Method, device and computer storage medium for predicting blood flow reserve fraction based on optical coherence tomography

By using multi-task learning model and fluid dynamics model in optical coherence tomography technology, fully automatic blood flow reserve score prediction is achieved, solving the problems of limited application scope and complex operation of the existing FFR detection method, and improving detection Accuracy and automation reduce patient pain and medical costs.

CN114463267BActive Publication Date: 2024-11-19SHENZHEN VIVOLIGHT MEDICAL DEVICE & TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111641922.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-29
Publication Date
2024-11-19
Estimated Expiration
2041-12-29

AI Technical Summary

Technical Problem

The existing blood flow reserve fraction (FFR) detection method requires the use of adenosine to induce maximum congestion, and the scope of application is limited. The insertion pressure guide wire is complicated and time-consuming, which leads to patient pain and high medical costs, and the doctor's subjective error is large, which reduces the Detection accuracy.

Method used

Using a multi-task learning model based on optical coherence tomography (OCT) imaging to identify lumens and bifurcations, combined with a fluid dynamic model, fully automatic FFR prediction is achieved, and the accuracy of healthy blood vessel recognition is improved through triple loss function and the doctor's subjective Intervention.

Benefits of technology

Improve the accuracy and automation of FFR prediction, reduce patient pain and medical costs, reduce diagnosis time, enhance the generalization ability of the model, apply to more patients, and provide more comprehensive vascular information to support personalized treatment plans. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114463267B_ABST
    Figure CN114463267B_ABST
Patent Text Reader

Abstract

The present invention discloses a blood flow reserve fraction prediction method, device and computer storage medium based on optical coherence tomography. By establishing a multi-task learning model, an end-to-end network structure is used to simultaneously identify the lumen and bifurcation in the OCT image. At the same time, considering that the lumen and bifurcation have certain structural characteristics and spatial position relationship, a multi-task learning model based on spatial structure constraints is constructed to achieve accurate recognition of the lumen and bifurcation, and solve the problems of structural information loss and category imbalance existing in the traditional network model, thereby improving the generalization ability of the model. In order to realize the automatic selection of the reference frame and solve the visual similarity problem between the OCT healthy blood vessels and the unhealthy blood vessels, an OCT healthy blood vessel classification network based on triplet loss is proposed. Then, by using the fluid dynamics program, combined with the lumen contour area, bifurcation and reference frame, a multi-parameter-based FFR prediction technology is proposed to accurately measure the FFR.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of medical technology, and in particular relates to a method and a device for predicting blood flow reserve fraction of optical coherence tomography imaging. Background Art

[0002] Optical Coherence Tomography (OCT) is an intracavitary imaging technology that uses the basic principle of weak coherent light interferometer to record the reflected light of biological tissues at different depths and then reconstruct the reflected light signal by computer to form an image of biological tissue structure that can be easily identified. Fast imaging, high resolution, accurate identification of plaques, and accurate evaluation of stent implantation effects have made OCT gradually become an important imaging technology in the diagnosis of coronary heart disease and PCI interventional treatment. Compared with intravascular ultrasound (IVUS), the resolution of OCT images is as high as 10 to 20 μm (IVUS resolution is about 100 μm), which can not only accurately distinguish the degree of anatomical stenosis, but also clearly see the type and shape of plaques in the vascular wall, so as to more accurately evaluate lesions.

[0003] Fractional Flow Reserve (FFR) refers to the ratio of the maximum blood flow provided by the coronary artery to the myocardial area it supplies when the coronary artery has a lesion to the maximum blood flow provided to the area in theory without the lesion. Long-term scientific research and clinical trials have shown that FFR can be used as the "gold standard" for the functional evaluation of coronary artery stenosis. FFR can accurately determine the relationship between coronary artery stenosis and myocardial ischemia, correctly guide coronary intervention, and evaluate the results and prognosis of coronary intervention. It is generally believed that interventional treatment is required when the FFR value of the diseased vessel is lower than 0.75, but FFR measurement has the following defects. For example, in order to make the patient's blood vessels reach the maximum congestion state, adenosine needs to be injected. For patients with renal dysfunction and patients who have adverse reactions to injected drugs, the measurement cannot be performed, and the patient's health cannot be guaranteed. In addition, the reaction time required for inserting the pressure guidewire is long, which will cause more pain to the patient; at the same time, the cost of in vivo detection of the pressure guidewire is generally high, and the medical expenses are high. Therefore, the prevalence of FFR detection is low at present.

