WO2022110445A1 - 一种卵圆孔未闭检测方法、系统、终端以及存储介质 - Google Patents

一种卵圆孔未闭检测方法、系统、终端以及存储介质 Download PDF

Info

Publication number
WO2022110445A1
WO2022110445A1 PCT/CN2020/139680 CN2020139680W WO2022110445A1 WO 2022110445 A1 WO2022110445 A1 WO 2022110445A1 CN 2020139680 W CN2020139680 W CN 2020139680W WO 2022110445 A1 WO2022110445 A1 WO 2022110445A1
Authority
WO
WIPO (PCT)
Prior art keywords
ctte
tte
image data
image
image sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/139680
Other languages
English (en)
French (fr)
Inventor
张世全
肖杨
杜丽娟
马腾
郑海荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Publication of WO2022110445A1 publication Critical patent/WO2022110445A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10132—Ultrasound image
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30048—Heart; Cardiac

Definitions

  • the present application belongs to the technical field of medical ultrasound imaging, and in particular relates to a patent foramen ovale detection method, system, terminal and storage medium.
  • the foramen ovale is an embryonic defect (small hole) in the atrial septum in the heart of the heart. During fetal growth and development, the foramen ovale serves as an important vital passageway, allowing oxygenated blood to enter the left atrium from the right atrium. In 75% of newborns, as the lungs develop and pulmonary resistance decreases, the consequent increase in pressure within the heart from the left to the right atrium causes the opening to close automatically and the foramen ovale to close functionally . However, patent foramen ovale is still present in 15%-35% of the adult population (Patent foramen ovale, PFO). According to statistics, PFO is the most common congenital heart abnormality in adults. Correct judgment of whether the foramen ovale is closed is an extremely important objective index for clinical diagnosis, identification and evaluation of curative effects.
  • the evaluation methods for whether the foramen ovale is closed mainly include:
  • cTCD contrast transcranial Doppler, enhanced transcranial Doppler ultrasound
  • TTE TransThoracic Echocardiography, transthoracic echocardiography
  • cTTE contrast Transthoracic echocardiography, contrast-enhanced transthoracic echocardiography
  • clinical gold standard TEE transesophageal echocardiography, transesophageal echocardiography
  • the ultrasound probe was inserted into the patient's esophagus to observe the blood flow changes at the foramen ovale of the heart, and whether the foramen ovale was closed was judged by whether there was blood flow shunting; during the cTTE examination, when The contrast agent was injected into the patient's body while the patient performed tile breathing, and it was observed whether the foramen ovale reached the left atrium through the foramen ovale within 3-5 cardiac cycles after filling of the right atrium to determine whether the foramen ovale was closed.
  • TEE is a semi-invasive testing method that cannot avoid risks. It is difficult to operate and requires high experience of clinicians. In China, this testing method is not popular, and it is often only carried out in large hospitals; cTCD has passed the test.
  • the detection of contrast agent microbubbles from PFO in the cerebral blood circulation can only indirectly infer the source of the microbubbles, but cannot directly prove the existence of PFO; although TTE is non-invasive and simple to operate, its detection rate and sensitivity are low; cTTE is a kind of The new detection method is non-invasive and can provide more detailed information on the right-to-left shunt from the heart, but it is susceptible to noise, motion artifacts, and lung gas, and has poor sensitivity and high specificity.
  • the present application provides a patent foramen ovale detection method, system, terminal and storage medium, aiming to solve one of the above technical problems in the prior art at least to a certain extent.
  • a patent foramen ovale detection method comprising:
  • the left atrium ROI is segmented for each frame of the dynamic image data from the apical four-chamber view perspective by using a convolutional neural network; wherein, the dynamic image data from the apical four-chamber view perspective includes a set number of cardiac cycles TTE image data and cTTE image data;
  • Microbubble detection and feature extraction are performed on the screened cTTE images, and the extracted features are input into the trained prediction model, and the foramen ovale closed state detection result is output through the prediction model.
  • the technical solutions adopted in the embodiments of the present application further include: before the use of the convolutional neural network to perform left atrial ROI segmentation on each frame of the dynamic image data from the apical four-chamber view perspective, the following further includes:
  • the MD5 algorithm is used to screen the TTE image sequence and the cTTE image sequence, and the repeated frames in the TTE image sequence and the cTTE image sequence are removed;
  • the filtered TTE image sequence and cTTE image sequence are filtered and denoised by GAD filtering algorithm.
  • the technical solutions adopted in the embodiments of the present application further include: the use of the convolutional neural network to perform left atrial ROI segmentation on each frame of the dynamic image data from the apical four-chamber view perspective is specifically:
  • a 2D U-net convolutional neural network is used to establish a TTE segmentation model and a cTTE segmentation model, respectively, and the TTE image sequence and cTTE image sequence are input into the TTE segmentation model and the cTTE segmentation model, respectively.
  • Continuous downsampling and pooling operations obtain the feature map of the input image, perform continuous upsampling on the feature map to restore the feature map to the size of the input, and output the TTE image sequence and the cTTE image sequence respectively.
  • the segmentation result of the left atrium ROI corresponding to each frame of image in .
  • the technical solution adopted in the embodiment of the present application further includes: the cTTE image of which the corresponding interval is selected according to the minimum grayscale distribution interval of the TTE image data is specifically:
  • the TTE image sequence and cTTE image sequence of a set number of cardiac cycles were screened out respectively, and bilinear The difference makes the frame numbers of the screened TTE image sequence and the cTTE image sequence consistent;
  • For the filtered TTE image sequence calculate the gray level in the ROI of each frame of TTE image frame by frame and calculate the mean value, generate the gray level mean value distribution map of the TTE image sequence, and calculate the gray level mean value according to the set interval range from the gray level mean value. Screen out the minimum gray distribution interval whose gray mean value is less than the set interval ratio in the distribution map;
  • a cTTE image sequence corresponding to the gray distribution interval is selected from the cTTE image sequence of the set number of cardiac cycles.
  • microbubble detection on the screened cTTE image is specifically:
  • the left atrium ROI in which microbubbles are detected first, based on the grayscale features and circularity morphological features of microbubbles, the feature quantities that characterize each superpixel block are extracted and the saliency map is calculated, and the grayscale threshold constraint is set to obtain the left atrial ROI.
  • the technical solutions adopted in the embodiments of the present application further include: the feature extraction of the screened cTTE images is specifically:
  • the shape feature, intensity feature and texture feature of the left atrium ROI of each frame of cTTE image in the cTTE image sequence with microbubbles were extracted by Pyradiomics algorithm, and the extracted features were screened by variance selection, t test and LASSO regression algorithm. and dimensionality reduction.
  • the prediction model is an SVM model.
  • Region of interest segmentation module used for using a convolutional neural network to perform left atrial ROI segmentation on each frame of the dynamic image data from the apical four-chamber view; wherein the dynamic image data from the apical four-chamber view includes: TTE image data and cTTE image data within a set number of cardiac cycles;
  • Artifact removal module used to calculate the grayscale of the left atrium ROI of each frame of image in the TTE image data, and filter out the minimum grayscale distribution interval of the TTE image data according to the set ratio, according to the TTE image data
  • the minimum grayscale distribution interval of the data selects the cTTE image of the corresponding interval from the cTTE image data;
  • Microbubble detection module used to perform microbubble detection on the screened cTTE images to obtain a sequence of microbubble images containing microbubbles;
  • Feature extraction module used for feature extraction on the microbubble image sequence
  • Prediction and classification module used to input the features extracted from the microbubble image sequence into the trained prediction model, and output the detection result of the closed state of the foramen ovale through the prediction model.
  • a terminal includes a processor and a memory coupled to the processor, wherein,
  • the memory stores program instructions for realizing the patent foramen ovale detection method
  • the processor is configured to execute the program instructions stored in the memory to control patent foramen ovale detection.
  • a storage medium storing program instructions executable by a processor, where the program instructions are used to execute the patent foramen ovale detection method.
  • the beneficial effects of the embodiments of the present application are: the patent foramen ovale detection method, system, terminal and storage medium of the embodiments of the present application collect TTE image data and cTTE image data, and after ROI segmentation is performed on the TTE image data and the cTTE image data respectively, the cTTE image sequence is screened according to the minimum grayscale distribution interval of the TTE image data, and the cTTE image sequence after motion artifact removal is obtained.
  • the cTTE image sequence is used for microbubble detection and feature extraction, the extracted features are input into the trained prediction model, and the detection result of the closed foramen ovale of the subject is output through the prediction model.
  • the embodiments of the present application realize non-invasive PFO detection, with simple operation and high accuracy, and improve the detection rate and detection efficiency of PFO detection.
  • Fig. 1 is the flow chart of the patent foramen ovale detection method of the embodiment of the present application
  • FIG. 2 is a schematic diagram of a prediction model training process according to an embodiment of the present application.
  • FIG. 3 is a schematic structural diagram of a patent foramen ovale detection system according to an embodiment of the application.
  • FIG. 4 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • the patent foramen ovale detection method of the embodiment of the present application collects TTE image data and cTTE image data of the apical four-chamber view of the subject's heart respectively, and analyzes the TTE image data and cTTE image data respectively.
