Processing method and device for evaluating and applying CATCH sequence scheme

Abstract

The embodiment of the present invention relates to a processing method and device for evaluating and applying a CATCH sequence scheme, the method comprising: configuring a CATCH sequence scanning scheme set; constructing MR-T1 and MR-HIP models; and collecting a first acquisition data set; and training the MR-T1 model; and performing magnetic resonance CATCH sequence scanning imaging on each subject based on each scanning scheme and forming a second acquisition data set based on the imaging result; and performing scheme evaluation on the scanning scheme set according to the first and second acquisition data sets and the MR-T1 model; and jointly training the MR-T1 model and the MR-HIP model according to the best scanning scheme and the first and second acquisition data sets; performing magnetic resonance CATCH sequence scanning imaging on any subject based on the best scanning scheme and visualizing coronary high-risk plaque identification based on the imaging result. The present invention can improve the accuracy and efficiency of prediction / evaluation of coronary high-risk plaques.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a processing method and device for evaluating and applying a CATCH sequence scheme. Background Art

[0002] Plaques in coronary arteries (CAs) that are prone to rupture and lead to thrombosis are called high-risk plaques (HIP). Magnetic resonance angiography (MRA) technology can provide patients with a non-invasive means of coronary examination. Currently, T1-weighted images based on coronary MRA dark blood imaging technology are widely used in the field of prediction and identification of HIP plaques.

[0003] However, the T1-weighted images generated by conventional MRA scanning sequences have some inherent deficiencies: 1) they are only sensitive to proximal blood vessels and cannot identify the stenosis position of the entire coronary artery; 2) the low spatial resolution makes it difficult to effectively improve the recognition accuracy of HIP plaques. For this reason, some researchers have proposed a new MRA scanning sequence, the CATCH (Coronary Atherosclerosis T1-weighted Characterization with Integrated Anatomical Reference) sequence in the public technical document "Coronary Atherosclerosis T1-Weighed Characterization With Integrated Anatomical Reference", and pointed out that imaging through the CATCH sequence can not only obtain black blood T1-weighted images but also synchronously obtain bright blood anatomical reference images (anatomical reference images) of the entire coronary artery, and the spatial resolution of imaging can be significantly improved.

[0004] If the CATCH sequence imaging technology can be applied to the coronary HIP plaque detection project, it can not only identify the stenosis position of the whole coronary artery, but also effectively improve the recognition accuracy of HIP plaques. However, there are very few medical institutions that can apply this technology. The main reason is that before applying this technology, each medical institution needs to use the HIP plaque detection gold standard (intravascular ultrasound IVUS test results) as a reference to perform a large amount of data comparison and parameter optimization work on their respective CATCH sequence scanning schemes, and this kind of work is difficult to complete without a systematic technical solution. The technical problem to be solved by the present invention is how to provide each medical institution with a coronary HIP plaque prediction mechanism based on CATCH sequence imaging technology, and how to help each medical institution improve the evaluation efficiency of their respective CATCH sequence scanning schemes. Summary of the invention

[0005] The purpose of the present invention is to provide a processing method, device, electronic device and computer-readable storage medium for evaluating and applying CATCH sequence schemes in view of the defects of the prior art. The present invention pre-constructs a deep learning model for identifying stenotic vascular segments, namely an MR-T1 model, and a deep learning model for identifying high-risk plaques, namely an MR-HIP model; and configures multiple whole-heart coronary artery CATCH sequence scanning schemes to form a scanning scheme set; and constructs a first acquisition data set by collecting CTA images, DSA image sets, T1-weighted images and plaque detection sets of subjects, and performs a magnetic resonance CATCH sequence scanning imaging on each subject based on each scanning scheme to construct a second acquisition data set; and pre-trains the MR-T1 model based on the first acquisition data set, and then evaluates the best scanning scheme based on the first and second acquisition data sets and the MR-T1 model, and then jointly trains the MR-T1 model and the MR-HIP model based on the best scanning scheme and the first and second acquisition data sets; and after the joint training is completed, magnetic resonance CATCH sequence scanning imaging can be performed on any subject based on the best scanning scheme, and visual coronary high-risk plaque identification processing can be performed based on the imaging results of this time. The present invention can not only provide various medical institutions with a coronary HIP plaque prediction mechanism based on CATCH sequence imaging technology, but also provide various medical institutions with an evaluation mechanism for CATCH sequence scanning solutions, and can also improve the prediction / evaluation accuracy and corresponding prediction / evaluation efficiency through a deep learning model.

[0006] To achieve the above object, a first aspect of an embodiment of the present invention provides a processing method for evaluating and applying a CATCH sequence scheme, the method comprising:

[0007] A plurality of whole-heart coronary artery CATCH sequence scanning schemes are configured to form a corresponding first scanning scheme set; the first scanning scheme set includes a plurality of first scanning schemes; each of the first scanning schemes is a whole-heart coronary artery CATCH sequence scanning scheme;

[0008] Construct a deep learning model for identifying narrowed blood vessel segments, recorded as the MR-T1 model; and construct a deep learning model for identifying high-risk plaques, recorded as the MR-HIP model; and collect first CTA images, first DSA image sets, first T1 weighted images and first plaque detection sets of multiple first subjects to form a corresponding first acquisition data set; and train the MR-T1 model based on the first acquisition data set; and perform a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and form a corresponding second acquisition data set based on the imaging results; and evaluate the first scanning scheme set based on the first and second acquisition data sets and the MR-T1 model, and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and jointly train the MR-T1 model and the MR-HIP model based on the optimal scanning scheme and the first and second acquisition data sets;

[0009] Based on the optimal scanning scheme, a magnetic resonance CATCH sequence scanning imaging is performed on any subject, and visual identification of high-risk coronary plaques is performed based on the imaging results.

[0010] Preferably, the MR-T1 model is used to identify the stenotic vessel segment on the input T1-weighted image and output an output T1-weighted image with a stenotic vessel segment detection frame mark; the input and output T1-weighted images are both three-dimensional T1-weighted images;

[0011] The MR-T1 model is formed by sequentially connecting a first vessel semantic segmentation module, a first vessel tree recognition module, a first stenotic vessel segment recognition module and a first fusion module;

[0012] The first vascular semantic segmentation module is implemented based on a U-Net network; the first vascular semantic segmentation module is used to perform vascular semantic segmentation processing on the input T1-weighted image to obtain a corresponding first vascular semantic segmentation map; and send the first vascular semantic segmentation map to the first vascular tree recognition module and the first stenotic vascular segment recognition module; the first vascular semantic segmentation map is a three-dimensional binary semantic segmentation map; the voxel point types corresponding to each voxel point on the first vascular semantic segmentation map include vascular semantic voxel points and non-vascular semantic voxel points;

[0013] The first vascular tree recognition module is implemented based on a convolutional neural network; the first vascular tree recognition module is used to identify one or more connected vascular networks on the first vascular semantic segmentation map and construct a corresponding vascular tree mask map based on each connected vascular network to obtain one or more corresponding first vascular tree mask maps to form a corresponding first vascular tree mask map set and send it to the first stenotic vascular segment recognition module;

[0014] The first stenotic vessel segment identification module is composed of a first vessel classification unit, a first stenotic vessel segment region identification unit and a first stenotic vessel segment feature fusion unit;

[0015] The first vessel classification unit is implemented based on a classifier model; the first vessel classification unit is used to classify the coronary vessel type according to the geometric shape features of each of the first vessel tree mask images to obtain a corresponding first classification vector and send it to the first stenotic vessel segment feature fusion unit; the first classification vector includes a plurality of first classification probabilities, each of which corresponds to a coronary vessel type;

[0016] The first stenotic vascular segment region recognition unit is implemented based on a target detection model; the first stenotic vascular segment region recognition unit is used to perform corresponding vascular tree semantic segmentation map segmentation processing on the first vascular semantic segmentation map according to each of the first vascular tree mask maps to obtain a corresponding first segmentation semantic segmentation map; and perform stenotic vascular segment region recognition processing based on each of the first segmentation semantic segmentation maps, and if one or more stenotic vascular segment regions are identified during the recognition process, a corresponding first target detection frame is generated for each identified stenotic vascular segment region to form a corresponding first target detection frame set, and if the stenotic vascular segment region cannot be identified during the processing, the corresponding first target detection frame set is set to empty; and all the obtained first target detection frame sets are sent to the first stenotic vascular segment feature fusion unit; the first segmentation semantic segmentation map and the first vascular tree mask are combined into a corresponding first target detection frame set. Figure 1 One-to-one correspondence; the first target detection frame set and the first blood vessel tree mask Figure 1 One-to-one correspondence; when the first target detection frame set is not empty, it is composed of one or more first target detection frames; each first target detection frame includes a first target frame center point and a first target frame size;

[0017] The first stenosis vascular segment feature fusion unit is used to extract all the first target detection frames of all the first target detection frame sets to form a corresponding first target detection frame set; and identify whether the first target detection frame set is empty; if the first target detection frame set is empty, set the corresponding first stenosis segment target frame set to be empty; if the first target detection frame set is not empty, use each of the first target detection frames of the first target detection frame set as the corresponding current target detection frame, and use the first vascular tree mask map corresponding to the current target detection frame as the corresponding current vascular tree mask map, perform vascular tree entry point identification on the current vascular tree mask map to obtain the corresponding current vascular entry point, and mark the position point corresponding to the first target frame center point of the current target detection frame on the current vascular tree mask map to obtain the corresponding current target frame center point, and mark the position point from the current vascular entry point to the current vascular tree mask map on the current vascular tree mask map The blood vessel length of the center point of the current target frame is estimated and the estimated length is used as the corresponding first target frame distance, and the coronary vessel type corresponding to the first classification probability with the largest probability value in the first classification vector corresponding to the current blood vessel tree mask map is used as the corresponding first target frame vessel type, and a corresponding first stenosis segment target frame is formed by the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance corresponding to the current target detection frame, and a corresponding first stenosis segment target frame set is formed by all the first stenosis segment target frames obtained; and the obtained first stenosis segment target frame set is output to the first fusion module; when the first stenosis segment target frame set is not empty, it is composed of one or more first stenosis segment target frames; the first stenosis segment target frame includes the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance;

[0018] The first fusion module is used to identify whether the first stenosis segment target frame set is empty; if the first stenosis segment target frame set is empty, the corresponding output T1 weighted image is set as the input T1 weighted image and output; if the first stenosis segment target frame set is not empty, the corresponding stenosis segment target frame marking processing is performed on the input T1 weighted image based on each of the first stenosis segment target frames in the first stenosis segment target frame set, and the marked T1 weighted image is used as the corresponding output T1 weighted image and output.