[0004] In order to solve the above-mentioned method of detecting blood flow reserve fraction, it is necessary to use vasodilator drugs such as adenosine triphosphate to induce the maximum hyperemia reaction. Patients who are not suitable for related drugs cannot be tested, which cannot guarantee the health of patients. It is an invasive operation. In addition, due to the insertion of pressure guide wire for measurement, it is not only troublesome for doctors to operate, but also time-consuming and painful for patients. With the development of technology, many researchers have developed some imaging methods to assist in the calculation of FFR. The method of calculating blood flow reserve fraction based on intracavitary imaging proposed by Hangzhou Pulse Flow Company is one of them.

[0005] CN111134651, publication date: 2020-05-12, describes a method, device, system and computer storage medium for calculating blood flow reserve fraction based on intracavitary imaging, wherein the method for obtaining the blood flow reserve fraction includes step S1, obtaining image data related to coronary arteries, and constructing a corresponding three-dimensional vascular model by processing the image data; wherein the image data related to the coronary arteries include: intracavitary image data and coronary angiography image data; step S2, calculating the blood flow equation according to the three-dimensional vascular model and the fluid dynamics method to obtain the hemodynamic parameter distribution of the coronary artery in the area expressed by the three-dimensional vascular model; step S3, calculating the blood flow reserve fraction according to the hemodynamic parameters obtained in step S2, and accurately calculating the patient's coronary artery FFR.

[0006] However, in actual practice, the existing technology is used to calculate the blood flow reserve fraction. Although it effectively identifies the lumen area, the selection of the reference frame (normal / healthy lumen) usually needs to be done manually, which not only takes more time, but also requires doctors to have relatively rich clinical experience, and is prone to introduce subjective errors, thereby reducing the prediction accuracy of FFR. Summary of the invention

[0007] The technical problem to be solved by the present invention is that, by establishing a multi-task learning model, an end-to-end network structure is used to simultaneously identify the lumen and bifurcation in the OCT image; at the same time, considering that the lumen and bifurcation have certain structural characteristics and spatial position relationships, a multi-task learning model based on spatial structural constraints is constructed to achieve accurate identification of the lumen and bifurcation, while solving the problems of missing structural information and category imbalance existing in the traditional network model, and improving the generalization ability of the model; in order to realize the automatic selection of the reference frame and solve the visual similarity problem between the OCT healthy blood vessels and unhealthy blood vessels, an OCT healthy blood vessel classification network based on triplet loss is proposed; then, by using the fluid dynamics program, combined with the lumen contour area, bifurcation and reference frame, it is proposed to construct a multi-parameter-based FFR prediction technology to accurately calculate the FFR.

[0008] In order to achieve the above technical tasks, the present invention adopts the following technical solutions:

[0009] A blood flow reserve fraction prediction device based on optical coherence tomography images includes a probe interface module, a display module, a detection module, a signal processing module, and a data calculation module, wherein the detection module, the probe interface module, the signal processing module, the data calculation module, and the display module are connected in sequence;

[0010] The probe interface module is used for transmitting and receiving optical signals and links the detection module to withdraw;

[0011] The detection module includes a catheter 1, an elastic member 2, an optical fiber 3, a probe 4, a guide wire 5, a transparent outer tube sleeve 6, and a flushing liquid outlet 7. The detection module is used to transmit the optical signal emitted by the probe interface module, and transmit the optical signal returned from the tissue back to the probe interface module. The above signal is sent to the signal processing module through the probe interface module. The detection module is mainly used to transmit the optical signal emitted by the probe interface module, and transmit the optical signal returned from the tissue back to the probe interface module. The above signal is sent to the signal processing module through the probe interface module.

