when a hemorrhagic lesion area and an exudative lesion area in the fundus image are marked and are then inputted to an FCN model for training. The requirements on the description ability of the network model are reduced, the model is easy to train, a lesion focus area is positioned and delineated for different lesions, and clinical screening by a doctor is facilitated.

The present application provides a recognition method and a system for pathological images. The recognition system comprises an image receiving device used for receiving a to-be-recognized pathological image; a feature extracting device used for extracting the features of the pathological image by utilizing multiple stages of feature processing units arranged in a convolutional neural network; anda recognition device used for conducting the fusion treatment on feature images outputted by at least two stages of feature processing units and then recognizing the pathological information in the pathological image based on the pathological features of fused images. According to the invention, low-dimensional and high-dimensional pathological features are effectively retained, so that lesion features can be accurately recognized. Meanwhile, the lesion features are accurately mapped to the pixel positions of the pathological image, so that the auxiliary diagnosis accuracy is improved.

The invention provides a diabetic retinopathy grade classification method based on deep learning. The diabetic retinopathy grade classification method comprises the steps of: constructing a sample library; removing backgrounds and noise of ophthalmoscope photographs in the sample library; normalizing the images of different brightness and different intensity to the same range by adopting a local mean value subtracting method; adopting random stretching and rotating methods for different samples for data augmentation, and constructing a training set and a test set; training an initial deep learning network model by establishing an input portion architecture, a multi-branch feature transformation portion architecture and an output portion architecture separately; and inputting samples to betested into the trained initial deep learning network model for diabetic retinopathy grade classification. Compared with the traditional processing method, the diabetic retinopathy grade classification method gets rid of the dependence on prior knowledge, and has good generalization ability; and by adopting the designed multiple grades, a small-sized convolution kernel can be used for extracting very tiny lesion features, thereby making the classification results more reliable.

A computer-aided diagnosis (CAD) method includes extracting a lesion feature value of a lesion feature of a lesion from a captured lesion image; receiving additional information; and diagnosing the lesion based on a combination of the extracted lesion feature value and the received additional information.

A computer-aided diagnosis (CAD) scheme to aid in the detection, characterization, diagnosis, and / or assessment of normal and diseased states (including lesions and / or images). The scheme employs lesion features for characterizing the lesion and includes non-parametric classification, to aid in the development of CAD methods in a limited database scenario to distinguish between malignant and benign lesions. The non-parametric classification is robust to kernel size.

The invention discloses a lesion monitoring method and device, a computer device and a storage medium. The lesion monitoring method comprises the following steps: inputting the sample data of an original CT image into a preset segmentation model for segmentation calculation, and outputting segmented liver image data; inputting the sample data of the original CT image and the segmented liver imagedata into a preset identification model for calculation, and outputting an identification result, wherein the segmentation model and the recognition model are respectively trained through a first convolution neural network and a second convolution neural network, and the first convolution neural network and the second convolution neural network are in cascade arrangement. According to the method,the neural network is used for learning liver features and lesion features in the original CT image, and the relation between the CT slices and the tags is established through the two cascaded full-convolution neural networks to distribute the task training model, so that the optimal network parameters can be found as soon as possible, and the disease treatment efficiency and the accuracy of disease analysis are improved.

The invention discloses a feature data mining and neural network-based tumor classification method. The method includes the following steps that: the man-made scoring data of the effective lesion features of a tumor ultrasonic image are selected as an original feature data set; a bi-clustering algorithm is adopted to acquire effective local diagnosis modes from the original training data set; features of higher levels are extracted according to the effective local diagnosis modes, so that new feature vectors are formed; with the new feature vectors adopted as the input of a neural network, training is carried out, so that an effective multi-class classifier can be obtained; and feature vectors are extracted from a test sample by using the same manner mentioned above, and the multi-class classifier which is obtained through training is utilized to classify the feature vectors, and the specific classification result of a tumor can be obtained. With the feature data mining and neural network-based tumor classification method adopted, the defect that a traditional computer-assisted method is limited to low-level image features can be eliminated; and higher-level effective features are mined from a large number of man-made scoring feature data sets, and the classifier which can identify many types of tumors can be obtained through training by using a popular neural network classification method.