[0019] Preferably, the MR-HIP model is used to identify high-risk plaques on the input T1-anatomical fusion image and output the corresponding high-risk plaque identification result; the T1-anatomical fusion image is formed by fusion of a T1-weighted image of a plaque position and its corresponding anatomical reference image; the high-risk plaque identification result includes high-risk plaques and non-high-risk plaques;

[0020] The MR-HIP model is formed by sequentially connecting a first feature extraction module, a first classification module and a first output module;

[0021] The first feature extraction module is implemented based on a residual network; the first feature extraction module is used to perform MR signal feature extraction processing on the input T1-anatomical fusion image to obtain a corresponding first feature tensor and send it to the first classification module;

[0022] The first classification module is implemented based on a binary classifier model; the first classification module is used to perform high-risk plaque classification prediction according to the first feature tensor to obtain a corresponding second classification vector and send it to the first output module; the second classification vector includes a high-risk plaque prediction probability and a non-high-risk plaque prediction probability;

[0023] The first output module is used to take the predicted probability with the largest probability value in the second classification vector as the corresponding current maximum predicted probability, and to set the corresponding high-risk plaque identification result as a high-risk plaque when the current maximum predicted probability is the high-risk plaque prediction probability, and to set the corresponding high-risk plaque identification result as a non-high-risk plaque when the current maximum predicted probability is the non-high-risk plaque prediction probability; and output the obtained high-risk plaque identification result.

[0024] Preferably, the first acquisition data set includes a plurality of first subject records; the first subject records correspond one-to-one to the first subject; the first subject records include the first CTA image, the first DSA image set, the first T1-weighted image and the first plaque detection set;

[0025] The first CTA image is a corresponding three-dimensional whole-heart coronary CTA image of the first subject; one or more first label detection frames are pre-marked on the first CTA image; each of the first label detection frames corresponds to a stenotic blood vessel segment; each of the first label detection frames includes a first label frame center point, a first label frame size, a first label frame blood vessel type and a first label frame distance; the first label frame blood vessel type is the blood vessel type of the corresponding coronary blood vessel where the first label detection frame is located; the first label frame distance is the blood vessel length between the corresponding first label frame center point and the blood vessel entry point of the coronary blood vessel where the first label detection frame is located;

[0026] The first DSA image set includes multiple first DSA images; each of the first DSA images corresponds to a coronary blood vessel type; the first DSA image is a three-dimensional DSA image of a type of coronary blood vessels of the corresponding first subject; one or more second label detection frames are pre-marked on the first DSA image; each of the second label detection frames corresponds to a stenotic blood vessel segment; each of the second label detection frames includes a second label frame center point, a second label frame size, a second label frame blood vessel type, and a second label frame distance; the second label frame blood vessel type is the blood vessel type of the corresponding coronary blood vessel where the second label detection frame is located; the second label frame distance is the blood vessel length between the corresponding second label frame center point and the blood vessel entry point of the coronary blood vessel where the second label detection frame is located; the second label detection frame corresponds to the first label detection frame one by one;

[0027] The first T1-weighted image is a three-dimensional whole-heart coronary T1-weighted image of the corresponding first subject, and the first T1-weighted image is generated based on a non-CATCH sequence; one or more third label detection frames are pre-marked on the first T1-weighted image; each of the third label detection frames corresponds to a stenotic vascular segment; each of the third label detection frames includes a third label frame center point, a third label frame size, a third label frame vessel type, and a third label frame distance; the third label frame vessel type is the vessel type of the corresponding coronary vessel where the third label detection frame is located; the third label frame distance is the vessel length between the corresponding third label frame center point and the vessel entry point of the coronary vessel where the third label detection frame is located; the third label detection frame corresponds to the second label detection frame one by one;

[0028] The first plaque detection set includes a plurality of first plaque detection results; the first plaque detection results include high-risk plaques and non-high-risk plaques; the first plaque detection results correspond one-to-one to the second label detection frame; the first plaque detection results are plaque detection results obtained through IVUS examination;

[0029] The second acquisition data set includes multiple first subject image sets; the first subject image sets correspond one-to-one to the first subject; the first subject image set includes multiple first plan image sets; the first plan image set corresponds one-to-one to the first scanning plan; the first plan image set includes a second T1-weighted image and a first anatomical reference image.

[0030] Preferably, the training of the MR-T1 model based on the first acquired data set specifically includes:

[0031] Step 51, traverse the first subject record of the first acquisition data set; and during the traversal, use the first subject record currently traversed as the corresponding current subject record; and use the first T1 weighted image of the current subject record as the corresponding first label T1 weighted image, and delete all the third label detection frames marked on the first label T1 weighted image to obtain the corresponding first training T1 weighted image, and form a corresponding first training data record with the obtained first training T1 weighted image and the first label T1 weighted image; and at the end of the traversal, all the obtained first training data records form the corresponding first training data set;

[0032] Step 52, extracting the first first training data record of the first training data set as the corresponding current training data record;

[0033] Step 53, inputting the T1-weighted image recorded in the current training data into the MR-T1 model to identify the stenotic blood vessel segment to obtain a corresponding first predicted T1-weighted image;

[0034] Step 54, bringing the first predicted T1 weighted image and the first label T1 weighted image of the current training data record into a preset first loss function to calculate and obtain a corresponding first loss value; the first loss function uses a cross entropy loss function by default;

[0035] Step 55, identifying whether the first loss value satisfies a preset first loss value range; if the first loss value satisfies the first loss value range, identifying whether the current training data record is the last first training data record of the first training data set, if so, going to step 56, if not, extracting the next first training data record of the first training data set as the new current training data record and returning to step 53 to continue training; if the first loss value does not satisfy the first loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 53 to continue training at the end of this parameter optimization; the model parameter optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm;

[0036] Step 56, solidify the model parameters of the MR-T1 model and confirm that the model training is completed.

[0037] Preferably, performing a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and forming a corresponding second acquisition data set based on the imaging results specifically includes:

[0038] Based on each of the first scanning schemes, a magnetic resonance CATCH sequence scanning imaging is performed on each of the first subjects to obtain the corresponding second T1-weighted image and the first anatomical reference image to form a corresponding first scheme image set; and all the first scheme image sets of each of the first subjects form a corresponding first subject image set; and all the obtained first subject image sets form a corresponding second acquisition data set.

[0039] Preferably, the performing scheme evaluation on the first scanning scheme set according to the first and second acquisition data sets and the MR-T1 model and taking the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme specifically includes:

[0040] Record the total number of the first subjects as a first total number M, and record the total number of the first scanning schemes in the first scanning scheme set as a second total number N; and extract the first CTA images, the first DSA image sets, and the first subject image sets corresponding to each of the first subjects from the first and second acquisition data sets to form a corresponding first subject data set;

[0041] Traversing all the first subject data sets; and during the traversal, taking the currently traversed first subject data set as the corresponding current subject data set; and taking the first CTA image, the first DSA image set and the first subject image set of the current subject data set as the corresponding current CTA image, the current DSA image set and the current subject image set; and performing vector conversion on each of the first label detection frames of the current CTA image to obtain a corresponding first label detection frame vector, and forming a corresponding first label frame vector set from all the obtained first label detection frame vectors; and performing vector conversion on each of the second label detection frames of the current DSA image set to obtain a corresponding second label detection frame vector, And all the obtained second label detection frame vectors form a corresponding second label frame vector set; and the second T1 weighted images of each of the first scheme image sets of the current subject image set are input into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third T1 weighted image, and each stenotic segment target frame marked on the current third T1 weighted image is vector-converted to obtain the corresponding first stenotic segment target frame vector, and all the first stenotic segment target frame vectors corresponding to the current third T1 weighted image form a corresponding first prediction frame vector set; and the matching degree of each of the first prediction frame vector sets with the corresponding first and second label frame vector sets is evaluated to obtain the corresponding first and second evaluation values ​​A. i,j , B i,j , and according to the first and second evaluation values ​​A i,j , Bi,j Calculate the corresponding third evaluation value C i,j =α×A i,j +β×B i,j , i is the subject index number of each of the first subjects, j is the scheme index number of each of the first scanning schemes, 1≤i≤M, 1≤j≤N, α and β are two preset weighting parameters, 0<α<β<1, α+β=1; and the N first scanning schemes corresponding to the current subject data set are evaluated according to the third evaluation value C i,j Sort them from high to low to get the corresponding first subject solution sequence;

[0042] After the traversal of all the first subject data sets is completed, all the first scanning schemes are traversed; and during the traversal, the first scanning scheme currently traversed is used as the corresponding current scanning scheme; and in the obtained M first subject scheme sequences, the number of times the current scanning scheme appears at the kth index position of the sequence is counted to obtain the corresponding first position number D k , 1≤k≤N, 0≤D k ≤M; and the N first position times D corresponding to the current scanning scheme k Calculate the corresponding

[0043] After traversing all the first scanning schemes, the first scanning scheme with the highest first scheme evaluation score is used as the corresponding optimal scanning scheme.

[0044] Preferably, the joint training of the MR-T1 model and the MR-HIP model according to the optimal scanning scheme and the first and second acquisition data sets specifically includes:

[0045] Step 81, traverse all the first subjects; and during the traversal, take the first subject currently traversed as the corresponding current subject; and take the first DSA image set and the first plaque detection set corresponding to the current subject in the first acquisition data set as the corresponding current DSA image set and current plaque detection set; and take the first scheme image set corresponding to the optimal scanning scheme in the first subject image set corresponding to the current subject in the second acquisition data set as the corresponding current scheme image set; and take each of the second label detection frames in the current DSA image set as the corresponding first label frame to form a corresponding first label frame set; and take each of the first label detection frames in the current plaque detection set as the corresponding first label frame set. The first plaque detection result corresponding to the frame is used as the corresponding first label type to form a corresponding first label type set; and the second T1 weighted image of the current scheme image set and the first anatomical reference image are subjected to image fusion processing to obtain a corresponding first fused image; and the second T1 weighted image of the current scheme image set and the corresponding first fused image are used as the corresponding second training T1 weighted image and the first training fused image; and the obtained second training T1 weighted image, the first training fused image, the first label frame set and the first label type set form a corresponding second training data record; and at the end of the traversal, all the obtained second training data records form a corresponding second training data set;

[0046] Step 82, extracting the first second training data record of the second training data set as the corresponding current training data record;

[0047] Step 83, inputting the second training T1 weighted image recorded in the current training data into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding second predicted T1 weighted image; extracting each stenotic segment target frame marked in the second predicted T1 weighted image as the corresponding first predicted frame; and pairing each first predicted frame with each first label frame recorded in the current training data one by one according to the shortest distance principle to obtain a plurality of first paired groups; the first paired group includes one first predicted frame and one matched first label frame;

[0048] Step 84, bringing all the first pairing groups obtained into a preset second loss function to calculate and obtain corresponding second loss values; the second loss function is an L2 loss function;

[0049] Step 85, identifying whether the second loss value satisfies a preset second loss value range; if the second loss value satisfies the second loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, going to step 86, if not, extracting the next second training data record of the second training data set as the new current training data record and returning to step 83 to continue training; if the second loss value does not satisfy the second loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 83 to continue training at the end of this parameter optimization; the model parameter optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm;

[0050] Step 86, solidifying the model parameters of the MR-T1 model once; and extracting the first second training data record of the second training data set as the corresponding current training data record;

[0051] Step 87, input the second training T1 weighted image of the current training data record into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third predicted T1 weighted image; and extract each stenotic segment target frame marked in the third predicted T1 weighted image as the corresponding second predicted frame; and according to the shortest distance principle, each second predicted frame is matched with each first label frame of the current training data record one by one to obtain a plurality of second pairing groups to form a corresponding second pairing group set; the second pairing group set includes a plurality of second pairing groups; the second pairing group includes one second predicted frame and one matched first label frame;

[0052] Step 88, taking each of the second pairing groups of the second pairing group set as the corresponding current pairing group; and extracting the sub-image area covered by the second prediction box of the current pairing group on the first training fusion image of the current training data record as the corresponding second training fusion image; and inputting the second training fusion image into the MR-HIP model to identify high-risk plaques and output the corresponding first prediction type; and taking the first label type corresponding to the first label box of the current pairing group in the first label type set of the current training data record as the corresponding second label type; and forming a corresponding first type pair by the first prediction type and the second label type obtained this time; and forming a corresponding first type pair set by all the first type pairs obtained; and the first type pair corresponds to the second pairing group one by one; the first type pair includes one first prediction type and one corresponding second label type; and the first prediction type includes high-risk plaques and non-high-risk plaques;

[0053] Step 89, bringing the first type pair set into a preset third loss function to calculate a corresponding third loss value; the third loss function is a cross entropy loss function;

[0054] Step 90, identifying whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, turning to step 91, if not, extracting the next second training data record of the second training data set as the new current training data record and returning to step 87 to continue training; if the third loss value does not satisfy the third loss value range, performing a parameter optimization of the MR-HIP model based on the model parameter optimization algorithm, and returning to step 88 to continue training at the end of this parameter optimization;

[0055] Step 91, solidify the model parameters of the MR-HIP model and confirm that the joint training is completed.