[0012] The signal processing module is used to convert the optical signal input by the probe interface module into image information; and perform data calibration and image enhancement on the image information;

[0013] The data acquisition unit superimposes and interferes the light signal reflected from the tissue at the same time with the light signal reflected from the reference reflector. The generated interference signal is detected by the photodetector, and the detector converts the interfered light signal into an electrical signal, thereby realizing data acquisition. Since the electrical signal output by the detector is relatively weak, a signal amplifier is required to amplify it, and the amplified electrical signal is sent to the data processing unit;

[0014] The data processing unit converts the electrical signal input by the data acquisition unit to obtain image information, and performs data calibration and image enhancement on the image information;

[0015] The data calculation module builds a joint lumen and bifurcation recognition framework based on OCT images and multi-task learning mechanism. By processing the signal input by the detection module, it obtains the edge, morphology, and structural space information of the vascular lumen, obtains the vascular lumen boundary, calculates the vascular lumen area, and obtains the bifurcation position;

[0016] By processing the image input by the data processing unit, a healthy blood vessel recognition model based on triples is constructed. By increasing the difference between classes and the similarity within classes, the healthy blood vessels in the pullback OCT image are identified, and the reference frame is selected in combination with the lumen area and bifurcation position.

[0017] The fluid dynamics model is further used to calculate the blood flow and flow resistance in the blood vessels in combination with the vascular lumen area and the reference frame area. The intravascular FFR value is obtained by combining the actual measured systolic and diastolic blood pressure data, thereby realizing the fully automatic calculation of FFR.

[0018] The display module is used to present the processing results obtained by the data calculation module on the display.

[0019] The present invention also discloses a method for predicting blood flow reserve fraction based on optical coherence tomography images, the method comprising the following steps:

[0020] Step 1: The optical signal returned by the near-infrared light from the tissue is superimposed and interfered with the optical signal reflected by the reference reflector. The generated interference signal is detected by the photodetector, and the detector converts the interfered optical signal into an electrical signal. Since the electrical signal output by the detector is relatively weak, a signal amplifier is required to amplify it, and the amplified electrical signal is sent to the signal processing module to complete signal data acquisition;

[0021] Step 2, performing signal conversion, image calibration, and image enhancement on the collected signal data to complete image data collection;

[0022] Step 3, construct a multi-task learning model for lumen and bifurcation recognition;

[0023] Step 4, establishing the prior constraints of the lumen and bifurcation contours;

[0024] Step 5, combining the contour prior constraints in step 4, executing the multi-task learning model learning task of lumen and bifurcation recognition, obtaining the edge, morphology, and structural space information of the vascular lumen, calculating the vascular lumen area, and obtaining the bifurcation position;

[0025] Step 6: construct a healthy blood vessel recognition model based on triples, identify healthy blood vessels in the pullback OCT image by increasing the differences between classes and the similarities within classes, and select the reference frame based on the lumen area and bifurcation position;

[0026] Step 7, using the fluid dynamics program, combined with the vascular lumen area and the reference frame area, to construct a fully automatic FFR prediction model to calculate the intravascular FFR value.

[0027] The present invention may also be applied to a storage medium made into any computer-readable instruction, which, when executed by one or more processors, enables the one or more processors to execute the overall optical coherence tomography image-based blood flow reserve fraction prediction method of the present invention.

[0028] Compared with the prior art, the present invention has the following positive and beneficial effects

[0029] (1) Optical coherence tomography has the remarkable feature of high resolution. Compared with traditional angiography imaging technology, its resolution can reach several microns, which is close to the level of tissue pathology sections and does not cause any damage to the tissue. OCT-FFR collects the necessary intravascular characteristic data for FFR prediction, and its predicted FFR value is more accurate and has a higher reference value for clinical diagnosis and treatment.

[0030] (2) Utilize the automatic reference frame recognition technology to realize fully automatic OCT-FFR technology, reduce the diagnosis time, reduce the subjective differences of doctors in FFR calculation images, and provide effective auxiliary diagnosis for junior doctors.

[0031] (3) Combining OCT with FFR, more accurate patient-specific coronary images can be obtained from OCT images, making virtual FFR prediction results more reliable and accurate. It can not only perform imaging evaluation of the degree of stenosis of myocardial vessels, but also analyze the impact of stenosis on distal blood flow and perform functional evaluation of coronary artery stenosis lesions. OCT-FFR is more accurate and detailed than conventional FFR. It can not only predict the FFR value of diseased vessels, but also accurately evaluate coronary atherosclerotic plaques through intravascular three-dimensional imaging, providing doctors with more comprehensive information on diseased vessels, thereby providing more theoretical references for formulating and selecting personalized treatment plans.

[0032] (4) The prediction of blood flow reserve fraction based on intravascular optical coherence tomography images is a non-invasive technology. Compared with the traditional pressure guide wire measurement of FFR, it does not require the introduction of adenosine to keep the patient's blood vessels in a state of maximum congestion, thereby alleviating the patient's pain, significantly reducing the diagnosis time, and effectively improving medical efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the structure of the device for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention.