The invention relates to the technical field of artificial intelligence, and provides a fundus image recognition method and device, fundus image recognition equipment and a storage medium. The methodcomprises: obtaining an fundus image; extracting first target data from the fundus image, and performing redundancy elimination processing on the first target data to obtain a first central concave feature; generating a macular area mask according to the first central concave feature; intercepting a macular area in the fundus image through a macular area mask to obtain a macular area image; and identifying senile macular degeneration lesion features in the macular area image, and classifying the macular area image according to the senile macular degeneration lesion features. The macular area image is cut from the fundus image through the mask generation model, the macular area image is classified according to AMD focus features in the macular area image, feature data in the macular area image is obvious and easy to recognize, and therefore the classification accuracy of the fundus image is effectively improved.

The invention discloses a disease recognition method based on a lightweight twin convolutional neural network. The method comprises the following steps of constructing a fine-grained lesion feature joint training model, wherein the fine-grained lesion feature joint training model comprises a data generator, the data generator is connected with a feature extractor, the feature extractor is connected with the twin convolutional neural network, and the twin convolutional neural network is connected with the feature discrimination network, training the fine-grained lesion feature joint training model, and generating a fine-grained lesion feature recognition model based on the fine-grained lesion feature joint training model with the minimum loss function value, and inputting the to-be-recognized image into the fine-grained lesion feature recognition model, and outputting a corresponding skin disease category. According to the method for carrying out positive and negative sample joint training based on the twin convolutional neural network, the model can extract more discriminative features, the conditions of small inter-class difference and large intra-class difference of lesion imagefeatures in an original data set are effectively relieved, and the feature discrimination capability of the model is enhanced.

The invention discloses a bi-clustering mining and AdaBoost-based tumor classification method. The method comprises the following steps of: firstly selecting digitalized scoring data of tumor lesion features to construct an original data set, screening features effective for distinguishing benign and malignant tumors from original features according to feature statistic information, mining important tumor diagnosis modes hidden behind the feature scoring data from the feature scoring data by utilizing a bi-clustering algorithm, and determining benign and malignant attributes of the diagnosis modes by adoption of support rate indexes according to benign and malignant priori knowledges, so as to convert locally consistent modes into effective diagnosis rules; constructing a simple weak classifier which is capable of carrying out classification in different feature spaces by adoption of a method of pairwise coupling benign and malignant rules, wherein the weak classifier takes similarity of matching between test samples and the benign and malignant rules as a classification rule; and finally training a high-correctness strong classifier from the weak classifier by adoption of an AdaBoost integration algorithm. The method plays an important role in improving the clinical diagnosis correctness of tumors.

The invention discloses a colonoscope pathological section screening and segmenting system based on a sliding window. The system comprises a computer, a screening and segmenting model is stored in a memory of the computer; when the system works, a pathological section picture with an original giant size is cut into small image blocks which can be input by a computer by using a sliding window; in order to better extract lesion features of an image, a feature extraction module based on massive natural image classification tasks is used as an image feature extractor, image resolution is recoveredlayer by layer based on a previous feature map and semantic information of a corresponding position, so that a more accurate lesion area segmentation result is obtained, and finally, the probabilityof illness of a patient is output according to the generated segmentation result. By means of the system, screening and diagnosis processes of pathologists can be effectively assisted, film reading pressure and time cost are greatly reduced, and the system has important significance in medical development of underdeveloped areas.

Extraction means analyzes a character string of an interpretation report to extract lesion portion information recorded in the interpretation report and a first lesion feature recorded in the interpretation report, and analysis means performs an image analysis process corresponding to the lesion portion information with respect to the image data corresponding to the interpretation report to acquire a second lesion feature based on a result of the image analysis process. Determination means determines whether the first lesion feature matches the second lesion feature, and registration means registers the image data corresponding to the interpretation report as learning correct answer data in a case where it is determined that the first lesion feature matches the second lesion feature.