[0056] Preferably, the performing of a magnetic resonance CATCH sequence scanning imaging on any subject based on the optimal scanning scheme and performing visual identification of high-risk coronary plaques based on the imaging results specifically includes:

[0057] Performing a magnetic resonance CATCH sequence scanning imaging on the current subject based on the optimal scanning scheme to obtain a corresponding fourth T1-weighted image and a second anatomical reference image;

[0058] and performing image fusion processing on the fourth T1-weighted image and the second anatomical reference image to obtain a corresponding second fused image;

[0059] and inputting the fourth T1 weighted image into the MR-T1 model to identify the stenotic blood vessel segment to obtain a corresponding fifth T1 weighted image;

[0060] When the number of stenosis target frames marked on the fifth T1 weighted image is not 0, extract each stenosis target frame marked on the fifth T1 weighted image as a corresponding first marking frame; perform equivalent marking on each of the first marking frames on the second anatomical reference image to obtain a corresponding second marking frame; perform equivalent marking on each of the first marking frames on the second fused image to obtain a corresponding third marking frame; extract a sub-image covered by each of the third marking frames on the second fused image as a corresponding third fused image; input each of the third fused images into the MR-HIP model to perform high-risk plaque identification and output a corresponding first high-risk plaque identification result; and perform plaque type marking on each of the third marking frames using the corresponding first high-risk plaque identification result on the third fused image;

[0061] The fifth T1-weighted image, the second anatomical reference image and the second fused image finally obtained are displayed.

[0062] A second aspect of an embodiment of the present invention provides a device for implementing the processing method for evaluating and applying the CATCH sequence scheme described in the first aspect, the device comprising: a CATCH sequence scanning scheme configuration module, a scanning scheme evaluation and deep learning model training module, and a coronary high-risk plaque visualization processing module;

[0063] The CATCH sequence scanning scheme configuration module is used to configure a plurality of whole heart coronary artery CATCH sequence scanning schemes to form a corresponding first scanning scheme set; the first scanning scheme set includes a plurality of first scanning schemes; each of the first scanning schemes is a whole heart coronary artery CATCH sequence scanning scheme;

[0064] The scanning scheme evaluation and deep learning model training module is used to construct a deep learning model for identifying narrowed blood vessel segments, which is recorded as an MR-T1 model; and to construct a deep learning model for identifying high-risk plaques, which is recorded as an MR-HIP model; and to collect first CTA images, first DSA image sets, first T1 weighted images and first plaque detection sets of multiple first subjects to form a corresponding first acquisition data set; and to train the MR-T1 model based on the first acquisition data set; and to perform a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and to form a corresponding second acquisition data set based on the imaging results; and to evaluate the first scanning scheme set based on the first and second acquisition data sets and the MR-T1 model and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and to jointly train the MR-T1 model and the MR-HIP model based on the optimal scanning scheme and the first and second acquisition data sets;

[0065] The coronary high-risk plaque visualization processing module is used to perform a magnetic resonance CATCH sequence scanning imaging on any subject based on the optimal scanning scheme and to perform visualized coronary high-risk plaque identification based on the imaging results.

[0066] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0067] The processor is used to be coupled to the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0068] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0069] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0070] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for evaluating and applying CATCH sequence schemes. As can be seen from the above content, the embodiment of the present invention pre-constructs a deep learning model for identifying stenotic vascular segments, namely, an MR-T1 model, and a deep learning model for identifying high-risk plaques, namely, an MR-HIP model; and configures multiple whole-heart coronary artery CATCH sequence scanning schemes to form a scanning scheme set; and constructs a first acquisition data set by collecting CTA images, DSA image sets, T1-weighted images and plaque detection sets of subjects, and performs a magnetic resonance CATCH sequence scanning imaging on each subject based on each scanning scheme to construct a second acquisition data set; and pre-trains the MR-T1 model based on the first acquisition data set, and then evaluates the best scanning scheme based on the first and second acquisition data sets and the MR-T1 model, and then jointly trains the MR-T1 model and the MR-HIP model based on the best scanning scheme and the first and second acquisition data sets; and after the joint training is completed, magnetic resonance CATCH sequence scanning imaging can be performed on any subject based on the best scanning scheme, and visual coronary high-risk plaque identification processing can be performed based on this imaging result. The embodiments of the present invention not only provide various medical institutions with a coronary HIP plaque prediction mechanism based on CATCH sequence imaging technology, but also provide various medical institutions with an evaluation mechanism for CATCH sequence scanning schemes, and also improve the prediction / evaluation accuracy and corresponding prediction / evaluation efficiency through a deep learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1A schematic diagram of a processing method for evaluating and applying a CATCH sequence scheme provided in Embodiment 1 of the present invention;

[0072] Figure 2 A module structure diagram of the MR-T1 model and the MR-HIP model provided in the first embodiment of the present invention;

[0073] Figure 3 A module structure diagram of a processing device for evaluating and applying a CATCH sequence scheme provided in Embodiment 2 of the present invention;

[0074] Figure 4 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] Embodiment 1 of the present invention provides a processing method for evaluating and applying a CATCH sequence scheme, such as Figure 1 As shown in the schematic diagram of a processing method for evaluating and applying a CATCH sequence scheme provided in the first embodiment of the present invention, the method mainly includes the following steps:

[0077] Step 1, configuring multiple whole-heart coronary artery CATCH sequence scanning schemes to form a corresponding first scanning scheme set;

[0078] The first scanning scheme set includes a plurality of first scanning schemes; each first scanning scheme is a whole-heart coronary artery CATCH sequence scanning scheme.

[0079] Step 2, construct a deep learning model for identifying narrowed vascular segments, recorded as the MR-T1 model; and construct a deep learning model for identifying high-risk plaques, recorded as the MR-HIP model; and collect a first CTA image, a first DSA image set, a first T1 weighted image and a first plaque detection set of multiple first subjects to form a corresponding first acquisition data set; and train the MR-T1 model based on the first acquisition data set; and perform a magnetic resonance CATCH sequence scanning imaging on each first subject based on each first scanning scheme and form a corresponding second acquisition data set based on the imaging results; and evaluate the first scanning scheme set according to the first and second acquisition data sets and the MR-T1 model, and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and jointly train the MR-T1 model and the MR-HIP model according to the optimal scanning scheme and the first and second acquisition data sets;

[0080] Specifically, the method comprises: step 21, constructing a deep learning model for identifying stenotic vascular segments, which is recorded as the MR-T1 model; and constructing a deep learning model for identifying high-risk plaques, which is recorded as the MR-HIP model;

[0081] Here, if Figure 2 The module structure diagram of the MR-T1 model and the MR-HIP model provided in the first embodiment of the present invention is as follows:

[0082] 1) MR-T1 model:

[0083] The MR-T1 model provided in the embodiment of the present invention is used to identify the stenotic blood vessel segment of the input T1 weighted image and output the output T1 weighted image with the stenotic blood vessel segment detection frame mark; wherein the input and output T1 weighted images are both three-dimensional T1 weighted images;

[0084] The MR-T1 model of the embodiment of the present invention is formed by sequentially connecting a first vessel semantic segmentation module, a first vessel tree recognition module, a first stenotic vessel segment recognition module and a first fusion module;

[0085] The first vascular semantic segmentation module is implemented based on the U-Net network; the first vascular semantic segmentation module is used to perform vascular semantic segmentation processing on the input T1 weighted image to obtain the corresponding first vascular semantic segmentation map; and send the first vascular semantic segmentation map to the first vascular tree recognition module and the first stenotic vascular segment recognition module; here, the first vascular semantic segmentation map is a three-dimensional binary semantic segmentation map; the voxel point types corresponding to each voxel point on the first vascular semantic segmentation map include vascular semantic voxel points and non-vascular semantic voxel points;

[0086] The first vascular tree recognition module is implemented based on a convolutional neural network; the first vascular tree recognition module is used to identify one or more connected vascular networks on the first vascular semantic segmentation map and construct a corresponding vascular tree mask map based on each connected vascular network to obtain one or more corresponding first vascular tree mask maps to form a corresponding first vascular tree mask map set and send it to the first stenotic vascular segment recognition module;

[0087] The first stenotic vessel segment identification module is composed of a first vessel classification unit, a first stenotic vessel segment region identification unit and a first stenotic vessel segment feature fusion unit;

[0088] The first vessel classification unit is implemented based on a classifier model; the first vessel classification unit is used to classify the coronary vessel type according to the geometric shape features of each first vessel tree mask image to obtain a corresponding first classification vector and send it to the first stenosis vessel segment feature fusion unit; wherein the first classification vector includes a plurality of first classification probabilities, each of which corresponds to a coronary vessel type;

[0089] The first stenotic vascular segment region recognition unit is implemented based on the target detection model; the first stenotic vascular segment region recognition unit is used to perform corresponding vascular tree semantic segmentation map segmentation processing on the first vascular semantic segmentation map according to each first vascular tree mask map to obtain a corresponding first segmentation semantic segmentation map; and perform stenotic vascular segment region recognition processing based on each first segmentation semantic segmentation map, and if one or more stenotic vascular segment regions are identified during the recognition process, a corresponding first target detection frame is generated for each identified stenotic vascular segment region to form a corresponding first target detection frame set, and if the stenotic vascular segment region cannot be identified during the processing, the corresponding first target detection frame set is set to empty; and all the obtained first target detection frame sets are sent to the first stenotic vascular segment feature fusion unit; wherein, the first segmentation semantic segmentation map and the first vascular tree mask map are combined to form a first target detection frame set. Figure 1 One-to-one correspondence: the first target detection frame set and the first blood vessel tree mask Figure 1 One-to-one correspondence; when the first target detection frame set is not empty, it is composed of one or more first target detection frames; each first target detection frame includes a first target frame center point and a first target frame size;