[0034] Figure 2 It is a structural schematic diagram of a data processing unit in a signal processing module of a blood flow reserve fraction prediction device based on optical coherence tomography images of the present invention.

[0035] Figure 3 Another structural schematic diagram of the data processing unit in the signal processing module of the device for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention.

[0036] Figure 4 It is a structural schematic diagram of a data calculation module of a blood flow reserve fraction prediction device based on optical coherence tomography images of the present invention.

[0037] Figure 5The flowchart of the method for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention is shown in FIG.

[0038] Figure 6 Schematic diagram of the network structure of the multi-task learning model for identifying lumens, bifurcations and stents of the present invention.

[0039] Figure 7 Schematic diagram of the triplet loss principle in the healthy blood vessel recognition model of the present invention

[0040] Figure 8 This is a diagram of the initial operation interface of the device for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention.

[0041] Fig. 9 The method for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention completes image calibration and image enhancement to obtain an OCT image.

[0042] Fig.10 This is a schematic diagram of the lumen and bifurcation identification effect in the blood flow reserve fraction prediction method based on optical coherence tomography images of the present invention.

[0043] Fig.11 The present invention is a schematic diagram of the effect of bifurcation position in the blood flow reserve fraction prediction method based on optical coherence tomography images.

[0044] Fig.12 The figure is a schematic diagram of the effect of automatic reference frame selection in the method for predicting blood flow reserve fraction based on optical coherence tomography images of the present invention.

[0045] Fig.13 The invention is a method for predicting blood flow reserve fraction based on optical coherence tomography images, and a schematic diagram of the FFR prediction results.

[0046] Figure 1 The symbols in the figure represent the following meanings respectively: 1. catheter; 2. elastic member; 3. optical fiber; 4. probe; 5. guide wire; 6. transparent outer tube sheath; 7. flushing fluid outlet; 8. blood vessel wall.

[0047] The specific contents of the present invention are further described in detail below in conjunction with the accompanying drawings and specific implementation methods. DETAILED DESCRIPTION

[0048] See also Figure 1 The blood flow reserve fraction prediction device based on optical coherence tomography imaging of the present invention comprises a probe interface module, a display module, a detection module, a signal processing module and a data calculation module, wherein the detection module, the probe interface module, the signal processing module, the data calculation module and the display module are connected in sequence;

[0049] The probe interface module is used for transmitting and receiving optical signals and links the detection module to withdraw;

[0050] The detection module includes a catheter 1, an elastic member 2, an optical fiber 3, a probe 4, a guide wire 5, a transparent outer tube sleeve 6, and a flushing liquid outlet 7. It is used to transmit the optical signal emitted by the probe interface module and transmit the optical signal returned from the tissue back to the probe interface module. The above signal is sent to the signal processing module through the probe interface module.

[0051] The signal processing module is used to convert the optical signal input by the probe interface module into image information; and perform data calibration and image enhancement on the image information;

[0052] The data calculation module builds a joint lumen and bifurcation recognition framework based on OCT images and multi-task learning mechanism. By processing the signal input by the detection module, it obtains the edge, morphology, and structural space information of the vascular lumen, obtains the vascular lumen boundary, calculates the vascular lumen area, and obtains the bifurcation position;

[0053] By processing the signal input by the signal processing module, a healthy blood vessel recognition model based on triples is constructed. By increasing the difference between classes and the similarity within classes, healthy blood vessels in the pullback OCT image are identified, and the reference frame is selected in combination with the lumen area and bifurcation position.

[0054] The fluid dynamics model is further used to calculate the blood flow and pressure loss in the blood vessels in combination with the vascular lumen area and the reference frame area. The intravascular FFR value is obtained by combining the actual measured systolic and diastolic blood pressure data, thereby realizing the fully automatic calculation of FFR.

[0055] The display module is used to present the processing results obtained by the data calculation module on the display.

[0056] See also Figure 2-Figure 3 The signal processing module at least comprises a data acquisition unit and a data processing unit connected in sequence. The data acquisition unit performs data acquisition on the signal acquired by the detection module through the detection interface module.