  • the cTTE image sequence was screened according to the minimum grayscale distribution interval of the left atrial region of interest in the TTE image data, and the cTTE image sequence after motion artifact removal was obtained.
  • the screened cTTE image sequence is subjected to microbubble detection and feature extraction, the extracted features are input into the trained prediction model, and the detection result of the closed foramen ovale of the subject is output through the prediction model.
  • FIG. 1 is a flowchart of a method for detecting a patent foramen ovale according to an embodiment of the present application.
  • the patent foramen ovale detection method of the embodiment of the present application comprises the following steps:
  • the image data includes TTE image data and cTTE image data;
  • the image data collection method is specifically: perform TTE and cTTE detection on the subject, and collect the TTE and cTTE of the subject respectively according to the unified collection specification and data specification Cardiac video image data of the first five cardiac cycles from the apical four-chamber view of the 2D ultrasound modality.
  • the cardiac cycle refers to from the start of one heartbeat to the start of the next heartbeat, and the number of cardiac cycles for collecting impact data can also be set according to actual operations.
  • the image data preprocessing specifically includes:
  • S111 utilize OpenCV to convert TTE image data and cTTE image data into frame-by-frame TTE image sequence and cTTE image sequence respectively;
  • S112 Screen the converted TTE image sequence and cTTE image sequence by using the MD5 algorithm, and remove duplicate frames in the TTE image sequence and the cTTE image sequence;
  • the embodiment of the present invention adopts the MD5 message digest algorithm (Message-Digest Algorithm). 5, MD5) Convert each frame of image in the TTE image sequence and the cTTE image sequence into a fixed-length MD5 value. If the MD5 values of several frames of images are exactly the same, it means that these frames are repeated Several frames of images are deduplicated, so that only one frame of image is retained for each MD5 value.
  • MD5 message digest algorithm Message-Digest Algorithm. 5
  • MD5 Convert each frame of image in the TTE image sequence and the cTTE image sequence into a fixed-length MD5 value. If the MD5 values of several frames of images are exactly the same, it means that these frames are repeated Several frames of images are deduplicated, so that only one frame of image is retained for each MD5 value.
  • GAD Gabor-based Anisotropic Diffusion, Gabor anisotropic diffusion
  • the embodiment of the present invention performs filtering and denoising processing on the images, so as to reduce the interference of noise, enhance the image edges and improve the contrast, and further improve the image quality.
  • S120 segment the left atrium region of interest from the preprocessed image data by using a convolutional neural network
  • the segmentation method of the region of interest in the left atrium is as follows: under the framework of deep learning Pytorch, a 2D U-net convolutional neural network is used to establish a TTE segmentation model and a cTTE segmentation model respectively. Input the TTE image sequence and cTTE image sequence into the TTE segmentation model and the cTTE segmentation model respectively, obtain the feature map of the input image through continuous downsampling and pooling operations, and then perform continuous upsampling on the feature map to restore the feature map to the input. The size of the left atrial region of interest corresponding to each frame in the TTE image sequence and the cTTE image sequence is output separately.
  • the manual annotation is used as a label for model training
  • the region-sensitive Dice index is used to measure the coincidence of the regions
  • the boundary-sensitive Hausdorff distance (HD) and mean absolute distance (mean absolute distance, The MAD) indicator measures the distance of the boundary, evaluates the segmentation model and adjusts the parameters, and realizes the automatic segmentation of the region of interest in the left atrium.
  • S130 Calculate the gray level of the region of interest of the left atrium of each frame of TTE images in the TTE image sequence respectively and calculate the mean value, obtain the gray level distribution map of the region of interest of the left atrium of all TTE images, and calculate the gray level of the region of interest from the left atrium according to the set ratio.
  • the minimum grayscale distribution interval is screened out in the mean distribution graph, and the cTTE image of the corresponding interval is screened according to the minimum grayscale distribution interval, and the cTTE image sequence with the motion artifact removed is obtained;
  • the motion artifact removal method is as follows: first, combining the area change of the region of interest of the left atrium and the cardiac cycle change of the TTE image and the cTTE image, respectively, screen out the TTE image sequence and cTTE image sequence of three complete cardiac cycles, respectively. And use the bilinear difference to make the frame numbers of all the screened image sequences consistent.
  • the interval range selects the minimum gray distribution interval whose gray mean value is less than the set interval ratio from the gray mean distribution map of the TTE image sequence; finally, selects the cTTE image sequence of three complete cardiac cycles according to the minimum gray distribution interval.
  • the cTTE image corresponding to the gray level distribution interval is obtained, and the cTTE image sequence after removing the motion artifact is obtained to reduce the interference of the motion artifact.
  • the interval ratio of the minimum grayscale distribution interval is set to 25%, which can be specifically set according to actual operations.
  • the motion artifact removal method of the embodiment of the present application is also applicable to motion artifact removal of other ultrasound images.
  • the average gray value distribution in the region of interest is calculated by using the image sequence of the conventional ultrasonic detection method to obtain the smallest gray value.
  • the distribution interval, the image sequence of the effective interval in the complex ultrasonic detection means is screened according to the gray distribution interval, etc.
  • S140 respectively perform microbubble detection and segmentation on the region of interest in the left atrium of each frame of cTTE image in the cTTE image sequence after the motion artifact has been removed, to obtain a microbubble image sequence;
  • the microbubble detection and segmentation methods are specifically: using simple linear iterative clustering based on color and spatial similarity constraints (simple linear iterative clustering) cluster, SLIC) algorithm performs superpixel segmentation to detect microbubbles in the region of interest in the left atrium of each frame of cTTE image; The foramen ovale of the examiner is closed.
  • SLIC simple linear iterative clustering
  • For the region of interest in the left atrium where microbubbles are detected first, based on the grayscale features and circularity morphological features of microbubbles, the feature quantities that characterize each superpixel block are extracted and the saliency map is calculated, and the grayscale threshold constraint is set to obtain the left atrial sense. Then, through the circular or elliptic-like roundness shape, set the geometric threshold constraint to obtain the microbubble fine segmentation result of the left atrium region of interest.
  • microbubble detection and segmentation methods of the embodiments of the present application are also applicable to contrast-enhanced ultrasound imaging or other microbubble detection involving the use of a contrast agent to generate an ultrasound image of microbubbles.
  • S150 Use the Pyradiomics algorithm to extract the shape features, intensity features and texture features of the region of interest in the left atrium of each frame of cTTE image in the microbubble image sequence, and use the variance selection, t-test and LASSO regression algorithm to extract the features. Screening and dimensionality reduction processing;
  • the open source software package Pyradiomics is used to extract 9 shape features for describing the geometric characteristics of the ROI, 18 intensity features for describing the first-order distribution of the ROI intensity, and 75 Describe the pattern of ROI intensity or the texture features of high-order distribution; further, in order to eliminate invalid and redundant features, the method of variance selection, t-test and LASSO regression is used to screen and reduce the dimension of a large number of extracted features. Good repeatability features.
  • S160 The features extracted and screened from the microbubble image sequence are input into the trained prediction model for classification, and the classification result of the closed state of the foramen ovale of the detected subject is output through the prediction model.
  • the prediction model is SVM (supported vector machine, support vector machine) model.
  • FIG. 2 is a schematic diagram of a training process of a prediction model according to an embodiment of the present application.
  • the training process of the prediction model in the embodiment of the present application includes the following steps:
  • the image data collection method is as follows: perform TTE and cTTE detection on 129 subjects (85 positive and 44 negative), and collect the data of each subject according to the unified collection specification and data specification.
  • the cardiac cycle refers to from the start of one heartbeat to the start of the next heartbeat, and the number of cardiac cycles for collecting impact data can also be set according to actual operations.
  • the preprocessing methods of TTE image data and cTTE image data specifically include:
  • S211 utilize OpenCV to convert TTE image data and cTTE image data into frame-by-frame TTE image sequence and cTTE image sequence respectively;
  • S212 Screen the converted TTE image sequence and cTTE image sequence by using the MD5 algorithm, and remove duplicate frames in the TTE image sequence and the cTTE image sequence;
  • the embodiment of the present invention adopts the MD5 message digest algorithm (Message-Digest Algorithm). 5, MD5) Convert each frame of image in the TTE image sequence and the cTTE image sequence into a fixed-length MD5 value. If the MD5 values of several frames of images are exactly the same, it means that these frames are repeated Several frames of images are deduplicated, so that only one frame of image is retained for each MD5 value.
  • MD5 message digest algorithm Message-Digest Algorithm. 5
  • MD5 Convert each frame of image in the TTE image sequence and the cTTE image sequence into a fixed-length MD5 value. If the MD5 values of several frames of images are exactly the same, it means that these frames are repeated Several frames of images are deduplicated, so that only one frame of image is retained for each MD5 value.
  • GAD Gabor-based Anisotropic Diffusion, Gabor anisotropic diffusion
  • the embodiment of the present invention performs filtering and denoising processing on the images, so as to reduce the interference of noise, enhance the image edges and improve the contrast, and further improve the image quality.