[0090] The first stenosis vascular segment feature fusion unit is used to extract all the first target detection frames of all the first target detection frame sets to form a corresponding first target detection frame set; and identify whether the first target detection frame set is empty; if the first target detection frame set is empty, set the corresponding first stenosis segment target frame set to be empty; if the first target detection frame set is not empty, use each first target detection frame of the first target detection frame set as the corresponding current target detection frame, and use the first vascular tree mask map corresponding to the current target detection frame as the corresponding current vascular tree mask map, and perform vascular tree entry point identification on the current vascular tree mask map to obtain the corresponding current vascular entry point, and mark the position point corresponding to the center point of the first target frame of the current target detection frame on the current vascular tree mask map to obtain the corresponding current target frame center point, and mark the position point from the current vascular entrance on the current vascular tree mask map The blood vessel length from the point to the center point of the current target frame is estimated and the estimated length is used as the corresponding first target frame distance, and the coronary vessel type corresponding to the first classification probability with the largest probability value in the first classification vector corresponding to the current blood vessel tree mask map is used as the corresponding first target frame vessel type, and a corresponding first stenosis segment target frame is formed by the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance corresponding to the current target detection frame, and a corresponding first stenosis segment target frame set is formed by all the obtained first stenosis segment target frames; and the obtained first stenosis segment target frame set is output to the first fusion module; here, when the first stenosis segment target frame set is not empty, it is composed of one or more first stenosis segment target frames; the first stenosis segment target frame includes the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance;

[0091] The first fusion module is used to identify whether the first stenosis segment target frame set is empty; if the first stenosis segment target frame set is empty, the corresponding output T1 weighted image is set as the input T1 weighted image and output; if the first stenosis segment target frame set is not empty, the corresponding stenosis segment target frame marking process is performed on the input T1 weighted image based on each first stenosis segment target frame of the first stenosis segment target frame set, and the marked T1 weighted image is used as the corresponding output T1 weighted image and output;

[0092] 2) MR-HIP model:

[0093] The MR-HIP model is used to identify high-risk plaques on the input T1-anatomical fusion image and output the corresponding high-risk plaque identification results; wherein the T1-anatomical fusion image is formed by the fusion of a T1-weighted image of a plaque position and its corresponding anatomical reference image; the high-risk plaque identification results include high-risk plaques and non-high-risk plaques;

[0094] The MR-HIP model is composed of a first feature extraction module, a first classification module, and a first output module connected in sequence;

[0095] The first feature extraction module is implemented based on a residual network; the first feature extraction module is used to perform MR signal feature extraction processing on the input T1-anatomical fusion image to obtain a corresponding first feature tensor and send it to the first classification module;

[0096] The first classification module is implemented based on a binary classifier model; the first classification module is used to perform high-risk plaque classification prediction according to the first feature tensor to obtain a corresponding second classification vector and send it to the first output module; wherein the second classification vector includes a high-risk plaque prediction probability and a non-high-risk plaque prediction probability;

[0097] The first output module is used to take the predicted probability with the largest probability value in the second classification vector as the corresponding current maximum predicted probability, and set the corresponding high-risk plaque identification result as a high-risk plaque when the current maximum predicted probability is the high-risk plaque prediction probability, and set the corresponding high-risk plaque identification result as a non-high-risk plaque when the current maximum predicted probability is the non-high-risk plaque prediction probability; and output the obtained high-risk plaque identification result;

[0098] Step 22, collecting a first CTA image, a first DSA image set, a first T1-weighted image, and a first plaque detection set of a plurality of first subjects to form a corresponding first collection data set;

[0099] Here, the first acquisition data set of the embodiment of the present invention includes a plurality of first subject records; the first subject records correspond to the first subject one by one; the first subject records include a first CTA image, a first DSA image set, a first T1-weighted image and a first plaque detection set;

[0100] The first CTA image is a three-dimensional whole-heart coronary CTA image of the corresponding first subject; one or more first label detection frames are pre-marked on the first CTA image; each first label detection frame corresponds to a stenotic blood vessel segment; each first label detection frame includes a first label frame center point, a first label frame size, a first label frame blood vessel type and a first label frame distance; the first label frame blood vessel type is the blood vessel type of the coronary blood vessel where the corresponding first label detection frame is located; the first label frame distance is the blood vessel length between the corresponding first label frame center point and the blood vessel entry point of the coronary blood vessel where the first label detection frame is located;

[0101] The first Digital subtraction angiography (DSA) image set includes a plurality of first DSA images; each first DSA image corresponds to a coronary blood vessel type; the first DSA image is a three-dimensional DSA image of a type of coronary blood vessels of the corresponding first subject; one or more second label detection frames are pre-marked on the first DSA image; each second label detection frame corresponds to a stenotic blood vessel segment; each second label detection frame includes a second label frame center point, a second label frame size, a second label frame blood vessel type and a second label frame distance; the second label frame blood vessel type is the blood vessel type of the coronary blood vessel where the corresponding second label detection frame is located; the second label frame distance is the blood vessel length between the corresponding second label frame center point and the blood vessel entry point of the coronary blood vessel where the second label detection frame is located; the second label detection frame corresponds to the first label detection frame one by one;

[0102] The first T1-weighted image is a three-dimensional whole-heart coronary T1-weighted image of the corresponding first subject, and the first T1-weighted image is generated based on a non-CATCH sequence; one or more third label detection frames are pre-marked on the first T1-weighted image; each third label detection frame corresponds to a stenotic blood vessel segment; each third label detection frame includes a third label frame center point, a third label frame size, a third label frame blood vessel type, and a third label frame distance; the third label frame blood vessel type is the blood vessel type of the coronary blood vessel where the corresponding third label detection frame is located; the third label frame distance is the blood vessel length between the corresponding third label frame center point and the blood vessel entry point of the coronary blood vessel where the third label detection frame is located; the third label detection frame corresponds one-to-one to the second label detection frame;

[0103] The first plaque detection set includes multiple first plaque detection results; the first plaque detection results include high-risk plaques and non-high-risk plaques; the first plaque detection results correspond to the second label detection frames one by one; the first plaque detection results are plaque detection results obtained through IVUS examination;

[0104] It should be noted that the first subjects in the embodiment of the present invention are all patients with coronary heart disease of varying degrees. The CTA images, DSA images, IVUS examination results and MR-T1 weighted images generated by non-CATCH sequences of this group of people are collected for the purpose of training the MR-T1 model and the MR-HIP model, and for evaluating and optimizing the CATCH sequence scanning scheme by using the gold standard for coronary stenosis examination (CTA and DSA) and the gold standard for coronary plaque examination (IVUS). The above four types of data collected for each first subject (the first CTA image, the first The correspondence between the DSA image set, the first T1-weighted image and the first plaque detection set is as follows: the first CTA image is a three-dimensional whole-heart coronary CTA image of a first subject. The vascular network of the left and right coronary arteries of the whole heart can be displayed simultaneously on the CTA image, which will have one or more (set as x) vascular stenosis locations, and each location is marked with a corresponding detection frame, namely the first label detection frame; from conventional DSA examinations, we know that during the examination, image acquisition will be performed based on the correspondence between contrast fluid flow and maximum vascular filling, that is, a DSA image set will be acquired and each DSA image can only be a pair of To display a part of the left and right coronary artery vascular network, it is necessary to collect a DSA image set, namely the first DSA image set, to align it with the first CTA image. The total number and position of vascular stenosis in the first DSA image set should be aligned with the first CTA image, so the second label detection frame used to mark the stenotic blood vessels should correspond one to one with the first label detection frame; the first T1-weighted image is a T1-weighted image generated by another type of non-CATCH sequence technology that has been widely used in coronary stenosis examination based on MR technology. The T1-weighted image can also display the vascular network of the left and right coronary arteries of the whole heart at the same time. The first T1-weighted image The total number and location of vascular stenosis in the image should be aligned with the first CTA image and the first DSA image set, so the third label detection frame used to mark the stenotic blood vessels should correspond to the first and second label detection frames one by one. The purpose of collecting the first T1-weighted image is that the model structure of the MR-T1 model is relatively complex. In order to improve the prediction performance of the MR-T1 model, the MR-T1 model will be trained in a generalized manner based on the first T1-weighted image, and then the MR-T1 model will be trained in an enhanced manner based on the T1-weighted image generated by the CATCH sequence technology, that is, the second T1-weighted image mentioned below.

[0105] Step 23, training the MR-T1 model based on the first acquired data set;

[0106] Here, the current step is to perform a round of generalization training on the MR-T1 model based on the first T1 weighted image in the first acquisition data set. Because the third label detection frame in the first T1 weighted image corresponds one-to-one with the first and second label detection frames in the first CTA image and the first DSA image set, the generalization prediction performance of the MR-T1 model is actually improved by referring to the examination results of the gold standard (CTA and DSA) for coronary artery stenosis examination.

[0107] Specifically, the method comprises: step 231, traversing the first subject record of the first acquisition data set; and during the traversal, taking the first subject record currently traversed as the corresponding current subject record; and taking the first T1 weighted image of the current subject record as the corresponding first label T1 weighted image, and deleting all third label detection frames marked on the first label T1 weighted image to obtain the corresponding first training T1 weighted image, and forming a corresponding first training data record with the obtained first training T1 weighted image and the first label T1 weighted image; and at the end of the traversal, forming the corresponding first training data set with all the obtained first training data records;

[0108] Step 232, extracting the first first training data record of the first training data set as the corresponding current training data record;

[0109] Step 233, inputting the T1-weighted image recorded in the current training data into the MR-T1 model to identify the stenotic blood vessel segment to obtain the corresponding first predicted T1-weighted image;

[0110] Step 234, bringing the first predicted T1 weighted image and the first label T1 weighted image of the current training data record into a preset first loss function to calculate and obtain a corresponding first loss value;

[0111] Among them, the first loss function uses the cross entropy loss function by default;

[0112] Step 235, identifying whether the first loss value satisfies a preset first loss value range; if the first loss value satisfies the first loss value range, identifying whether the current training data record is the last first training data record of the first training data set, if so, turning to step 236, if not, extracting the next first training data record of the first training data set as a new current training data record and returning to step 233 to continue training; if the first loss value does not satisfy the first loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 233 to continue training at the end of this parameter optimization;

[0113] Among them, the model parameter optimization algorithms include SDG optimization algorithm and Adam optimization algorithm;

[0114] Here, the first loss value range is a preset loss value range;

[0115] Step 236, solidify the model parameters of the MR-T1 model and confirm that the model training is completed;

[0116] Step 24, performing a magnetic resonance CATCH sequence scanning imaging on each first subject based on each first scanning scheme and forming a corresponding second acquisition data set based on the imaging result;

[0117] Specifically, a magnetic resonance CATCH sequence scanning imaging is performed on each first subject based on each first scanning scheme to obtain a corresponding second T1-weighted image and a first anatomical reference image to form a corresponding first scheme image set; and all first scheme image sets of each first subject form a corresponding first subject image set; and all obtained first subject image sets form a corresponding second acquisition data set;

[0118] Here, the second acquisition data set of the embodiment of the present invention includes a plurality of first subject image sets; the first subject image sets correspond one-to-one to the first subject; the first subject image sets include a plurality of first scheme image sets; the first scheme image sets correspond one-to-one to the first scanning scheme; the first scheme image sets include the second T1-weighted image and the first anatomical reference image;