[0057] The data acquisition unit superimposes and interferes the light signal reflected from the tissue at the same time with the light signal reflected from the reference reflector. The generated interference signal is detected by the photodetector, and the detector converts the interfered light signal into an electrical signal, thereby realizing data acquisition. Since the electrical signal output by the detector is relatively weak, a signal amplifier is required to amplify it, and the amplified electrical signal is sent to the data processing unit;

[0058] The data processing unit at least includes a signal conversion module, an image calibration module and an image enhancement module. The signal conversion module is used to convert the electrical signal output by the data acquisition unit into image information; the image calibration module is used to perform data calibration on the image obtained by the signal conversion module or the image cropping module respectively; the image enhancement module is used to perform image enhancement on the image reconstructed by the image calibration module.

[0059] The data calculation module unit at least includes lumen and bifurcation modules connected in sequence; lumen area calculation module, reference frame automatic identification module, and FFR fully automatic prediction module; the lumen and bifurcation identification module is used to extract contour information from the vascular lumen and bifurcation information; the lumen area calculation module is used to obtain the vascular lumen boundary and calculate the vascular lumen area based on the vascular lumen edge, morphology, and structural space information; the reference frame automatic calculation module uses the triplet loss function to build a healthy lumen recognition model to increase the correlation within the class and the difference between classes, and then automatically selects the reference frame position in combination with the bifurcation position; the FFR fully automatic prediction module uses the fluid dynamics model, combined with the vascular lumen area and the reference frame area, to calculate the blood flow and flow resistance in the blood vessel, and combines the actually measured systolic and diastolic pressure data to obtain the intravascular FFR value.

[0060] See also Figure 3 This is another structural schematic diagram of the blood flow reserve fraction prediction device based on optical coherence tomography images of the present invention, wherein the data processing unit of the signal processing module, the data preprocessing unit also includes an image cropping module, which is used to segment the image information obtained by the signal conversion module and extract specific image areas.

[0061] A cleaning module is also arranged between the probe interface module and the detection module.

[0062] See also Figure 1 In the device for predicting blood flow reserve fraction based on optical coherence tomography imaging of the present invention, the detection module also includes a catheter 1, an elastic member 2, and an optical fiber 3. The catheter 1 includes an inner part and an outer part of the body. A cleaning module is arranged on the outer part of the body of the catheter 1. The detection module is connected with the probe interface module through the outer part of the body of the catheter 1. A probe 4, an optical fiber 3, and an elastic member 2 are arranged inside the body of the catheter 1. The probe 4 is connected with the probe interface module through the optical fiber 3, and the elastic member 2 is arranged on the optical fiber 3. The elastic member 2 can be any member that can produce elastic deformation, such as a spring tube, so that under the deformation constraint condition of the elastic member 2, the probe 4 can realize reciprocating motion along the axial direction of the catheter to obtain data information related to the lumen of the blood vessel and the external elastic membrane.

[0063] A flushing liquid outlet 7 is provided at one end of the body of the detection module catheter 1 away from the body of the catheter 1. The flushing liquid is injected into the catheter 1 from the cleaning module and discharged from the flushing liquid outlet 7.

[0064] A transparent outer sleeve 6 is also sleeved on the outside of the inner body of the catheter 1 , and the depth of the transparent outer sleeve 6 in the inner body of the catheter 1 does not exceed the axial position of the probe 4 in the inner body of the catheter 1 .

[0065] The transparent outer sleeve 6 is arranged outside the catheter 1 in the body, and its main purpose is to help fix the position of the catheter 1 in the body.

[0066] See also Figure 5 The present invention provides a method for predicting blood flow reserve fraction based on optical coherence tomography images, the method comprising the following steps:

[0067] Step 1: The optical signal returned by the near-infrared light from the tissue is superimposed and interfered with the optical signal reflected by the reference reflector. The generated interference signal is detected by the photodetector, and the detector converts the interfered optical signal into an electrical signal. Since the electrical signal output by the detector is relatively weak, a signal amplifier is required to amplify it, and the amplified electrical signal is sent to the signal processing module to complete signal data acquisition;

[0068] Step 2, performing signal conversion, image calibration, and image enhancement on the collected signal data to complete image data collection;

[0069] Step 3, construct a multi-task learning model for lumen and bifurcation recognition;

[0070] Step 4, establishing the prior constraints of the lumen and bifurcation contours;

[0071] Step 5, combining the contour prior constraints in step 4, executing the multi-task learning model learning task of lumen and bifurcation recognition, obtaining the edge, morphology, and structural space information of the vascular lumen, calculating the vascular lumen area, and obtaining the bifurcation position;

[0072] Step 6: construct a healthy blood vessel recognition model based on triples, identify healthy blood vessels in the pullback OCT image by increasing the differences between classes and the similarities within classes, and select the reference frame based on the lumen area and bifurcation position;

[0073] Step 7, using the fluid dynamics program, combined with the vascular lumen area and the reference frame area, to construct a fully automatic FFR prediction model to calculate the intravascular FFR value.