  • S220 segment the left atrium region of interest of each frame of image from the preprocessed TTE image data and the cTTE image data by using a convolutional neural network;
  • the segmentation method of the region of interest in the left atrium is as follows: under the framework of deep learning Pytorch, a 2D U-net convolutional neural network is used to establish a TTE segmentation model and a cTTE segmentation model respectively. Input the TTE image sequence and cTTE image sequence into the TTE segmentation model and the cTTE segmentation model respectively, obtain the feature map of the input image through continuous downsampling and pooling operations, and then perform continuous upsampling on the feature map to restore the feature map to the input. The size of the left atrial region of interest corresponding to each frame in the TTE image sequence and the cTTE image sequence is output separately.
  • the manual annotation is used as a label for model training
  • the region-sensitive Dice index is used to measure the coincidence of the regions
  • the boundary-sensitive Hausdorff distance (HD) and mean absolute distance (mean absolute distance, The MAD) indicator measures the distance of the boundary, evaluates the segmentation model and adjusts the parameters, and realizes the automatic segmentation of the region of interest in the left atrium.
  • S230 Calculate the gray level of the region of interest in the left atrium of each frame of the TTE image in the TTE image sequence and calculate the mean value, obtain the gray level distribution map of the region of interest in the left atrium of all TTE images, and calculate the gray level of the region of interest from the left atrium according to the set ratio.
  • the minimum grayscale distribution interval is screened out in the mean distribution graph, and the cTTE image of the corresponding interval is screened according to the minimum grayscale distribution interval, and the cTTE image sequence with the motion artifact removed is obtained;
  • the motion artifact removal method is as follows: first, combining the area change of the left atrial region of interest and the cardiac cycle change of the TTE image and the cTTE image, respectively, screen out the TTE of each subject's three complete cardiac cycles. Image sequence and cTTE image sequence, and use bilinear difference to make the frame number of all the screened image sequences consistent.
  • the interval range selects the minimum gray distribution interval whose gray mean value is less than the set interval ratio from the gray mean distribution map of the TTE image sequence; finally, selects the cTTE image sequence of three complete cardiac cycles according to the minimum gray distribution interval.
  • the cTTE image corresponding to the gray distribution interval is obtained, and the cTTE image sequence after removing the motion artifact is obtained.
  • the interval ratio of the minimum grayscale distribution interval is set to 25%, which can be specifically set according to actual operations.
  • S240 Perform microbubble detection and segmentation on the left atrial region of interest of each frame of cTTE image in the cTTE image sequence after motion artifact removal, and divide the cTTE image sequence into a microbubble image sequence and a microbubble-free image sequence according to the detection results image sequence;
  • the microbubble detection and segmentation methods are specifically: using simple linear iterative clustering based on color and spatial similarity constraints (simple linear iterative clustering) cluster, SLIC) algorithm performs superpixel segmentation to detect microbubbles in the region of interest of the left atrium of each frame of cTTE image; if no microbubbles are detected in the region of interest of the left atrium in all cardiac cycles of a subject, Indicates that the subject's foramen ovale is closed.
  • simple linear iterative clustering based on color and spatial similarity constraints (simple linear iterative clustering) cluster, SLIC) algorithm performs superpixel segmentation to detect microbubbles in the region of interest of the left atrium of each frame of cTTE image; if no microbubbles are detected in the region of interest of the left atrium in all cardiac cycles of a subject, Indicates that the subject's foramen ovale is closed.
  • the feature quantities that characterize each superpixel block are extracted, and the saliency map is calculated, and the grayscale threshold constraint is set to obtain the left atrial sense. Then, through the circular or elliptic-like roundness shape, set the geometric threshold constraint to obtain the microbubble fine segmentation result of the left atrium region of interest.
  • S250 Use the Pyradiomics algorithm to extract the shape features, intensity features and texture features of the left atrial region of interest in each frame of the cTTE image in the microbubble image sequence, and use the variance selection, t-test and LASSO regression algorithm to extract the features. Screening and dimensionality reduction processing;
  • the open source software package Pyradiomics is used to extract 9 shape features for describing the geometric characteristics of the ROI, 18 intensity features for describing the first-order distribution of the ROI intensity, and 75 Describe the pattern of ROI intensity or the texture features of high-order distribution; further, in order to eliminate invalid and redundant features, the method of variance selection, t-test and LASSO regression is used to screen and reduce the dimension of a large number of extracted features. Good repeatability features.
  • S260 Input the microbubble-free image sequence and the features extracted and screened from the suspected microbubble image sequence together into a prediction model for training, and output the PFO classification result of each subject through the prediction model;
  • the SVM model is trained by combining cTTE images with no microbubbles detected and cTTE images with detected microbubbles and feature extraction and screening, and at the same time, the leave-one-out method is used to perform cross-validation on the model.
  • the model's accuracy, sensitivity, specificity, and positive predictive rate value, ppv), negative prediction rate (negative The predictive value (npv) and the matthews correlation coefficient (mcc) are used to evaluate the model performance and optimize the parameters to obtain the optimal predictive model.
  • the PFO classification results of the 129 subjects are described in Table 1 below:
  • the patent foramen ovale detection method of the embodiment of the present application combines cTTE images with no microvesicles detected and cTTE images with detected microvesicles and feature extraction and screening to train a prediction model, and collects the apex of the subject's heart by collecting
  • the TTE image data and cTTE image data of the four-chamber cardiac view are divided into ROIs respectively, and the cTTE image sequence is screened according to the minimum grayscale distribution interval of the ROI of the TTE image data to obtain the removal of motion.
  • microbubble detection and feature extraction are performed on the screened cTTE image sequence, the extracted features are input into the trained prediction model, and the detection result of the closed foramen ovale of the subject is output through the prediction model.
  • FIG. 3 is a schematic structural diagram of a patent foramen ovale detection system according to an embodiment of the present application.
  • the patent foramen ovale detection system 40 according to the embodiment of the present application includes:
  • Image data processing module 41 used to preprocess the image data of the subject; wherein, the image data includes TTE image data and cTTE image data;
  • Region of interest segmentation module 42 for segmenting the region of interest of the left atrium corresponding to each frame of image from the preprocessed TTE image data and the cTTE image data by using the convolutional neural network;
  • the artifact removal module 43 is used to separately calculate the gray level of the region of interest in the left atrium of each frame of the TTE image in the TTE image sequence and calculate the mean value, and obtain the gray level distribution map of the region of interest in the left atrial region of all TTE images. Set the ratio to select the minimum grayscale distribution interval from the grayscale mean distribution map, and select the cTTE image of the corresponding interval according to the minimum grayscale distribution interval, and obtain a cTTE image sequence with motion artifacts removed;
  • the microbubble detection module 44 is used to detect and segment the region of interest of the left atrium of each frame of cTTE image in the cTTE image sequence after motion artifact removal, respectively, to obtain a microbubble image sequence with microbubbles;
  • Feature extraction module 45 used to extract shape features, intensity features and texture features of the region of interest in the left atrium of each frame of cTTE image in the microbubble image sequence by using the Pyradiomics algorithm, and use variance selection, t-test and LASSO regression algorithm Screening and dimensionality reduction of the extracted features;
  • Prediction classification module 46 used to input the features extracted and screened from the microvesicle image sequence into the trained prediction model for classification, and output the classification result of the closed foramen ovale of the subject through the prediction model.
  • FIG. 4 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • the terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .
  • the memory 52 stores program instructions for implementing the aforementioned patent foramen ovale detection method.
  • the processor 51 is configured to execute program instructions stored in the memory 52 to control the detection of the patent foramen ovale.
  • the processor 51 may also be referred to as a CPU (Central Processing Unit, central processing unit).
  • the processor 51 may be an integrated circuit chip with signal processing capability.
  • the processor 51 may also be a general purpose processor, digital signal processor (DSP), application specific integrated circuit (ASIC), off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components .
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA off-the-shelf programmable gate array
  • a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • FIG. 5 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • the storage medium of this embodiment of the present application stores a program file 61 capable of implementing all the above methods, wherein the program file 61 may be stored in the above-mentioned storage medium in the form of a software product, and includes several instructions to make a computer device (which can be A personal computer, a server, or a network device, etc.) or a processor (processor) executes all or part of the steps of the methods in the various embodiments of the present invention.
  • a computer device which can be A personal computer, a server, or a network device, etc.
  • processor processor
  • the aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes , or terminal devices such as computers, servers, mobile phones, and tablets.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