[0119] Step 25, performing scheme evaluation on the first scanning scheme set according to the first and second acquisition data sets and the MR-T1 model and taking the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme;

[0120] Specifically, the method comprises: step 251, recording the total number of first subjects as a first total number M, and recording the total number of first scanning schemes in the first scanning scheme set as a second total number N; and extracting the first CTA images, the first DSA image sets and the first subject image sets corresponding to each first subject from the first and second acquisition data sets to form a corresponding first subject data set;

[0121] Step 252, traverse all first subject data sets; and during the traversal, use the currently traversed first subject data set as the corresponding current subject data set; and use the first CTA image, the first DSA image set and the first subject image set of the current subject data set as the corresponding current CTA image, the current DSA image set and the current subject image set; and perform vector conversion on each first label detection frame of the current CTA image to obtain a corresponding first label detection frame vector, and form a corresponding first label frame vector set from all the obtained first label detection frame vectors; and perform vector conversion on each second label detection frame of the current DSA image set to obtain a corresponding second label detection frame vector. The amount is obtained, and all the obtained second label detection frame vectors are used to form a corresponding second label frame vector set; and the second T1 weighted images of each first scheme image set of the current subject image set are input into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third T1 weighted image, and each stenotic segment target frame marked on the current third T1 weighted image is vector-converted to obtain the corresponding first stenotic segment target frame vector, and all the first stenotic segment target frame vectors corresponding to the current third T1 weighted image are used to form a corresponding first prediction frame vector set; and the matching degree of each first prediction frame vector set with the corresponding first and second label frame vector sets is evaluated to obtain the corresponding first and second evaluation values ​​A. i,j , B i,j , and according to the first and second evaluation values ​​A i,j , B i,j Calculate the corresponding third evaluation value c i,j =α×A i,j +β×B i,j , i is the subject index number of each first subject, j is the scheme index number of each first scanning scheme, 1≤i≤M, 1≤j≤N, α and β are two preset weighting parameters, 0<α<β<1, α+β=1; and the N first scanning schemes corresponding to the current subject data set are evaluated according to the third evaluation value C i,j Sort them from high to low to get the corresponding first subject solution sequence;

[0122] Here, the embodiment of the present invention evaluates the imaging quality of T1-weighted images obtained by different CATCH sequence scanning schemes by evaluating the matching degree of each first prediction frame vector set with the corresponding first and second label frame vector sets. The first and second evaluation values ​​A i,j , B i,j The higher the third evaluation value C is, the more closely the position of the stenotic vessel detected on the current T1-weighted image matches the examination results of the gold standard for coronary artery stenosis (CTA and DSA), and the better the imaging quality of the CATCH sequence scanning scheme corresponding to the current T1-weighted image. i,j A is the first and second evaluation valuesi,j , B i,j The weighted sum of , because the clarity of DSA images is higher than that of CTA images, the weighting parameter β corresponding to DSA images is greater than the weighting parameter α corresponding to CTA images;

[0123] It should be noted that in practical applications, there are multiple evaluation methods for evaluating the matching degree between each first prediction box vector set and the corresponding first and second label box vector sets, and the evaluation method can be set based on application requirements; one of the evaluation methods is:

[0124] All first stenosis segment target frame vectors in the first prediction frame vector set are marked as unused vectors in advance; and each first label detection frame vector of the first label frame vector set is traversed; and when traversing, the currently traversed first label detection frame vector is used as the corresponding current label vector; and the first label frame blood vessel type and the first label frame distance of the current label vector are used as the corresponding current label type and the current label distance; and the first stenosis segment target frame vector whose target frame blood vessel type of the stenosis segment target frame in the first prediction frame vector set matches the current label type and is marked as an unused vector forms a corresponding first vector set; and whether the first vector set is empty is identified; if so, the corresponding first vector matching degree is set to 0; if not, the vascular inlet of the corresponding stenosis segment target frame center point and the coronary vessel in each first stenosis segment target frame vector in the first vector set is further matched. The target frame distance feature of the blood vessel length between the mouth points is recorded as the corresponding first target distance, and the absolute difference between the current label distance and each first target distance is calculated to obtain the corresponding first distance difference, and the shortest first distance difference is used as the corresponding first shortest distance difference, and when the first shortest distance difference is greater than a preset small distance threshold, the corresponding first vector matching degree is set to 0, and when the first shortest distance difference is less than or equal to the small distance threshold, the corresponding first vector matching degree is set to (1-first distance difference / small distance threshold); and after obtaining the first vector matching degree, whether the first vector matching degree is 0 is identified, and if it is not 0, the first stenosis segment target frame vector corresponding to the first shortest distance difference in the first prediction frame vector set is marked as a used vector; and at the end of the traversal, the average of all the first vector matching degrees obtained is calculated and the calculation result is used as the corresponding first evaluation value A i,j ;

[0125] All first stenosis target frame vectors in the first prediction frame vector set are marked as unused vectors in advance; each second label detection frame vector of the second label frame vector set is traversed; and when traversing, the currently traversed second label detection frame vector is used as the corresponding current label vector; and the second label frame blood vessel type and the second label frame distance of the current label vector are used as the corresponding current label type and current label distance; and the target frame blood vessel type of the stenosis target frame in the first prediction frame vector set matches the current label type and the first stenosis target frame vector marked as an unused vector is used to form a corresponding second vector set; and whether the second vector set is empty is identified; if so, the corresponding second vector matching degree is set to 0; if not, the blood vessel of the corresponding stenosis target frame center point and the coronary blood vessel in each first stenosis target frame vector in the second vector set is further matched. The target frame distance feature of the blood vessel length between the entry points is recorded as the corresponding second target distance, and the absolute difference between the current label distance and each second target distance is calculated to obtain the corresponding second distance difference, and the shortest second distance difference is used as the corresponding second shortest distance difference, and when the second shortest distance difference is greater than a small distance threshold, the corresponding second vector matching degree is set to 0, and when the second shortest distance difference is less than or equal to the small distance threshold, the corresponding second vector matching degree is set to (1-second distance difference / small distance threshold); and after obtaining the second vector matching degree, whether the second vector matching degree is 0 is identified, and if it is not 0, the first stenosis segment target frame vector corresponding to the second shortest distance difference in the first prediction frame vector set is marked as a used vector; and at the end of the traversal, the average of all the obtained second vector matching degrees is calculated and the calculation result is used as the corresponding second evaluation value B i,j ;

[0126] Step 253, after the traversal of all first subject data sets is completed, all first scanning schemes are traversed; and during the traversal, the first scanning scheme currently traversed is used as the corresponding current scanning scheme; and the number of times the current scanning scheme appears at the kth index position of the sequence in the obtained M first subject scheme sequences is counted to obtain the corresponding first position number D k , 1≤k≤N, 0≤D k ≤M; and the number of N first positions corresponding to the current scanning scheme D k Calculate the corresponding

[0127] Step 254: After the traversal of all first scanning schemes is completed, the first scanning scheme with the highest first scheme evaluation score is used as the corresponding optimal scanning scheme;

[0128] For example, there are M = 4 first subjects; there are N = 3 first scanning plans, namely plan 1, plan 2, and plan 3;

[0129] It is known that the M first subject solution sequences are:

[0130] The first subject's solution sequence 1 {Solution 1, Solution 2, Solution 3},

[0131] The first subject's solution sequence 2 {Solution 1, Solution 3, Solution 2},

[0132] The first subject's solution sequence 3 {Solution 2, Solution 1, Solution 3},

[0133] The first subject's solution sequence 4 {Solution 3, Solution 2, Solution 1},

[0134] Then, in step 253, the number of times that solution 1 appears in the first, second, and third index positions in the first subject's solution sequence 1-4 is counted to obtain the corresponding first position number D k＝1、2、3 =2, 1, 1, and the calculated first solution evaluation score 1 = 3×2+2×1+1×1=9; the number of times solution 2 appears in the first, second, and third index positions in the first subject's solution sequence 1-4 is counted to obtain the corresponding first position number D k＝1、2、3 =1, 2, 1, and the calculated first solution evaluation score 2 = 3×1+2×2+1×1=8; the number of times solution 3 appears in the first, second, and third index positions in the first subject's solution sequence 1-4 is counted to obtain the corresponding first position number D k＝1、2、3 =1, 1, 2, and the calculated evaluation score of the first plan is 3=3×1+2×1+1×2=7;

[0135] Then, step 254 is performed to select the maximum value, that is, the plan 1 corresponding to the first plan evaluation score 1, from the three plan evaluation scores, that is, the first plan evaluation scores 1, 2, and 3 as the best scanning plan; this screening method can reduce the plan evaluation error caused by the individual differences of the subjects, and achieve the purpose of improving the standardization and accuracy of the evaluation;

[0136] Step 26, jointly training the MR-T1 model and the MR-HIP model according to the optimal scanning scheme and the first and second acquisition data sets;

[0137] Here, in the embodiment of the present invention, when joint training is performed, the MR-T1 model is firstly subjected to a round of enhanced training using the T1 weighted images obtained from the optimal scanning scheme, and then the MR-HIP model is trained; it should be noted that when training the MR-T1 model, the initial parameters of the MR-T1 model should be the model parameters solidified after the training in step 23 above, and the MR-T1 model obtained through this round of enhanced training not only has good generalization performance but also can perform more accurate prediction of stenotic blood vessels on the T1 weighted images generated by the CATCH sequence, and can also further improve the training effect of the subsequent MR-HIP model training;

[0138] Specifically, it includes: step 26-1, traversing all first subjects; and during the traversal, taking the currently traversed first subject as the corresponding current subject; taking the first DSA image set and the first plaque detection set corresponding to the current subject in the first acquisition data set as the corresponding current DSA image set and the current plaque detection set; taking the first scheme image set corresponding to the best scanning scheme in the first subject image set corresponding to the current subject in the second acquisition data set as the corresponding current scheme image set; taking each second label detection frame of the current DSA image set as the corresponding first label frame to form a corresponding first label frame set; and taking each first label frame in the current plaque detection set as the corresponding first label frame set. The corresponding first plaque detection result is used as the corresponding first label type to form a corresponding first label type set; and the second T1 weighted image of the current scheme image set and the first anatomical reference image are subjected to image fusion processing to obtain the corresponding first fused image; and the second T1 weighted image of the current scheme image set and the corresponding first fused image are used as the corresponding second training T1 weighted image and the first training fused image; and the obtained second training T1 weighted image, the first training fused image, the first label frame set and the first label type set form a corresponding second training data record; and at the end of the traversal, all the obtained second training data records form a corresponding second training data set;

[0139] Step 26-2, extracting the first second training data record of the second training data set as the corresponding current training data record;

[0140] Step 26-3, inputting the second training T1 weighted image recorded in the current training data into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding second predicted T1 weighted image; extracting each stenotic segment target frame marked in the second predicted T1 weighted image as the corresponding first predicted frame; and pairing each first predicted frame with each first label frame recorded in the current training data one by one according to the shortest distance principle to obtain multiple first pairing groups;

[0141] The first pairing group includes a first prediction frame and a matching first label frame;