[0074] See also Figure 6, Schematic diagram of the network structure of the multi-task learning model for identifying lumens and bifurcations of the present invention. Considering that the lumen and bifurcation have their own structural and contour characteristics, in order to utilize the structural information without increasing the labeling cost, this technology extracts the contour of the mask and constructs two parallel convolutional neural networks. The convolution kernel sizes used include 5x5 convolution and 3x3 convolution. Considering that the lumen and bifurcation have different sizes in different images, skip connections are used to connect the shallow downsampling layer and the deep upsampling layer to achieve feature fusion at different semantic levels. The nonlinear activation function uses Relu, and the pooling layer uses maximum pooling. The input size of the image is 704x704. The mean and variance used in data preprocessing are both the mean [0.406, 0.456, 0.485] and variance [0.225, 0.224, 0.229] of ImageNet. No pre-trained network is used. The size of each batch during training is 32, the maximum value of the data epoch is 100, the initial learning rate is 0.01, and the Adam optimizer is used. The present invention uses a multi-task learning model to construct a mask prediction and contour drawing multi-task learning model, which can be plug-and-play. The module consists of two parallel convolution filters and can learn two related tasks at the same time. The module handles class imbalance, reduces outliers, alleviates structural information loss, and can handle multi-instance object segmentation well. The learning model helps the network learn multiple related tasks in parallel, so that the network has better generalization ability. Among them, mask prediction and contour extraction are both classification tasks.

[0075] Furthermore, the contour prior constraints of the contour and bifurcation in step 4 are to add the contour loss function to the network loss function so that the model can learn more structural information. The specific formula is:

[0076] L 总损失 =λ1(L 管腔掩膜 +L 分叉掩膜 )+λ2(L 管腔轮廓 +L 分叉轮廓 )

[0077] Where L represents the loss function, which uses the negative likelihood logarithm, that is,

[0078]

[0079] Where x represents the position of the pixel in the image. represents the predicted probability of the network model relative to the true label l, λ1 and λ2 represent the weight parameters of the mask and contour in multi-task learning, respectively.

[0080] In order to realize the automatic selection of reference frames, this patent adopts ResNet101 network to classify OCT images into healthy blood vessels and unhealthy blood vessels. Considering the problem that healthy blood vessels and unhealthy blood vessels have large intra-class differences and small inter-class differences, this patent proposes a triplet healthy blood vessel recognition model, see Figure 7 By adopting the triplet loss function, the distance between healthy blood vessels can be reduced, while the distance between healthy and unhealthy blood vessels can be increased, thereby improving the classification accuracy. Specifically, the triplet loss function is:

[0081] L = max(d(f 健康1 -f 健康2 )-d(f 健康1 -f 非健康1 )+margin,0)

[0082] The function d(·) represents the Euclidean distance, and f 健康1 、f 健康2 represents the deep features extracted from healthy blood vessels, f 非健康1 It represents the feature extracted from the unhealthy blood vessels, and margin is a bias term. Then the reference frame position can be selected as the position of the largest lumen area of ​​the healthy blood vessels within 10 mm near and far from the minimum lumen area and not at the bifurcation.

[0083] Furthermore, combining the Bernoulli equation, momentum equation, Navier-Stokes equation, lumen area, and reference frame area in fluid dynamics, the multi-parameter FFR model used in this patent is:

[0084]

[0085] ΔP=FV+FV 2

[0086]

[0087] Among them, P 近端 represents the proximal pressure of OCT data, ΔP is the intravascular pressure loss, and μ is the viscosity of blood, usually 4.0×10 -3 Pa·S, L represents the length of the vascular lumen, A 自动 represents the automatically selected reference frame, A s The lumen area at the narrow part, V represents the blood flow velocity, k represents the influence coefficient of the inlet and outlet on the pressure drop, which is usually taken as 1, and ρ represents the blood density, which is taken as ρ = 1050 kg / m 3 .