本申请涉及一种卵圆孔未闭检测方法、系统、终端以及存储介质。包括：对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割；其中，心尖四腔心切面视角动态影像数据包括设定数量的心动周期内的TTE影像数据和cTTE影像数据；计算TTE影像数据中每一帧图像的左心房ROI的灰度，并筛选出TTE影像数据的最小灰度分布区间，根据TTE影像数据的最小灰度分布区间筛选出cTTE影像数据中对应区间的cTTE图像；对筛选的cTTE图像进行微泡检测和特征提取，并输入训练好的预测模型，通过预测模型输出卵圆孔闭合状态检测结果。本申请实现了无创的PFO检测，操作简单且准确率高，提高了检出率和检测效率。

Description

一种卵圆孔未闭检测方法、系统、终端以及存储介质 技术领域
本申请属于医学超声成像技术领域，特别涉及一种卵圆孔未闭检测方法、系统、终端以及存储介质。
 背景技术
卵圆孔是心脏中心房房间隔的一种胚胎缺陷(小孔)。在胎儿生长发育过程中，卵圆孔作为重要的生命通道，它允许含氧血液从右心房进入左心房。在新生儿中有75%的比例，随着肺的发育和肺部阻力的下降，随之而来增加了心脏内从左到右心房的压力，导致开口自动关闭，卵圆孔发生功能性闭合。然而，在15%-35%的成年人群中仍然存在卵圆孔未闭(Patent foramen ovale, PFO)。据统计，PFO是目前成年人中最为常见的先天性心脏异常。卵圆孔是否闭合的正确判断对于临床诊断、鉴别和疗效评估等都是极其重要的客观指标。
目前，卵圆孔是否闭合的评估方法主要包括：
①临床医学影像诊断；临床医生借助医学影像手段，通过cTCD（contrast transcranial Doppler，增强经颅多谱勒超声）、TTE（TransThoracic Echocardiography，经胸超声心动图）、cTTE（contrast transthoracic echocardiography，经胸超声心动图造影）和临床金标准TEE（transesophageal echocardiography, 经食管超声）进行检测评估。其中，在进行TEE检查时，将超声探头插入患者的食道中观察心脏卵圆孔处的血流变化，通过是否存在血流分流的现象来判断卵圆孔是否闭合；在进行cTTE检查时，当造影剂被注射到患者体内同时伴随着患者进行瓦式呼吸，观察在右心房充盈后3-5个心动周期内，是否有微泡通过卵圆孔到达左心房来判断卵圆孔是否闭合。
②临床探究方法；Nakayama等人于2019年在Journal of the American Society of Echocardiography上发表题为《Identification of High-Risk Patent Foramen Ovale Associated With Cryptogenic Stroke: Development of a Scoring System》的文章，提出了一套评估卵圆孔未闭合的风险引发隐源性脑卒中的评估系统；Ravi等人于2020年在European Heart Journal - Cardiovascular Imaging上发表题为《Improved differential diagnosis of intracardiac and extracardiac shunts using acoustic intensity mapping of saline contrast studies》的文章，提出通过声学造影超声心动图筛查从右心房到左心房分流的病例，绘制左右心脏内声音时间强度曲线的变化来识别和定义分流的起源，对于鉴别右向左分流、区分卵圆孔未闭合。
综上所述，现有的卵圆孔闭合评估方法存在以下不足：
一、TEE为不能免除风险的半侵入性检测手段，其操作难度大，对临床医生经验要求较高，在国内，该检测手段并不普及，往往只有大型医院才会开展此项检测；cTCD通过脑血循环探测来自PFO的造影剂微泡，其只能间接推测微泡的来源，不能直接证明PFO的存在；TTE虽然无创且操作简单，但检出率以及敏感性均较低；cTTE作为一种新型的检测手段，无创且能更为详细地提供从心脏内右向左分流的信息，然而易受噪声、运动伪影和肺气等的影响，敏感性差，且特异性高。
二、现有的临床探究方法涉及实验群体有限且结论通用性不强，其效果有待进一步证实。
三、目前没有采用人工智能的方法对卵圆孔是否闭合进行有效评估。
 技术问题
本申请提供了一种卵圆孔未闭检测方法、系统、终端以及存储介质，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
 技术解决方案
一种卵圆孔未闭检测方法，包括：
利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割；其中，所述心尖四腔心切面视角动态影像数据包括设定数量的心动周期内的TTE影像数据和cTTE影像数据；
计算所述TTE影像数据中每一帧图像的左心房ROI的灰度，并按照设定比例筛选出所述TTE影像数据的最小灰度分布区间，根据所述TTE影像数据的最小灰度分布区间从所述cTTE影像数据中筛选出对应区间的cTTE图像；
对所述筛选的cTTE图像进行微泡检测和特征提取，并将提取的特征输入训练好的预测模型，通过所述预测模型输出卵圆孔闭合状态检测结果。
本申请实施例采取的技术方案还包括：所述利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割前还包括：
分别将所述TTE影像数据和cTTE影像数据转化成逐帧的TTE图像序列和cTTE图像序列；
采用MD5算法对所述TTE图像序列和cTTE图像序列进行筛选，去除所述TTE图像序列和cTTE图像序列中的重复帧；
采用GAD滤波算法对筛选后的TTE图像序列和cTTE图像序列进行滤波祛噪处理。
本申请实施例采取的技术方案还包括：所述利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割具体为：
利用2D U-net卷积神经网络分别建立TTE分割模型和cTTE分割模型，将所述TTE图像序列和cTTE图像序列分别输入TTE分割模型和cTTE分割模型，所述TTE分割模型和cTTE分割模型分别经过连续的下采样、池化操作获得输入图像的特征图，对所述特征图进行连续的上采样，使所述特征图恢复到输入时的尺寸，并分别输出所述TTE图像序列和cTTE图像序列中每一帧图像对应的左心房ROI分割结果。
本申请实施例采取的技术方案还包括：所述根据所述TTE影像数据的最小灰度分布区间筛选出对应区间的cTTE图像具体为：
结合所述TTE图像序列和cTTE图像序列中每一帧图像的左心房ROI的面积变化和心动周期变化，分别筛选出设定数量的心动周期的TTE图像序列和cTTE图像序列，并利用双线性差值使所述筛选后的TTE图像序列和cTTE图像序列的帧数一致；
针对筛选后的TTE图像序列，逐帧计算每一帧TTE图像的ROI内的灰度并求均值，生成TTE图像序列的灰度均值分布图，并按照设定的区间范围从所述灰度均值分布图中筛选出灰度均值小于设定区间比例的最小灰度分布区间；
根据所述最小灰度分布区间从所述设定数量的心动周期的cTTE图像序列中筛选出与该灰度分布区间相对应的cTTE图像序列。
本申请实施例采取的技术方案还包括：所述对所述筛选的cTTE图像进行微泡检测具体为：
利用基于色彩和空间相似度约束的简单线性迭代聚类算法对所述筛选出的cTTE图像序列中每一帧cTTE图像的左心房ROI分别进行超像素分割检测微泡；
对于检测到微泡的左心房ROI，首先基于微泡的灰度特征和圆度形态特征提取表征各超像素块的特征量并计算显著图，设定灰度阈值约束获得所述左心房ROI的微泡粗分割结果；然后，通过类圆或类椭圆性的圆度形态，设定几何阈值约束获得所述左心房ROI的微泡细分割结果。
本申请实施例采取的技术方案还包括：所述对所述筛选的cTTE图像进行特征提取具体为：
利用Pyradiomics算法对存在微泡的cTTE图像序列中每一帧cTTE图像的左心房ROI分别进行形状特征、强度特征和纹理特征提取，并利用方差选择、t检验和LASSO回归算法对提取的特征进行筛选和降维处理。
本申请实施例采取的技术方案还包括：所述预测模型为SVM模型。
本申请实施例采取的另一技术方案为：一种卵圆孔未闭检测系统，包括：
感兴趣区域分割模块：用于利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割；其中，所述心尖四腔心切面视角动态影像数据包括设定数量的心动周期内的TTE影像数据和cTTE影像数据；
伪影祛除模块：用于计算所述TTE影像数据中每一帧图像的左心房ROI的灰度，并按照设定比例筛选出所述TTE影像数据的最小灰度分布区间，根据所述TTE影像数据的最小灰度分布区间从所述cTTE影像数据中筛选出对应区间的cTTE图像；
微泡检测模块：用于对所述筛选的cTTE图像进行微泡检测，得到存在微泡的微泡图像序列；
特征提取模块：用于对所述微泡图像序列进行特征提取；
预测分类模块：用于将所述从微泡图像序列中提取的特征输入训练好的预测模型，通过所述预测模型输出卵圆孔闭合状态检测结果。
本申请实施例采取的又一技术方案为：一种终端，所述终端包括处理器、与所述处理器耦接的存储器，其中，
所述存储器存储有用于实现所述卵圆孔未闭检测方法的程序指令；
所述处理器用于执行所述存储器存储的所述程序指令以控制卵圆孔未闭检测。
本申请实施例采取的又一技术方案为：一种存储介质，存储有处理器可运行的程序指令，所述程序指令用于执行所述卵圆孔未闭检测方法。 有益效果