Automated diagnostic decision support (104) in the imaging of potentially malignant lesions is distributed and streamlined to protect patient confidentiality and to lower bandwidth and transaction costs. At a client hospital site (108a, 108b), a software agent (132) monitors a database and responsively accesses an image of a lesion and ground truth that the lesion is malignant / benign (S310-S330).After computing at least one feature of the lesion based on the image (S340, S350), the software agent transmits the feature(s) and ground truth externally from the hospital, to a central diagnostic decision support server (S360, S370). When a client hospital site needs automatic diagnostic support, the lesion feature(s) of the new patient are likewise extracted and transmitted to the external server in a query message (S440). The classifier located on the server will return a diagnosis (benign / malignant) and a confidence level (S450, S460).

The invention discloses a method for measuring lesion characteristics of hypertensive retinopathy. The method comprises the following steps: acquiring a fundus image; recognizing an optic disc region of the fundus image and dividing the fundus image into at least three regions including a first region, a second region and a third region based on the optic disc region; performing artery and vein segmentation on the fundus image by using an arteriovenous segmentation model based on deep learning to obtain arteriovenous segmentation results, wherein the arteriovenous blood vessel marking results include an artery marking result, a vein marking result and a small blood vessel marking result; and based on the three regions and the arteriovenous segmentation result, measuring lesion characteristics in the fundus image, wherein the lesion characteristics comprise at least one of arteriovenous cross-indentation characteristics, arteriovenous local stenosis characteristics and arteriovenous general stenosis characteristics. According to the present disclosure, it is possible to provide a measurement method and a measurement system for efficiently and objectively measuring lesion characteristics of hypertensive retinopathy.

The invention belongs to the field of medical image classification, and relates to deep learning, in particular to a diabetic retinopathy classification device based on local lesion features, which isused for solving the problems of poor classification effect of a depth model on a fundus image, lack of interpretability of a result and the like. According to the method, the global features and thelocal lesion information of the fundus image are extracted respectively, so that the information in the image with serious lesion degree can be fully utilized, and the problem of poor application effect of deep learning in the field of sugar network lesion classification caused by insufficient and unbalanced data volume is solved; in addition, in the local focus information extraction process, the labeling result of the local focus information in the fundus image is output, the problem that the interpretability degree of a deep learning model result is low is solved, and the auxiliary effecton disease diagnosis of ophthalmologists is improved.

The invention discloses a puncture method based on phase registration. The puncture method comprises the steps that any 3D image in 4D images is selected as a planning image for planning; acquiring a breathing curve of the patient; acquiring the position of a lesion feature in the 4D image to obtain a lesion feature motion curve; performing phase matching on the respiration curve and the focus characteristic motion curve to obtain a mapping relation between the respiration curve and the focus characteristic motion curve; according to the phase point of the planning image on the focus characteristic motion curve and the mapping relation, obtaining the corresponding amplitude of the planning image in the respiration curve, and recording the amplitude as a target amplitude; monitoring the breathing of the patient in real time, and executing the planning result when the breathing amplitude of the patient reaches the target amplitude. The corresponding relation is established between the phase point of the planned image and the breathing amplitude of the breathing curve, the puncture phase can be prompted to be reached through the amplitude change of the external breathing curve, and precise puncture is achieved.

The invention discloses a breast cancer HER2 state prediction method and related equipment, and the method comprises the steps: inputting a to-be-predicted breast MR image into a first classification model and a second classification model, and controlling the first classification model and the second classification model to determine that the HER2 category corresponding to the mammary gland MR image is an HER2 high expression category, an HER2 low expression category or an HER2 zero expression category. According to the application, the first classification model and the second classification model are adopted to learn the mammary gland MR image carrying mammary gland lesion features in different aspects, so that lesion information mined in different aspects is associated, and the HER2 category is determined to be an HER2 high expression category, an HER2 low expression category or an HER2 zero expression category through the associated lesion information. Therefore, the accuracy of identifying the HER2 category as an HER2 high expression category, an HER2 low expression category or an HER2 zero expression category is improved.