[0142] Here, each first prediction frame is matched with each first label frame of the current training data record one by one according to the shortest distance principle to obtain multiple first pairing groups, specifically: each first prediction frame is traversed; and during the traversal, the currently traversed first prediction frame is used as the corresponding current prediction frame; and the target frame blood vessel type and the target frame distance of the current prediction frame are used as the corresponding current frame blood vessel type and the current frame distance; and the corresponding current label frame set is composed of one or more first label frames matching the current frame blood vessel type in the current training data record; and whether the current label frame set is empty is identified; if the current label frame set is empty, a preset invalid label frame and the current prediction frame form a corresponding first pairing group; if the current label frame set is not empty, the absolute difference between the current frame distance and the label frame distance of each first label frame in the current label frame set is calculated to obtain the corresponding first absolute difference, and the first label frame corresponding to the smallest first absolute difference and the current prediction frame form a corresponding first pairing group;

[0143] It should be noted that the invalid label box mentioned in the above steps is a pre-set label box data object, and the data format of the invalid label box is consistent with the data format of the first label box, but each data item inside is set in an invalid data mode, and the so-called invalid data mode can be an all-zero mode, a maximum value mode, or a negative value mode, which can be specifically configured based on actual application requirements;

[0144] Step 26-4, bringing all the obtained first pairing groups into a preset second loss function for calculation to obtain corresponding second loss values;

[0145] Among them, the second loss function is the L2 loss function;

[0146] Step 26-5, identifying whether the second loss value satisfies a preset second loss value range; if the second loss value satisfies the second loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, turning to step 26-6, if not, extracting the next second training data record of the second training data set as a new current training data record and returning to step 26-3 to continue training; if the second loss value does not satisfy the second loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 26-3 to continue training at the end of this parameter optimization;

[0147] Among them, the model parameter optimization algorithms include SDG optimization algorithm and Adam optimization algorithm;

[0148] Here, the second loss value range is a preset loss value range;

[0149] Step 26-6, solidifying the model parameters of the MR-T1 model once; and extracting the first second training data record of the second training data set as the corresponding current training data record;

[0150] Step 26-7, inputting the second training T1 weighted image recorded in the current training data into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third predicted T1 weighted image; extracting each stenotic segment target frame marked in the third predicted T1 weighted image as the corresponding second predicted frame; and pairing each second predicted frame with each first label frame recorded in the current training data one by one according to the shortest distance principle to obtain multiple second pairing groups to form a corresponding second pairing group set;

[0151] The second pairing group set includes multiple second pairing groups; the second pairing group includes a second prediction box and a matching first label box;

[0152] Here, the pairing method of pairing each second prediction frame with each first label frame of the current training data record one by one according to the shortest distance principle in the current step 26-7 is similar to the pairing method of pairing each first prediction frame with each first label frame of the current training data record one by one according to the shortest distance principle in the aforementioned step 26-3, and will not be further repeated here;

[0153] Step 26-8, taking each second pairing group of the second pairing group set as the corresponding current pairing group; extracting the sub-image area covered by the second prediction box of the current pairing group on the first training fusion image of the current training data record as the corresponding second training fusion image; inputting the second training fusion image into the MR-HIP model for high-risk plaque identification and outputting the corresponding first prediction type; taking the first label type corresponding to the first label box of the current pairing group in the first label type set of the current training data record as the corresponding second label type; and forming a corresponding first type pair by the first prediction type and the second label type obtained this time; and forming a corresponding first type pair set by all the first type pairs obtained;

[0154] The first type pair corresponds to the second pairing group one by one; the first type pair includes a first prediction type and a corresponding second label type; the first prediction type includes high-risk plaques and non-high-risk plaques;

[0155] Here, it is known that the first label type set is composed of the gold standard for coronary plaque examination, namely, IVUS examination results. Using the first label type set as the training label of the MR-HIP model is to improve the high-risk plaque prediction accuracy of the MR-HIP model based on the IVUS examination results;

[0156] Step 26-9, bringing the first type pair set into a preset third loss function to calculate and obtain a corresponding third loss value;

[0157] Among them, the third loss function is the cross entropy loss function;

[0158] Step 26-10, identifying whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, turning to step 26-11, if not, extracting the next second training data record of the second training data set as the new current training data record and returning to step 26-7 to continue training; if the third loss value does not satisfy the third loss value range, performing a parameter optimization of the MR-HIP model based on the model parameter optimization algorithm, and returning to step 26-8 to continue training at the end of this parameter optimization;

[0159] Here, the third loss value range is a preset loss value range;

[0160] Step 26-11, solidify the model parameters of the MR-HIP model and confirm that the joint training is completed.

[0161] Therefore, through the above steps 1-2, the embodiment of the present invention, on the one hand, completes the optimization of multiple whole-heart coronary artery CATCH sequence scanning schemes, and achieves the purpose of refining the CATCH sequence scanning scheme; on the other hand, it also completes the training of two deep learning models (MR-T1 model, MR-HIP model), which can not only help improve the optimization efficiency of the CATCH sequence scanning scheme, but also improve the prediction accuracy and efficiency of vascular stenosis segments in actual application, as well as the prediction accuracy and efficiency of high-risk plaques; the following step 3 is how to complete the processing process of CATCH sequence imaging analysis based on the MR-T1 model and the MR-HIP model.

[0162] Step 3: Perform a magnetic resonance CATCH sequence scan on any subject based on the optimal scanning scheme and perform visual identification of high-risk coronary plaques based on the imaging results;

[0163] Specifically, it includes: step 31, performing a magnetic resonance CATCH sequence scanning imaging on the current subject based on the optimal scanning scheme to obtain a corresponding fourth T1-weighted image and a second anatomical reference image;

[0164] Step 32, performing image fusion processing on the fourth T1-weighted image and the second anatomical reference image to obtain a corresponding second fused image;

[0165] Step 33, inputting the fourth T1-weighted image into the MR-T1 model to identify the stenotic blood vessel segment to obtain a corresponding fifth T1-weighted image;

[0166] Step 34, when the number of stenosis target frames marked on the fifth T1 weighted image is not 0, extract each stenosis target frame marked on the fifth T1 weighted image as a corresponding first marking frame; perform equivalent marking on each first marking frame on the second anatomical reference image to obtain a corresponding second marking frame; perform equivalent marking on each first marking frame on the second fused image to obtain a corresponding third marking frame; extract a sub-image covered by each third marking frame on the second fused image as a corresponding third fused image; input each third fused image into the MR-HIP model to perform high-risk plaque identification and output a corresponding first high-risk plaque identification result; and perform plaque type marking on each third marking frame using the corresponding first high-risk plaque identification result on the third fused image;

[0167] Among them, the first high-risk plaque identification result includes high-risk plaques and non-high-risk plaques;

[0168] Step 35 , the fifth T1-weighted image, the second anatomical reference image and the second fused image finally obtained are displayed.

[0169] Figure 3 The module structure diagram of a processing device for evaluating and applying a CATCH sequence scheme provided in the second embodiment of the present invention, the device is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment, for example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a CATCH sequence scanning scheme configuration module 201, a scanning scheme evaluation and deep learning model training module 202, and a coronary high-risk plaque visualization processing module 203.

[0170] The CATCH sequence scanning scheme configuration module 201 is used to configure a plurality of whole-heart coronary artery CATCH sequence scanning schemes to form a corresponding first scanning scheme set; the first scanning scheme set includes a plurality of first scanning schemes; each first scanning scheme is a whole-heart coronary artery CATCH sequence scanning scheme.

[0171] The scanning scheme evaluation and deep learning model training module 202 is used to construct a deep learning model for identifying narrowed vascular segments, recorded as the MR-T1 model; and to construct a deep learning model for identifying high-risk plaques, recorded as the MR-HIP model; and to collect first CTA images, first DSA image sets, first T1 weighted images and first plaque detection sets of multiple first subjects to form a corresponding first acquisition data set; and to train the MR-T1 model based on the first acquisition data set; and to perform a magnetic resonance CATCH sequence scanning imaging on each first subject based on each first scanning scheme and to form a corresponding second acquisition data set based on the imaging results; and to evaluate the first scanning scheme set based on the first and second acquisition data sets and the MR-T1 model and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and to jointly train the MR-T1 model and the MR-HIP model based on the optimal scanning scheme and the first and second acquisition data sets.

[0172] The coronary high-risk plaque visualization processing module 203 is used to perform a magnetic resonance CATCH sequence scanning imaging on any subject based on the optimal scanning scheme and to perform visualized coronary high-risk plaque identification based on the imaging results.

[0173] An embodiment of the present invention provides a processing device for evaluating and applying a CATCH sequence scheme, which can execute the method steps in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0174] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the scanning scheme evaluation and deep learning model training modules can be separately established processing elements, or they can be integrated in a chip of the above device. In addition, they can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned module is determined. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in software form.

[0175] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0176] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the above method embodiments are generated. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0177] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device may be a terminal device or a server for implementing the method of the aforementioned embodiment, or may be a terminal device or a server for implementing the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 4As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303. Various instructions may be stored in the memory 302 to complete various processing functions and implement the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention also includes: a power supply 304, a system bus 305 and a communication port 306. The system bus 305 is used to realize the communication connection between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0178] exist Figure 4 The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0179] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (Network Processor, NP), a graphics processing unit (Graphics Processing Unit, GPU), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0180] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the method and processing process provided in the above embodiments.

[0181] An embodiment of the present invention further provides a chip for executing instructions, wherein the chip is used to execute the processing steps described in the aforementioned method embodiment.

[0182] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for evaluating and applying CATCH sequence schemes. As can be seen from the above content, the embodiment of the present invention pre-constructs a deep learning model for identifying stenotic vascular segments, namely, an MR-T1 model, and a deep learning model for identifying high-risk plaques, namely, an MR-HIP model; and configures multiple whole-heart coronary artery CATCH sequence scanning schemes to form a scanning scheme set; and constructs a first acquisition data set by collecting CTA images, DSA image sets, T1-weighted images and plaque detection sets of subjects, and performs a magnetic resonance CATCH sequence scanning imaging on each subject based on each scanning scheme to construct a second acquisition data set; and pre-trains the MR-T1 model based on the first acquisition data set, and then evaluates the best scanning scheme based on the first and second acquisition data sets and the MR-T1 model, and then jointly trains the MR-T1 model and the MR-HIP model based on the best scanning scheme and the first and second acquisition data sets; and after the joint training is completed, magnetic resonance CATCH sequence scanning imaging can be performed on any subject based on the best scanning scheme, and visual coronary high-risk plaque identification processing can be performed based on this imaging result. The embodiments of the present invention not only provide various medical institutions with a coronary HIP plaque prediction mechanism based on CATCH sequence imaging technology, but also provide various medical institutions with an evaluation mechanism for CATCH sequence scanning schemes, and also improve the prediction / evaluation accuracy and corresponding prediction / evaluation efficiency through a deep learning model.