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的卵圆孔未闭检测方法、系统、终端及存储介质通过采集被检测者心尖四腔心切面的TTE影像数据和cTTE影像数据，并对TTE影像数据和cTTE影像数据分别进行ROI分割后，根据TTE影像数据的最小灰度分布区间对cTTE图像序列进行筛选，得到祛除运动伪影后的cTTE图像序列，对筛选后的cTTE图像序列进行微泡检测和特征提取，将提取的特征输入训练好的预测模型，通过预测模型输出被检测者的卵圆孔闭合状态检测结果。基于上述方案，本申请实施例实现了无创的PFO检测，操作简单且准确率高，提高了PFO检测的检出率和检测效率。
 附图说明
图1是本申请实施例的卵圆孔未闭检测方法的流程图；
图2是本申请实施例的预测模型训练过程示意图；
图3为本申请实施例的卵圆孔未闭检测系统结构示意图；
图4为本申请实施例的终端结构示意图；
图5为本申请实施例的存储介质的结构示意图。
 本发明的实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
针对现有技术的不足，本申请实施例的卵圆孔未闭检测方法通过分别采集被检测者心尖四腔心切面的TTE影像数据和cTTE影像数据，并对TTE影像数据和cTTE影像数据分别进行左心房感兴趣区域(region of interest, ROI)分割后，根据TTE影像数据的左心房感兴趣区域的最小灰度分布区间对cTTE图像序列进行筛选，得到祛除运动伪影后的cTTE图像序列，对筛选后的cTTE图像序列进行微泡检测和特征提取，将提取的特征输入训练好的预测模型，通过预测模型输出被检测者的卵圆孔闭合状态检测结果。
具体的，请参阅图1，是本申请实施例的卵圆孔未闭检测方法的流程图。本申请实施例的卵圆孔未闭检测方法包括以下步骤：
S100：采集被检测者的心尖四腔心切面视角动态影像数据；
本步骤中，影像数据包括TTE影像数据和cTTE影像数据；影像数据采集方式具体为：对被检测者进行TTE和cTTE检测，按照统一的采集规范和数据规范，分别采集被检测者的TTE和cTTE的2D超声模态心尖四腔心切面视角的前五个心动周期的心脏视频影像数据。其中，心动周期（cardiac cycle）指从一次心跳的起始到下一次心跳的起始，采集影响数据的心动周期数量也可以根据实际操作进行设定。
S110：对采集的影像数据进行预处理；
本步骤中，影像数据预处理具体包括：
S111:利用OpenCV分别将TTE影像数据和cTTE影像数据转化成逐帧的TTE图像序列和cTTE图像序列；
S112：采用MD5算法对转换后的TTE图像序列和cTTE图像序列进行筛选，去除TTE图像序列和cTTE图像序列中的重复帧；
其中，由于心脏跳动和成像设备的采样频率过高，导致成像时产生重复帧。为避免重复帧干扰后续训练深度学习分割模型和分类任务，本发明实施例采用MD5信息摘要算法(Message-Digest Algorithm 5, MD5)将TTE图像序列和cTTE图像序列中的每一帧图像分别转换为一个长度固定的MD5值，若几帧图像的MD5值完全相同，表示这几帧图像为重复帧，则对这几帧图像进行去重处理，使每个MD5值仅保留一帧图像。
S113: 采用GAD(Gabor-based Anisotropic Diffusion,伽柏各向异性扩散）滤波算法对去重处理后的TTE图像序列和cTTE图像序列进行滤波祛噪处理；
其中，由于超声图像中往往存在斑点噪声等干扰，为了提高图像的质量和信噪比，本发明实施例对图像进行滤波祛噪处理，以减少噪声的干扰，增强图像边缘和提高对比度，进而提高图像的质量。
S120：利用卷积神经网络从预处理后的影像数据中分割出左心房感兴趣区域；
本步骤中，左心房感兴趣区域分割方式具体为：在深度学习Pytorch框架下，利用2D U-net卷积神经网络分别建立TTE分割模型和cTTE分割模型。将TTE图像序列和cTTE图像序列分别输入TTE分割模型和cTTE分割模型，经过连续的下采样、池化操作获得输入图像的特征图，接着对特征图进行连续的上采样，使特征图恢复到输入时的尺寸，并分别输出TTE图像序列和cTTE图像序列中每一帧图像对应的左心房感兴趣区域分割结果。在训练TTE分割模型和cTTE分割模型时，通过人工标注作为标签进行模型训练，利用对区域敏感的Dice指标衡量区域的重合度，并利用对边界敏感的豪斯多夫（hausdorff distance, HD）和平均绝对距离（mean absolute distance, MAD）指标衡量边界的距离，对分割模型进行评估和参数调整，实现左心房感兴趣区域的自动分割。
S130：分别计算TTE图像序列中每一帧TTE图像的左心房感兴趣区域的灰度并求均值，获得所有TTE图像的左心房感兴趣区域灰度均值分布图，并根据设定比例从该灰度均值分布图中筛选出最小灰度分布区间，根据最小灰度分布区间筛选出对应区间的cTTE图像，得到祛除运动伪影的cTTE图像序列；
本步骤中，由于受呼吸、肌肉运动、心脏运动以及设备噪声等因素的影响，在成像过程中会造成图像上出现运动伪影，影响后续的图像分析。因此，需要对图像进行运动伪影祛除。具体的，运动伪影祛除方式为：首先，分别结合TTE图像和cTTE图像的左心房感兴趣区域的面积变化和心动周期变化，分别筛选出三个完整心动周期的TTE图像序列和cTTE图像序列，并利用双线性差值使筛选出的所有图像序列的帧数一致。其次，针对筛选后的所有TTE图像序列，逐帧计算每一帧TTE图像的左心房感兴趣区域内的灰度并求均值，生成所有TTE图像序列的灰度均值分布图，并按照设定的区间范围从TTE图像序列的灰度均值分布图中筛选出灰度均值小于设定区间比例的最小灰度分布区间；最后，根据最小灰度分布区间从三个完整心动周期的cTTE图像序列中筛选出与该灰度分布区间相对应的cTTE图像，得到祛除运动伪影后的cTTE图像序列，以减少运动伪影的干扰。具体的，本申请实施例设定最小灰度分布区间的区间比例为25%，具体可根据实际操作进行设定。
可以理解，本申请实施例的运动伪影祛除方法同样适用于其他超声图像的运动伪影祛除，例如利用常规超声检测手段的图像序列计算感兴趣区域内的灰度均值分布，获得最小的灰度分布区间，根据灰度分布区间筛选复杂超声检测手段中的有效区间的图像序列等。
S140：对祛除运动伪影后的cTTE图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行微泡检测及分割，得到微泡图像序列；
本步骤中，微泡检测及分割方式具体为：利用基于色彩和空间相似度约束的简单线性迭代聚类(simple linear iterative cluster，SLIC)算法对每一帧cTTE图像的左心房感兴趣区域分别进行超像素分割检测微泡；如果被检测者所有心动周期内的左心房感兴趣区域均未检测到微泡，表示该被检测者的卵圆孔为闭合状态。对于检测到微泡的左心房感兴趣区域，首先基于微泡的灰度特征和圆度形态特征提取表征各超像素块的特征量并计算显著图，设定灰度阈值约束获得该左心房感兴趣区域的微泡粗分割结果；然后，再通过类圆或类椭圆性的圆度形态，设定几何阈值约束获得该左心房感兴趣区域的微泡细分割结果。
可以理解，本申请实施例的微泡检测及分割方法同样适用于超声造影成像或其他涉及使用造影剂而产生微泡的超声图像的微泡检测。
S150：利用Pyradiomics算法对微泡图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行形状特征、强度特征和纹理特征提取，并利用方差选择、t检验和LASSO回归算法对提取的特征进行筛选和降维处理；
本步骤中，针对微泡图像序列，首先利用开源软件包Pyradiomics分别提取出9个用于描述ROI内几何特性的形状特征、18个用于描述ROI强度一阶分布的强度特征以及75个用于描述ROI强度的模式或高阶分布的纹理特征；进一步地，为了剔除无效冗余的特征，利用方差选择、t检验和LASSO回归的方法对提取出的大量特征进行筛选降维，得到有效和可重复性好的特征。
S160：将从微泡图像序列中提取并筛选的特征输入训练好的预测模型进行分类，通过预测模型输出被检测者的卵圆孔闭合状态分类结果。
本申请实施例中，预测模型为SVM (supported vector machine,支持向量机)模型。进一步地，请参阅图2，是本申请实施例的预测模型训练过程示意图。本申请实施例的预测模型训练过程包括以下步骤：
S200：采集多个受试者的心尖四腔心切面视角的TTE影像数据和cTTE影像数据；
本步骤中，影像数据采集方式具体为：对129例受试者（85例阳性，44例阴性）进行TTE和cTTE检测，按照统一的采集规范和数据规范，分别采集每一位受试者的TTE和cTTE的2D超声模态心尖四腔心切面视角的前五个心动周期的心脏视频影像数据。其中，心动周期（cardiac cycle）指从一次心跳的起始到下一次心跳的起始，采集影响数据的心动周期数量也可以根据实际操作进行设定。
S210：对采集的TTE影像数据和cTTE影像数据进行预处理；
本步骤中，TTE影像数据和cTTE影像数据的预处理方式具体包括：
S211:利用OpenCV分别将TTE影像数据和cTTE影像数据转化成逐帧的TTE图像序列和cTTE图像序列；
S212：采用MD5算法对转换后的TTE图像序列和cTTE图像序列进行筛选，去除TTE图像序列和cTTE图像序列中的重复帧；
其中，由于心脏跳动和成像设备的采样频率过高，导致成像时产生重复帧。为避免重复帧干扰后续训练深度学习分割模型和分类任务，本发明实施例采用MD5信息摘要算法(Message-Digest Algorithm 5, MD5)将TTE图像序列和cTTE图像序列中的每一帧图像分别转换为一个长度固定的MD5值，若几帧图像的MD5值完全相同，表示这几帧图像为重复帧，则对这几帧图像进行去重处理，使每个MD5值仅保留一帧图像。