The invention provides a three-dimensional precise intraluminal radiation therapy method for cancer treatment. The method includes the steps: (1) performing three-dimensional quantitative lesion measurement and analysis on three-dimensional scanned images of lesion lumens; (2) performing comprehensive analysis according to three-dimensional quantitative lesion measurement results and lesion features, and calculating a three-dimensional lesion radiation dose distribution diagram; (3) according to the three-dimensional lesion radiation dose distribution diagram, selecting proper radiation sources and radiation doses to produce stents, and performing three-dimensional precise intraluminal radiation therapy. The invention further provides a method for implementing the intraluminal radiation therapy method. Lesion tissue in the lesion lumens is subjected to three-dimensional reconstruction by means of modern medicine three-dimensional image reconstruction technology, most reasonable radiation therapy particles are mounted at proper positions of the stents, three-dimensional precise intraluminal radiation therapy is realized, cancers are treated with the optimized radiation dose, and normal tissue is protection to the largest extent.

The invention relates to an intelligent gastroscope image processing device, including a gastroscope device housing; a probe is arranged on one side of the gastroscope device housing; a 3D image display and a split screen touch display are arranged on the front side of the gastroscope device housing; the connected end of the split screen touch display is provided with an image editing unit; a PLCis arranged inside the housing; the PLC is internally provided with a coding classification unit; the input end of the PLC is provided with an image processing system and a data processing system; theinput end of the image processing system is provided with an image pickup module; the image pickup module is internally provided with an A coding unit; and the connection end of the image processingsystem is provided with a first database. The invention can perform color contrast and lesion feature contrast on the photographed picture through the setting of the image processing system, and the setting of a picture search engine can search lesion pictures uploaded on the network, conveniently assisting a doctor to quickly judge the condition, shortening the time for a patient to do the gastroscope and reducing the pain.

The invention discloses an auxiliary diagnosis system based on a Mask R-CNN network and an auxiliary diagnosis information generation method, and belongs to the technical field of medical image processing and segmentation, and the system comprises: a data uploading end which is used for uploading a CT image of a patient and corresponding medical auxiliary information, and the CT image carries the cerebral hemorrhage information of the patient; the image processor that is connected with the data uploading end and is used for enhancing and graying the CT image to obtain a grayscale image; the target detection module that is connected with the image processor, and is used for detecting a grayscale image by using a trained Mask R-CNN network model so as to identify and extract lesion feature information; and the diagnosis analysis module that is connected with the target detection module, and is used for matching the focus characteristic information with the structured medical record in the medical record database, and synthesizing medical auxiliary information to generate auxiliary diagnosis information. According to the invention, the trained Mask R-CNN network model is utilized to scan the brain CT image to generate the auxiliary diagnosis information, so that the efficiency of cerebral hemorrhage screening in emergency treatment is improved.

The invention discloses a judgment method for simultaneously identifying stages and lesion features of diabetic retinopathy. The judgment method comprises the following steps: acquiring a fundus image; and inputting the fundus image into a deep learning-based multi-task network to simultaneously identify a stage to which the fundus image belongs and lesion features of the fundus image, the multi-task network at least comprising a backbone network, a feature head network and a stage head network, the backbone network receiving the fundus image and outputting a first feature map, the feature head network receiving the first feature map, judging lesion features of the fundus image and obtaining a second feature map through a middle layer of the feature head network, and the staging head network receiving the first feature map and obtaining a third feature map based on a middle layer of the staging head network; and performing feature fusion on the third feature map and the second feature map to judge the stage to which the eye fundus image belongs. Therefore, the causal relationship between the lesion features and the stages can be simulated, and the recognition performance of the stages and the lesion features can be improved at the same time.

The invention provides an OCT image recommendation method and device based on artificial intelligence, electronic equipment and a storage medium, and the method comprises the steps: obtaining fundus OCT images, and carrying out the preprocessing of all the obtained fundus OCT images, and obtaining an OCT to-be-detected image set; inputting images in the OCT to-be-detected image set into a network detection model for focus detection to obtain a focus image set; constructing a lesion vector, and screening the lesion image set based on the lesion vector to obtain an effective lesion image set; calculating a focus characteristic index of each image in the effective focus image set; and evaluating images in the effective lesion image set based on the lesion feature indexes to obtain an OCT recommended image. According to the method, the focus characteristic indexes of the OCT images are calculated, and the vectors are constructed according to the focus characteristic indexes for operation to quickly recommend similar OCT images, so that the reading and diagnosis efficiency of ophthalmologists is improved.