[0183] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0184] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0185] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for evaluating and applying a CATCH sequence scheme, characterized in that: The method comprises: A plurality of whole-heart coronary artery CATCH sequence scanning schemes are configured to form a corresponding first scanning scheme set; the first scanning scheme set includes a plurality of first scanning schemes; each of the first scanning schemes is a whole-heart coronary artery CATCH sequence scanning scheme; Construct a deep learning model for identifying narrowed blood vessel segments, recorded as the MR-T1 model; and construct a deep learning model for identifying high-risk plaques, recorded as the MR-HIP model; and collect first CTA images, first DSA image sets, first T1 weighted images and first plaque detection sets of multiple first subjects to form a corresponding first acquisition data set; and train the MR-T1 model based on the first acquisition data set; and perform a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and form a corresponding second acquisition data set based on the imaging results; and evaluate the first scanning scheme set based on the first and second acquisition data sets and the MR-T1 model, and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and jointly train the MR-T1 model and the MR-HIP model based on the optimal scanning scheme and the first and second acquisition data sets; Based on the optimal scanning scheme, a magnetic resonance CATCH sequence scanning imaging is performed on any subject, and visual identification of high-risk coronary plaques is performed based on the imaging results.

2. The method for evaluating and applying a CATCH sequence scheme according to claim 1, characterized in that: The MR-T1 model is used to identify the stenotic blood vessel segment of the input T1 weighted image and output the output T1 weighted image with the stenotic blood vessel segment detection frame mark; the input and output T1 weighted images are both three-dimensional T1 weighted images; The MR-T1 model is formed by sequentially connecting a first vessel semantic segmentation module, a first vessel tree recognition module, a first stenotic vessel segment recognition module and a first fusion module; The first vascular semantic segmentation module is implemented based on a U-Net network; the first vascular semantic segmentation module is used to perform vascular semantic segmentation processing on the input T1-weighted image to obtain a corresponding first vascular semantic segmentation map; and send the first vascular semantic segmentation map to the first vascular tree recognition module and the first stenotic vascular segment recognition module; the first vascular semantic segmentation map is a three-dimensional binary semantic segmentation map; the voxel point types corresponding to each voxel point on the first vascular semantic segmentation map include vascular semantic voxel points and non-vascular semantic voxel points; The first vascular tree recognition module is implemented based on a convolutional neural network; the first vascular tree recognition module is used to identify one or more connected vascular networks on the first vascular semantic segmentation map and construct a corresponding vascular tree mask map based on each connected vascular network to obtain one or more corresponding first vascular tree mask maps to form a corresponding first vascular tree mask map set and send it to the first stenotic vascular segment recognition module; The first stenotic vessel segment identification module is composed of a first vessel classification unit, a first stenotic vessel segment region identification unit and a first stenotic vessel segment feature fusion unit; The first vessel classification unit is implemented based on a classifier model; the first vessel classification unit is used to classify the coronary vessel type according to the geometric shape features of each of the first vessel tree mask images to obtain a corresponding first classification vector and send it to the first stenotic vessel segment feature fusion unit; the first classification vector includes a plurality of first classification probabilities, each of which corresponds to a coronary vessel type; The first stenotic vascular segment region identification unit is implemented based on the target detection model; the first stenotic vascular segment region identification unit is used to perform corresponding vascular tree semantic segmentation map segmentation processing on the first vascular semantic segmentation map according to each of the first vascular tree mask maps to obtain a corresponding first segmentation semantic segmentation map; and perform stenotic vascular segment region identification processing based on each of the first segmentation semantic segmentation maps, and if one or more stenotic vascular segment regions are identified during the identification process, a corresponding first target detection frame is generated for each identified stenotic vascular segment region to form a corresponding first target detection frame set, and if the stenotic vascular segment region cannot be identified during the processing process, the corresponding first target detection frame set is set to be empty; and all the obtained first target detection frame sets are sent to the first stenotic vascular segment feature fusion unit; the first segmentation semantic segmentation map corresponds to the first vascular tree mask map one-to-one; the first target detection frame set corresponds to the first vascular tree mask map one-to-one; when the first target detection frame set is not empty, it is composed of one or more of the first target detection frames; each of the first target detection frames includes a first target frame center point and a first target frame size; The first stenotic blood vessel segment feature fusion unit is used to extract all the first target detection frames of all the first target detection frame sets to form a corresponding first target detection frame set; and identify whether the first target detection frame set is empty; If the entire set of the first target detection frames is empty, the corresponding first stenosis segment target frame set is set to be empty; if the entire set of the first target detection frames is not empty, each of the first target detection frames in the entire set of the first target detection frames is used as the corresponding current target detection frame, and the first vascular tree mask map corresponding to the current target detection frame is used as the corresponding current vascular tree mask map, and the vascular tree entry point of the current vascular tree mask map is identified to obtain the corresponding current vascular entry point, and the position point corresponding to the first target frame center point of the current target detection frame is marked on the current vascular tree mask map to obtain the corresponding current target frame center point, and the vascular length from the current vascular entry point to the current target frame center point is estimated on the current vascular tree mask map and the estimated length is used as the corresponding first target frame distance, and The coronary vessel type corresponding to the first classification probability with the largest probability value in the first classification vector corresponding to the current vessel tree mask map is used as the corresponding first target frame vessel type, and a corresponding first stenosis segment target frame is formed by the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance corresponding to the current target detection frame, and a corresponding first stenosis segment target frame set is formed by all the obtained first stenosis segment target frames; and the obtained first stenosis segment target frame set is output to the first fusion module; when the first stenosis segment target frame set is not empty, it is composed of one or more first stenosis segment target frames; the first stenosis segment target frame includes the first target frame center point, the first target frame size, the first target frame vessel type and the first target frame distance; The first fusion module is used to identify whether the first stenosis segment target frame set is empty; if the first stenosis segment target frame set is empty, the corresponding output T1 weighted image is set as the input T1 weighted image and output; if the first stenosis segment target frame set is not empty, the corresponding stenosis segment target frame marking processing is performed on the input T1 weighted image based on each of the first stenosis segment target frames in the first stenosis segment target frame set, and the marked T1 weighted image is used as the corresponding output T1 weighted image and output.

3. The method for evaluating and applying a CATCH sequence scheme according to claim 1, characterized in that: The MR-HIP model is used to identify high-risk plaques on the input T1-anatomical fusion image and output the corresponding high-risk plaque identification result; the T1-anatomical fusion image is formed by fusion of a T1-weighted image of a plaque position and its corresponding anatomical reference image; the high-risk plaque identification result includes high-risk plaques and non-high-risk plaques; The MR-H IP model is formed by sequentially connecting a first feature extraction module, a first classification module and a first output module; The first feature extraction module is implemented based on a residual network; the first feature extraction module is used to perform MR signal feature extraction processing on the input T1-anatomical fusion image to obtain a corresponding first feature tensor and send it to the first classification module; The first classification module is implemented based on a binary classifier model; the first classification module is used to perform high-risk plaque classification prediction according to the first feature tensor to obtain a corresponding second classification vector and send it to the first output module; the second classification vector includes a high-risk plaque prediction probability and a non-high-risk plaque prediction probability; The first output module is used to take the predicted probability with the largest probability value in the second classification vector as the corresponding current maximum predicted probability, and to set the corresponding high-risk plaque identification result as a high-risk plaque when the current maximum predicted probability is the high-risk plaque prediction probability, and to set the corresponding high-risk plaque identification result as a non-high-risk plaque when the current maximum predicted probability is the non-high-risk plaque prediction probability; and output the obtained high-risk plaque identification result.

4. The method for evaluating and applying a CATCH sequence scheme according to claim 1, characterized in that: The first acquisition data set includes a plurality of first subject records; the first subject records correspond one-to-one to the first subject; the first subject records include the first CTA image, the first DSA image set, the first T1-weighted image and the first plaque detection set; The first CTA image is a corresponding three-dimensional whole-heart coronary CTA image of the first subject; one or more first label detection frames are pre-marked on the first CTA image; each of the first label detection frames corresponds to a stenotic blood vessel segment; each of the first label detection frames includes a first label frame center point, a first label frame size, a first label frame blood vessel type and a first label frame distance; the first label frame blood vessel type is the blood vessel type of the corresponding coronary blood vessel where the first label detection frame is located; the first label frame distance is the blood vessel length between the corresponding first label frame center point and the blood vessel entry point of the coronary blood vessel where the first label detection frame is located; The first DSA image set includes multiple first DSA images; each of the first DSA images corresponds to a coronary blood vessel type; the first DSA image is a three-dimensional DSA image of a type of coronary blood vessels of the corresponding first subject; one or more second label detection frames are pre-marked on the first DSA image; each of the second label detection frames corresponds to a stenotic blood vessel segment; each of the second label detection frames includes a second label frame center point, a second label frame size, a second label frame blood vessel type, and a second label frame distance; the second label frame blood vessel type is the blood vessel type of the corresponding coronary blood vessel where the second label detection frame is located; the second label frame distance is the blood vessel length between the corresponding second label frame center point and the blood vessel entry point of the coronary blood vessel where the second label detection frame is located; the second label detection frame corresponds to the first label detection frame one by one; The first T1-weighted image is a three-dimensional whole-heart coronary T1-weighted image of the corresponding first subject, and the first T1-weighted image is generated based on a non-CATCH sequence; one or more third label detection frames are pre-marked on the first T1-weighted image; each of the third label detection frames corresponds to a stenotic vascular segment; each of the third label detection frames includes a third label frame center point, a third label frame size, a third label frame vessel type, and a third label frame distance; the third label frame vessel type is the vessel type of the corresponding coronary vessel where the third label detection frame is located; the third label frame distance is the vessel length between the corresponding third label frame center point and the vessel entry point of the coronary vessel where the third label detection frame is located; the third label detection frame corresponds to the second label detection frame one by one; The first plaque detection set includes a plurality of first plaque detection results; the first plaque detection results include high-risk plaques and non-high-risk plaques; the first plaque detection results correspond one-to-one to the second label detection frame; the first plaque detection results are plaque detection results obtained by I VUS examination; The second acquisition data set includes multiple first subject image sets; the first subject image sets correspond one-to-one to the first subject; the first subject image set includes multiple first plan image sets; the first plan image set corresponds one-to-one to the first scanning plan; the first plan image set includes a second T1-weighted image and a first anatomical reference image.

5. The method for evaluating and applying a CATCH sequence scheme according to claim 4, characterized in that: The training of the MR-T1 model based on the first acquired data set specifically includes: Step 51, traverse the first subject record of the first acquisition data set; and during the traversal, use the first subject record currently traversed as the corresponding current subject record; and use the first T1 weighted image of the current subject record as the corresponding first label T1 weighted image, and delete all the third label detection frames marked on the first label T1 weighted image to obtain the corresponding first training T1 weighted image, and form a corresponding first training data record with the obtained first training T1 weighted image and the first label T1 weighted image; and at the end of the traversal, all the obtained first training data records form the corresponding first training data set; Step 52, extracting the first first training data record of the first training data set as the corresponding current training data record; Step 53, inputting the T1-weighted image recorded in the current training data into the MR-T1 model to identify the stenotic blood vessel segment to obtain a corresponding first predicted T1-weighted image; Step 54, bringing the first predicted T1 weighted image and the first label T1 weighted image of the current training data record into a preset first loss function to calculate and obtain a corresponding first loss value; the first loss function uses a cross entropy loss function by default; Step 55, identifying whether the first loss value satisfies a preset first loss value range; if the first loss value satisfies the first loss value range, identifying whether the current training data record is the last first training data record of the first training data set, if so, going to step 56, if not, extracting the next first training data record of the first training data set as the new current training data record and returning to step 53 to continue training; if the first loss value does not satisfy the first loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 53 to continue training at the end of this parameter optimization; the model parameter optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm; Step 56, solidify the model parameters of the MR-T1 model and confirm that the model training is completed.