S213: 采用GAD (Gabor-based Anisotropic Diffusion,伽柏各向异性扩散）滤波算法对去重处理后的TTE图像序列和cTTE图像序列进行滤波祛噪处理；
其中，由于超声图像中往往存在斑点噪声等干扰，为了提高图像的质量和信噪比，本发明实施例对图像进行滤波祛噪处理，以减少噪声的干扰，增强图像边缘和提高对比度，进而提高图像的质量。
S220：利用卷积神经网络从预处理后的TTE影像数据和cTTE影像数据中分别分割出每一帧图像的左心房感兴趣区域；
本步骤中，左心房感兴趣区域分割方式具体为：在深度学习Pytorch框架下，利用2D U-net卷积神经网络分别建立TTE分割模型和cTTE分割模型。将TTE图像序列和cTTE图像序列分别输入TTE分割模型和cTTE分割模型，经过连续的下采样、池化操作获得输入图像的特征图，接着对特征图进行连续的上采样，使特征图恢复到输入时的尺寸，并分别输出TTE图像序列和cTTE图像序列中每一帧图像对应的左心房感兴趣区域分割结果。在训练TTE分割模型和cTTE分割模型时，通过人工标注作为标签进行模型训练，利用对区域敏感的Dice指标衡量区域的重合度，并利用对边界敏感的豪斯多夫（hausdorff distance, HD）和平均绝对距离（mean absolute distance, MAD）指标衡量边界的距离，对分割模型进行评估和参数调整，实现左心房感兴趣区域的自动分割。
S230：分别计算TTE图像序列中每一帧TTE图像的左心房感兴趣区域的灰度并求均值，获得所有TTE图像的左心房感兴趣区域灰度均值分布图，并根据设定比例从该灰度均值分布图中筛选出最小灰度分布区间，根据最小灰度分布区间筛选出对应区间的cTTE图像，得到祛除运动伪影的cTTE图像序列；
本步骤中，由于受呼吸、肌肉运动、心脏运动以及设备噪声等因素的影响，在成像过程中会造成图像上出现运动伪影，影响后续的图像分析。因此，需要对图像进行运动伪影祛除。具体的，运动伪影祛除方式为：首先，分别结合TTE图像和cTTE图像的左心房感兴趣区域的面积变化和心动周期变化，分别筛选出每一位受试者的三个完整心动周期的TTE图像序列和cTTE图像序列，并利用双线性差值使筛选出的所有图像序列的帧数一致。其次，针对筛选后的所有TTE图像序列，逐帧计算每一帧TTE图像的左心房感兴趣区域内的灰度并求均值，生成所有TTE图像序列的灰度均值分布图，并按照设定的区间范围从TTE图像序列的灰度均值分布图中筛选出灰度均值小于设定区间比例的最小灰度分布区间；最后，根据最小灰度分布区间从三个完整心动周期的cTTE图像序列中筛选出与该灰度分布区间相对应的cTTE图像，得到祛除运动伪影后的cTTE图像序列。具体的，本申请实施例设定最小灰度分布区间的区间比例为25%，具体可根据实际操作进行设定。
S240：对祛除运动伪影后的cTTE图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行微泡检测及分割，并根据检测结果将cTTE图像序列分为微泡图像序列和无微泡图像序列；
本步骤中，微泡检测及分割方式具体为：利用基于色彩和空间相似度约束的简单线性迭代聚类(simple linear iterative cluster，SLIC)算法对每一帧cTTE图像的左心房感兴趣区域分别进行超像素分割检测微泡；如果某一位受试者所有心动周期内的左心房感兴趣区域均未检测到微泡，表示该受试者的卵圆孔为闭合状态。对检测到微泡的左心房感兴趣区域，首先基于微泡的灰度特征和圆度形态特征提取表征各超像素块的特征量并计算显著图，设定灰度阈值约束获得该左心房感兴趣区域的微泡粗分割结果；然后，再通过类圆或类椭圆性的圆度形态，设定几何阈值约束获得该左心房感兴趣区域的微泡细分割结果。
S250：利用Pyradiomics算法对微泡图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行形状特征、强度特征和纹理特征提取，并利用方差选择、t检验和LASSO回归算法对提取的特征进行筛选和降维处理；
本步骤中，针对微泡图像序列，首先利用开源软件包Pyradiomics分别提取出9个用于描述ROI内几何特性的形状特征、18个用于描述ROI强度一阶分布的强度特征以及75个用于描述ROI强度的模式或高阶分布的纹理特征；进一步地，为了剔除无效冗余的特征，利用方差选择、t检验和LASSO回归的方法对提取出的大量特征进行筛选降维，得到有效和可重复性好的特征。
S260：将无微泡图像序列和从疑似微泡图像序列中提取并筛选的特征一起输入预测模型进行训练，通过预测模型输出每一位受试者的PFO分类结果；
本步骤中，本申请实施例通过联合未检测出微泡的cTTE图像和检测出微泡并进行特征提取和筛选的cTTE图像进行SVM模型的训练，同时采用留一法对模型进行交叉验证，根据模型的准确率accuracy、敏感性sensitivity、特异性specificity、阳性预测率(positive predictive value, ppv)、阴性预测率(negative predictive value, npv)和马修相关系数 (matthews correlation coefficient, mcc)，对模型性能进行评估和参数优化，进而获得最优预测模型。129例受试者的PFO分类结果如下表1所述：
表1129例受试者的PFO分类结果
评价指标 accuracy sensitivity specificity ppv npv mcc
结果 0.7755 0.8333 0.6154 0.8571 0.5714 0.4385
基于上述，本申请实施例的卵圆孔未闭检测方法联合未检测出微泡的cTTE图像和检测出微泡并进行特征提取和筛选的cTTE图像进行预测模型的训练，通过采集被检测者心尖四腔心切面的TTE影像数据和cTTE影像数据，并对TTE影像数据和cTTE影像数据分别进行ROI分割后，根据TTE影像数据的ROI的最小灰度分布区间对cTTE图像序列进行筛选，得到祛除运动伪影后的cTTE图像序列，对筛选后的cTTE图像序列进行微泡检测和特征提取，将提取的特征输入训练好的预测模型，通过预测模型输出被检测者的卵圆孔闭合状态检测结果。基于上述方案，本申请实施例实现了无创的PFO检测，操作简单且准确率高，提高了PFO检测的检出率和检测效率。
请参阅图3，为本申请实施例的卵圆孔未闭检测系统的结构示意图。本申请实施例的卵圆孔未闭检测系统40包括：
影像数据处理模块41：用于对被检测者的影像数据进行预处理；其中，影像数据包括TTE影像数据和cTTE影像数据；
感兴趣区域分割模块42：用于利用卷积神经网络从预处理后的TTE影像数据和cTTE影像数据中分别分割出每一帧图像对应的左心房感兴趣区域；
伪影祛除模块43：用于分别计算TTE图像序列中每一帧TTE图像的左心房感兴趣区域的灰度并求均值，获得所有TTE图像的左心房感兴趣区域灰度均值分布图，并根据设定比例从该灰度均值分布图中筛选出最小灰度分布区间，根据最小灰度分布区间筛选出对应区间的cTTE图像，得到祛除运动伪影的cTTE图像序列；
微泡检测模块44：用于对祛除运动伪影后的cTTE图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行微泡检测及分割，得到存在微泡的微泡图像序列；
特征提取模块45：用于利用Pyradiomics算法对微泡图像序列中每一帧cTTE图像的左心房感兴趣区域分别进行形状特征、强度特征和纹理特征提取，并利用方差选择、t检验和LASSO回归算法对提取的特征进行筛选和降维处理；
预测分类模块46：用于将从微泡图像序列中提取并筛选的特征输入训练好的预测模型进行分类，通过预测模型输出被检测者的卵圆孔闭合状态分类结果。
请参阅图4，为本申请实施例的终端结构示意图。该终端50包括处理器51、与处理器51耦接的存储器52。
存储器52存储有用于实现上述卵圆孔未闭检测方法的程序指令。
处理器51用于执行存储器52存储的程序指令以控制卵圆孔未闭检测。
其中，处理器51还可以称为CPU（Central Processing Unit，中央处理单元）。处理器51可能是一种集成电路芯片，具有信号的处理能力。处理器51还可以是通用处理器、数字信号处理器（DSP）、专用集成电路（ASIC）、现成可编程门阵列（FPGA）或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
请参阅图5，为本申请实施例的存储介质的结构示意图。本申请实施例的存储介质存储有能够实现上述所有方法的程序文件61，其中，该程序文件61可以以软件产品的形式存储在上述存储介质中，包括若干指令用以使得一台计算机设备（可以是个人计算机，服务器，或者网络设备等）或处理器（processor）执行本发明各个实施方式方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器（ROM，Read-Only Memory）、随机存取存储器（RAM，Random Access Memory）、磁碟或者光盘等各种可以存储程序代码的介质，或者是计算机、服务器、手机、平板等终端设备。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本申请中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本申请所示的这些实施例，而是要符合与本申请所公开的原理和新颖特点相一致的最宽的范围。

Claims (10)

  1. 一种卵圆孔未闭检测方法，其特征在于，包括：
    利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割；其中，所述心尖四腔心切面视角动态影像数据包括设定数量的心动周期内的TTE影像数据和cTTE影像数据；
    计算所述TTE影像数据中每一帧图像的左心房ROI的灰度，并按照设定比例筛选出所述TTE影像数据的最小灰度分布区间，根据所述TTE影像数据的最小灰度分布区间从所述cTTE影像数据中筛选出对应区间的cTTE图像；
    对所述筛选的cTTE图像进行微泡检测和特征提取，并将提取的特征输入训练好的预测模型，通过所述预测模型输出卵圆孔闭合状态检测结果。
  2. 根据权利要求1所述的卵圆孔未闭检测方法，其特征在于，所述利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割前还包括：
    分别将所述TTE影像数据和cTTE影像数据转化成逐帧的TTE图像序列和cTTE图像序列；
    采用MD5算法对所述TTE图像序列和cTTE图像序列进行筛选，去除所述TTE图像序列和cTTE图像序列中的重复帧；
    采用GAD滤波算法对筛选后的TTE图像序列和cTTE图像序列进行滤波祛噪处理。
  3. 根据权利要求2所述的卵圆孔未闭检测方法，其特征在于，所述利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割具体为：
    利用2D U-net卷积神经网络分别建立TTE分割模型和cTTE分割模型，将所述TTE图像序列和cTTE图像序列分别输入TTE分割模型和cTTE分割模型，所述TTE分割模型和cTTE分割模型分别经过连续的下采样、池化操作获得输入图像的特征图，对所述特征图进行连续的上采样，使所述特征图恢复到输入时的尺寸，并分别输出所述TTE图像序列和cTTE图像序列中每一帧图像对应的左心房ROI分割结果。
  4. 根据权利要求3所述的卵圆孔未闭检测方法，其特征在于，所述根据所述TTE影像数据的最小灰度分布区间筛选出对应区间的cTTE图像具体为：
    结合所述TTE图像序列和cTTE图像序列中每一帧图像的左心房ROI的面积变化和心动周期变化，分别筛选出设定数量的心动周期的TTE图像序列和cTTE图像序列，并利用双线性差值使所述筛选后的TTE图像序列和cTTE图像序列的帧数一致；
    针对筛选后的TTE图像序列，逐帧计算每一帧TTE图像的ROI内的灰度并求均值，生成TTE图像序列的灰度均值分布图，并按照设定的区间范围从所述灰度均值分布图中筛选出灰度均值小于设定区间比例的最小灰度分布区间；
    根据所述最小灰度分布区间从所述设定数量的心动周期的cTTE图像序列中筛选出与该灰度分布区间相对应的cTTE图像序列。
  5. 根据权利要求4所述的卵圆孔未闭检测方法，其特征在于，所述对所述筛选的cTTE图像进行微泡检测具体为：
    利用基于色彩和空间相似度约束的简单线性迭代聚类算法对所述筛选出的cTTE图像序列中每一帧cTTE图像的左心房ROI分别进行超像素分割检测微泡；
    对于检测到微泡的左心房ROI，首先基于微泡的灰度特征和圆度形态特征提取表征各超像素块的特征量并计算显著图，设定灰度阈值约束获得所述左心房ROI的微泡粗分割结果；然后，通过类圆或类椭圆性的圆度形态，设定几何阈值约束获得所述左心房ROI的微泡细分割结果。
  6. 根据权利要求5所述的卵圆孔未闭检测方法，其特征在于，所述对所述筛选的cTTE图像进行特征提取具体为：
    利用Pyradiomics算法对存在微泡的cTTE图像序列中每一帧cTTE图像的左心房ROI分别进行形状特征、强度特征和纹理特征提取，并利用方差选择、t检验和LASSO回归算法对提取的特征进行筛选和降维处理。
  7. 根据权利要求1至6任一项所述的卵圆孔未闭检测方法，其特征在于，所述预测模型为SVM模型。
  8. 一种卵圆孔未闭检测系统，其特征在于，包括：
    感兴趣区域分割模块：用于利用卷积神经网络对心尖四腔心切面视角动态影像数据中的每一帧图像分别进行左心房ROI分割；其中，所述心尖四腔心切面视角动态影像数据包括设定数量的心动周期内的TTE影像数据和cTTE影像数据；
    伪影祛除模块：用于计算所述TTE影像数据中每一帧图像的左心房ROI的灰度，并按照设定比例筛选出所述TTE影像数据的最小灰度分布区间，根据所述TTE影像数据的最小灰度分布区间从所述cTTE影像数据中筛选出对应区间的cTTE图像；
    微泡检测模块：用于对所述筛选的cTTE图像进行微泡检测，得到存在微泡的微泡图像序列；
    特征提取模块：用于对所述微泡图像序列进行特征提取；
    预测分类模块：用于将所述从微泡图像序列中提取的特征输入训练好的预测模型，通过所述预测模型输出卵圆孔闭合状态检测结果。
  9. 一种终端，其特征在于，所述终端包括处理器、与所述处理器耦接的存储器，其中，
    所述存储器存储有用于实现权利要求1-7任一项所述的卵圆孔未闭检测方法的程序指令；
    所述处理器用于执行所述存储器存储的所述程序指令以控制卵圆孔未闭检测。
  10. 一种存储介质，其特征在于，存储有处理器可运行的程序指令，所述程序指令用于执行权利要求1-7任一项所述卵圆孔未闭检测方法。
PCT/CN2020/139680 2020-11-30 2020-12-25 一种卵圆孔未闭检测方法、系统、终端以及存储介质 Ceased WO2022110445A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011373095.3A CN112435247B (zh) 2020-11-30 2020-11-30 一种卵圆孔未闭检测方法、系统、终端以及存储介质
CN202011373095.3 2020-11-30