6. The method for evaluating and applying a CATCH sequence scheme according to claim 4, characterized in that: The step of performing a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and forming a corresponding second acquisition data set based on the imaging results specifically includes: Based on each of the first scanning schemes, a magnetic resonance CATCH sequence scanning imaging is performed on each of the first subjects to obtain the corresponding second T1-weighted image and the first anatomical reference image to form a corresponding first scheme image set; and all the first scheme image sets of each of the first subjects form a corresponding first subject image set; and all the obtained first subject image sets form a corresponding second acquisition data set.

7. The method for evaluating and applying a CATCH sequence scheme according to claim 4, characterized in that: The step of evaluating the first scanning scheme set according to the first and second acquisition data sets and the MR-T1 model and taking the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme specifically includes: Record the total number of the first subjects as a first total number M, and record the total number of the first scanning schemes in the first scanning scheme set as a second total number N; and extract the first CTA images, the first DSA image sets, and the first subject image sets corresponding to each of the first subjects from the first and second acquisition data sets to form a corresponding first subject data set; Traversing all the first subject data sets; and during the traversal, taking the currently traversed first subject data set as the corresponding current subject data set; and taking the first CTA image, the first DSA image set and the first subject image set of the current subject data set as the corresponding current CTA image, the current DSA image set and the current subject image set; and performing vector conversion on each of the first label detection frames of the current CTA image to obtain a corresponding first label detection frame vector, and forming a corresponding first label frame vector set from all the obtained first label detection frame vectors; and performing vector conversion on each of the second label detection frames of the current DSA image set to obtain a corresponding second label detection frame vector, And all the obtained second label detection frame vectors form a corresponding second label frame vector set; and the second T1 weighted images of each of the first scheme image sets of the current subject image set are input into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third T1 weighted image, and each stenotic segment target frame marked on the current third T1 weighted image is vector-converted to obtain the corresponding first stenotic segment target frame vector, and all the first stenotic segment target frame vectors corresponding to the current third T1 weighted image form a corresponding first prediction frame vector set; and the matching degree of each of the first prediction frame vector sets with the corresponding first and second label frame vector sets is evaluated to obtain the corresponding first and second evaluation values ​​A. i,j , B i,j , and according to the first and second evaluation values ​​A i,j , B i,j Calculate the corresponding third evaluation value C i,j =α×A i,j +β×B i,j , i is the subject index number of each of the first subjects, j is the scheme index number of each of the first scanning schemes, 1≤i≤M, 1≤j≤N, α and β are two preset weighting parameters, 0<α<β<1, α+β=1; and the N first scanning schemes corresponding to the current subject data set are evaluated according to the third evaluation value C i,j Sort them from high to low to get the corresponding first subject solution sequence; After the traversal of all the first subject data sets is completed, all the first scanning schemes are traversed; and during the traversal, the first scanning scheme currently traversed is used as the corresponding current scanning scheme; and in the obtained M first subject scheme sequences, the number of times the current scanning scheme appears at the kth index position of the sequence is counted to obtain the corresponding first position number D k , 1≤k≤N, 0≤D k ≤M; and the N first position times D corresponding to the current scanning scheme k Calculate the corresponding After traversing all the first scanning schemes, the first scanning scheme with the highest first scheme evaluation score is used as the corresponding optimal scanning scheme.

8. The method for evaluating and applying a CATCH sequence scheme according to claim 4, characterized in that: The jointly training the MR-T1 model and the MR-H IP model according to the optimal scanning scheme and the first and second acquisition data sets specifically includes: Step 81, traverse all the first subjects; and during the traversal, take the first subject currently traversed as the corresponding current subject; and take the first DSA image set and the first plaque detection set corresponding to the current subject in the first acquisition data set as the corresponding current DSA image set and current plaque detection set; and take the first scheme image set corresponding to the optimal scanning scheme in the first subject image set corresponding to the current subject in the second acquisition data set as the corresponding current scheme image set; and take each of the second label detection frames in the current DSA image set as the corresponding first label frame to form a corresponding first label frame set; and take each of the first label detection frames in the current plaque detection set as the corresponding first label frame set. The first plaque detection result corresponding to the frame is used as the corresponding first label type to form a corresponding first label type set; and the second T1 weighted image of the current scheme image set and the first anatomical reference image are subjected to image fusion processing to obtain a corresponding first fused image; and the second T1 weighted image of the current scheme image set and the corresponding first fused image are used as the corresponding second training T1 weighted image and the first training fused image; and the obtained second training T1 weighted image, the first training fused image, the first label frame set and the first label type set form a corresponding second training data record; and at the end of the traversal, all the obtained second training data records form a corresponding second training data set; Step 82, extracting the first second training data record of the second training data set as the corresponding current training data record; Step 83, inputting the second training T1 weighted image recorded in the current training data into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding second predicted T1 weighted image; extracting each stenotic segment target frame marked in the second predicted T1 weighted image as the corresponding first predicted frame; and pairing each first predicted frame with each first label frame recorded in the current training data one by one according to the shortest distance principle to obtain a plurality of first paired groups; the first paired group includes one first predicted frame and one matched first label frame; Step 84, bringing all the first pairing groups obtained into a preset second loss function to calculate and obtain corresponding second loss values; the second loss function is an L2 loss function; Step 85, identifying whether the second loss value satisfies a preset second loss value range; if the second loss value satisfies the second loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, going to step 86, if not, extracting the next second training data record of the second training data set as the new current training data record and returning to step 83 to continue training; if the second loss value does not satisfy the second loss value range, performing a parameter optimization on the MR-T1 model based on a preset model parameter optimization algorithm, and returning to step 83 to continue training at the end of this parameter optimization; the model parameter optimization algorithm includes an SDG optimization algorithm and an Adam optimization algorithm; Step 86, solidifying the model parameters of the MR-T1 model once; and extracting the first second training data record of the second training data set as the corresponding current training data record; Step 87, input the second training T1 weighted image of the current training data record into the MR-T1 model to identify the stenotic vascular segment to obtain the corresponding third predicted T1 weighted image; and extract each stenotic segment target frame marked in the third predicted T1 weighted image as the corresponding second predicted frame; and according to the shortest distance principle, each second predicted frame is matched with each first label frame of the current training data record one by one to obtain a plurality of second pairing groups to form a corresponding second pairing group set; the second pairing group set includes a plurality of second pairing groups; the second pairing group includes one second predicted frame and one matched first label frame; Step 88, taking each of the second pairing groups of the second pairing group set as the corresponding current pairing group; extracting the sub-image area covered by the second prediction box of the current pairing group on the first training fusion image of the current training data record as the corresponding second training fusion image; inputting the second training fusion image into the MR-H IP model to identify high-risk plaques and output the corresponding first prediction type; taking the first label type corresponding to the first label box of the current pairing group in the first label type set of the current training data record as the corresponding second label type; and forming a corresponding first type pair with the first prediction type and the second label type obtained this time; and forming a corresponding first type pair set with all the first type pairs obtained; the first type pair corresponds to the second pairing group one by one; the first type pair includes one first prediction type and one corresponding second label type; the first prediction type includes high-risk plaques and non-high-risk plaques; Step 89, bringing the first type pair set into a preset third loss function to calculate a corresponding third loss value; the third loss function is a cross entropy loss function; Step 90, identifying whether the third loss value satisfies a preset third loss value range; if the third loss value satisfies the third loss value range, identifying whether the current training data record is the last second training data record of the second training data set, if so, turning to step 91, if not, extracting the next second training data record of the second training data set as the new current training data record and returning to step 87 to continue training; if the third loss value does not satisfy the third loss value range, performing a parameter optimization of the MR-H IP model based on the model parameter optimization algorithm, and returning to step 88 to continue training at the end of this parameter optimization; Step 91, solidify the model parameters of the MR-HIP model and confirm that the joint training is completed.

9. The method for evaluating and applying a CATCH sequence scheme according to claim 4, characterized in that: The method of performing a magnetic resonance CATCH sequence scanning imaging on any subject based on the optimal scanning scheme and visually identifying high-risk coronary plaques based on the imaging results specifically includes: Performing a magnetic resonance CATCH sequence scanning imaging on the current subject based on the optimal scanning scheme to obtain a corresponding fourth T1-weighted image and a second anatomical reference image; and performing image fusion processing on the fourth T1-weighted image and the second anatomical reference image to obtain a corresponding second fused image; and inputting the fourth T1 weighted image into the MR-T1 model to identify the stenotic blood vessel segment to obtain a corresponding fifth T1 weighted image; When the number of stenosis target frames marked on the fifth T1 weighted image is not 0, extract each stenosis target frame marked on the fifth T1 weighted image as a corresponding first marking frame; perform equivalent marking on each of the first marking frames on the second anatomical reference image to obtain a corresponding second marking frame; perform equivalent marking on each of the first marking frames on the second fused image to obtain a corresponding third marking frame; extract a sub-image covered by each of the third marking frames on the second fused image as a corresponding third fused image; input each of the third fused images into the MR-HIP model to perform high-risk plaque identification and output a corresponding first high-risk plaque identification result; and perform plaque type marking on each of the third marking frames using the corresponding first high-risk plaque identification result on the third fused image; The fifth T1-weighted image, the second anatomical reference image and the second fused image finally obtained are displayed.

10. A device for executing the processing method for evaluating and applying a CATCH sequence scheme according to any one of claims 1 to 9, characterized in that: The device comprises: a CATCH sequence scanning scheme configuration module, a scanning scheme evaluation and deep learning model training module, and a coronary high-risk plaque visualization processing module; The CATCH sequence scanning scheme configuration module is used to configure a plurality of whole heart coronary artery CATCH sequence scanning schemes to form a corresponding first scanning scheme set; the first scanning scheme set includes a plurality of first scanning schemes; each of the first scanning schemes is a whole heart coronary artery CATCH sequence scanning scheme; The scanning scheme evaluation and deep learning model training module is used to construct a deep learning model for identifying narrowed blood vessel segments, which is recorded as an MR-T1 model; and to construct a deep learning model for identifying high-risk plaques, which is recorded as an MR-HIP model; and to collect first CTA images, first DSA image sets, first T1 weighted images and first plaque detection sets of multiple first subjects to form a corresponding first acquisition data set; and to train the MR-T1 model based on the first acquisition data set; and to perform a magnetic resonance CATCH sequence scanning imaging on each of the first subjects based on each of the first scanning schemes and to form a corresponding second acquisition data set based on the imaging results; and to evaluate the first scanning scheme set based on the first and second acquisition data sets and the MR-T1 model and use the first scanning scheme with the highest evaluation score as the corresponding optimal scanning scheme; and to jointly train the MR-T1 model and the MR-HIP model based on the optimal scanning scheme and the first and second acquisition data sets; The coronary high-risk plaque visualization processing module is used to perform a magnetic resonance CATCH sequence scanning imaging on any subject based on the optimal scanning scheme and to perform visualized coronary high-risk plaque identification based on the imaging results.

11. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 9; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 9.

Images