Publications (1)

Publication Number Publication Date
WO2022110445A1 true WO2022110445A1 (zh) 2022-06-02

Family

ID=74699348

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/139680 Ceased WO2022110445A1 (zh) 2020-11-30 2020-12-25 一种卵圆孔未闭检测方法、系统、终端以及存储介质

Country Status (2)

Country Link
CN (1) CN112435247B (zh)
WO (1) WO2022110445A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118010782A (zh) * 2024-04-10 2024-05-10 广东省农业科学院农业资源与环境研究所 一种基于农业废弃物的富硅生物炭检测方法及系统
CN118553438A (zh) * 2024-07-30 2024-08-27 首都医科大学附属北京天坛医院 一种卵圆孔未闭患者的电子病历信息处理方法
CN118940537A (zh) * 2024-08-08 2024-11-12 深圳技术大学 复合件结合强度的预测方法、预测装置以及计算机可读存储介质

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113180737B (zh) * 2021-05-06 2022-02-08 中国人民解放军总医院 基于人工智能的卵圆孔闭合检测方法、系统、设备及介质
US12554796B2 (en) * 2021-05-28 2026-02-17 Nvidia Corporation Optimizing parameter estimation for training neural networks
CN116167957B (zh) * 2021-11-15 2025-02-18 四川大学华西医院 cTTE图像处理方法、计算机设备、系统和存储介质
CN114565622B (zh) * 2022-03-03 2023-04-07 北京安德医智科技有限公司 房间隔缺损长度的确定方法及装置、电子设备和存储介质
CN114612421A (zh) * 2022-03-07 2022-06-10 河南科技大学 一种基于深度学习的卵圆孔未闭微泡计数方法
CN116823697A (zh) * 2022-03-18 2023-09-29 中国科学院深圳先进技术研究院 一种微泡检测方法、装置、存储介质及终端设备
CN115601604B (zh) * 2022-11-29 2023-04-07 西南石油大学 一种基于长短时记忆网络的多任务微泡轨迹追踪方法
CN117197594B (zh) * 2023-11-07 2024-01-02 西南石油大学 一种基于深度神经网络的心脏分流分类系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6708055B2 (en) * 1998-08-25 2004-03-16 University Of Florida Method for automated analysis of apical four-chamber images of the heart
US7805011B2 (en) * 2006-09-13 2010-09-28 Warner Bros. Entertainment Inc. Method and apparatus for providing lossless data compression and editing media content
CN106991694A (zh) * 2017-03-17 2017-07-28 西安电子科技大学 基于显著区域面积匹配的心脏ct与超声图像配准方法
CN107361791A (zh) * 2017-07-21 2017-11-21 北京大学 一种快速超分辨血流成像方法
CN109087318A (zh) * 2018-07-26 2018-12-25 东北大学 一种基于优化U-net网络模型的MRI脑肿瘤图像分割方法
CN109801294A (zh) * 2018-12-14 2019-05-24 深圳先进技术研究院 三维左心房分割方法、装置、终端设备及存储介质
CN110719755A (zh) * 2017-06-08 2020-01-21 皇家飞利浦有限公司 超声成像方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080095417A1 (en) * 2006-10-23 2008-04-24 Gianni Pedrizzetti Method for registering images of a sequence of images, particularly ultrasound diagnostic images
US10617396B2 (en) * 2016-10-17 2020-04-14 International Business Machines Corporation Detection of valve disease from analysis of doppler waveforms exploiting the echocardiography annotations
CN109118528A (zh) * 2018-07-24 2019-01-01 西安工程大学 基于区域分块的奇异值分解图像匹配算法
CN111493935B (zh) * 2020-04-29 2021-01-15 中国人民解放军总医院 基于人工智能的超声心动图自动预测识别方法及系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6708055B2 (en) * 1998-08-25 2004-03-16 University Of Florida Method for automated analysis of apical four-chamber images of the heart
US7805011B2 (en) * 2006-09-13 2010-09-28 Warner Bros. Entertainment Inc. Method and apparatus for providing lossless data compression and editing media content
CN106991694A (zh) * 2017-03-17 2017-07-28 西安电子科技大学 基于显著区域面积匹配的心脏ct与超声图像配准方法
CN110719755A (zh) * 2017-06-08 2020-01-21 皇家飞利浦有限公司 超声成像方法
CN107361791A (zh) * 2017-07-21 2017-11-21 北京大学 一种快速超分辨血流成像方法
CN109087318A (zh) * 2018-07-26 2018-12-25 东北大学 一种基于优化U-net网络模型的MRI脑肿瘤图像分割方法
CN109801294A (zh) * 2018-12-14 2019-05-24 深圳先进技术研究院 三维左心房分割方法、装置、终端设备及存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
DU YAJUAN, ZHANG YUSHUN, CHENG GESHEN: "Application Stud on Diagnosis and Assessment of Shunt Direction in Adult Patients with PFO by TTE Combined with cTTE", CHINESE JOURNAL OF ULTRASOUND IN MEDICINE, vol. 30, no. 9, 1 September 2014 (2014-09-01), pages 800 - 803, XP055934867 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118010782A (zh) * 2024-04-10 2024-05-10 广东省农业科学院农业资源与环境研究所 一种基于农业废弃物的富硅生物炭检测方法及系统
CN118553438A (zh) * 2024-07-30 2024-08-27 首都医科大学附属北京天坛医院 一种卵圆孔未闭患者的电子病历信息处理方法
CN118940537A (zh) * 2024-08-08 2024-11-12 深圳技术大学 复合件结合强度的预测方法、预测装置以及计算机可读存储介质

Also Published As

Publication number Publication date
CN112435247B (zh) 2022-03-25
CN112435247A (zh) 2021-03-02

Similar Documents

Publication Publication Date Title
CN112435247B (zh) 一种卵圆孔未闭检测方法、系统、终端以及存储介质
CN102800089B (zh) 基于颈部超声图像的主颈动脉血管提取和厚度测量方法
CN104794708B (zh) 一种基于多特征学习的动脉粥样硬化斑块成分分割方法
CN104637044B (zh) 钙化斑块及其声影的超声图像提取系统
CN110232383A (zh) 一种基于深度学习模型的病灶图像识别方法及病灶图像识别系统
CN110613483B (zh) 一种基于机器学习检测胎儿颅脑异常的系统
CN110163809A (zh) 基于U-net生成对抗网络DSA成像方法及装置
CN108241865B (zh) 基于超声图像的多尺度多子图肝纤维化多级量化分期方法
CN111297399B (zh) 一种基于超声视频的胎心定位和胎心率提取方法
CN115456890A (zh) 基于多尺度双域判别器的生成对抗医学ct图像去噪方法
WO2021209887A1 (en) Rapid, accurate and machine-agnostic segmentation and quantification method and device for coronavirus ct-based diagnosis
CN102247144A (zh) 基于时间强度特征的乳腺病灶良恶性诊断计算机辅助方法
CN112419282A (zh) 脑部医学影像中动脉瘤的自动检测方法及系统
CN108670297A (zh) 基于多模态经颅超声的帕金森病辅助诊断系统及方法
Yang et al. The efficiency of a Machine learning approach based on Spatio-Temporal information in the detection of patent foramen ovale from contrast transthoracic echocardiography Images: A primary study
Sulas et al. Automatic detection of complete and measurable cardiac cycles in antenatal pulsed-wave Doppler signals
Raja et al. Segment based detection and quantification of kidney stones and its symmetric analysis using texture properties based on logical operators with ultrasound scanning
Fan et al. Diagnosis of brain abnormality using both structural and functional MR images
Bansal et al. Applying deep learning techniques in automated analysis of echocardiograms, CMRs and phonocardiograms for the detection and localization of cardiac diseases
Chen et al. Intelligent interpretation of four lung ultrasonographic features with split attention based deep learning model
CN116167957B (zh) cTTE图像处理方法、计算机设备、系统和存储介质
Mihăilescu et al. Automatic evaluation of steatosis by ultrasound image analysis
Matsakou et al. Automated detection of the carotid artery wall in longitudinal B-mode images using active contours initialized by the Hough transform
Caban et al. Texture-based computer-aided diagnosis system for lung fibrosis
Zhu et al. Automatic 3D segmentation of human airway tree in CT image

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20963333

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20963333

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 27.05.2024)

122 Ep: pct application non-entry in european phase

Ref document number: 20963333

Country of ref document: EP

Kind code of ref document: A1