[0088] The following is an actual test case involved in the present invention, which interprets the key technical effects of the entire solution through the operation interface.

[0089] First, after data collection, the data processing interface is opened ( Figure 8 ), the system will complete image calibration and image enhancement to obtain IVUS images ( Fig. 9 ).

[0090] The blood flow reserve fraction prediction method based on optical coherence tomography imaging of the present invention, wherein the lumen and bifurcation recognition effect is schematically shown Fig.10 .

[0091] See also Fig.11 , which is a schematic diagram of bifurcation position identification of the method of the present invention.

[0092] Fig.12 A schematic diagram of the effect of the automatic reference frame selection result of the present invention.

[0093] Fig.13 Schematic diagram of OCT-FFR results of the present invention.

[0094] The present invention may also be applied to a storage medium made into any computer-readable instruction, which, when executed by one or more processors, enables the one or more processors to execute the overall optical coherence tomography-based blood flow reserve fraction prediction method of the present invention.

Claims

1. A device for predicting blood flow reserve fraction based on optical coherence tomography images, comprising a probe interface module and a display module, characterized in that: It also includes a detection module, a signal processing module and a data calculation module, wherein the detection module, the probe interface module, the signal processing module, the data calculation module and the display module are connected in sequence; The probe interface module is used for transmitting and receiving optical signals and links the detection module to withdraw; The detection module comprises a catheter (1), an elastic member (2), an optical fiber (3), a probe (4), a guide wire (5), a transparent outer tube sleeve (6), and a flushing liquid outlet (7), and is used to transmit the optical signal emitted by the probe interface module and transmit the optical signal returned from the tissue back to the probe interface module. The above signal is sent to the signal processing module through the probe interface module; The signal processing module is used to convert the optical signal emitted or received by the detection module into image information; and perform data calibration and image enhancement on the image information; The data calculation module builds a joint lumen and bifurcation recognition framework based on OCT images and multi-task learning mechanism. By processing the signal input by the detection module, it obtains the edge, morphology, and structural space information of the vascular lumen, obtains the vascular lumen boundary, calculates the vascular lumen area, and obtains the bifurcation position; By constructing a healthy blood vessel recognition model based on triples, the healthy blood vessels in the pullback OCT image are identified by increasing the differences between classes and the similarities within classes, and the reference frame is selected in combination with the lumen area and bifurcation position; the healthy blood vessel recognition model of the triplet is: L=max(d(f 健康1 -f 健康2 )-d(f 健康1 -f 非健康1 )+margin,0) The function d(·) represents the Euclidean distance, and f 健康1 、f 健康2 represents the deep features extracted from healthy blood vessels, f 非健康1 Represents the features extracted from non-healthy blood vessels, and margin is a bias term; The fluid dynamics model is further used to calculate the blood flow and flow resistance in the blood vessels in combination with the vascular lumen area and the reference frame area. The intravascular FFR value is obtained by combining the actual measured systolic and diastolic blood pressure data, thereby realizing the fully automatic calculation of FFR. The FFR calculation model is as follows: ΔP=FV+FV 2 Among them, P 近端 represents the proximal pressure of OCT data, ΔP is the intravascular pressure loss, and μ is the viscosity of blood, which is 4.0×10 - 3 Pa·S, L represents the length of the vascular lumen, A 自动 represents the automatically selected reference frame, A s The lumen area at the stenosis, V represents the blood flow velocity, k represents the influence coefficient of the inlet and outlet on the pressure drop, which is taken as 1, and ρ represents the blood density, which is taken as ρ = 1050 kg / m 3 ; The display module is used to present the processing results obtained by the data calculation module on the display.

2. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 1, characterized in that: The signal processing module at least includes a data acquisition unit and a data processing unit connected in sequence, and the data acquisition unit performs data acquisition on the signal acquired by the detection module through the detection interface module; The data processing unit at least includes a signal conversion module, an image calibration module and an image enhancement module, wherein the signal conversion module is used to convert the electrical signal received by the data acquisition unit into image information; An image calibration module, used to perform data calibration on the image obtained by the signal conversion module or the image cropping module; An image enhancement module, used for performing image enhancement on the image calibrated and reconstructed by the image calibration module; The data calculation module unit at least includes a lumen, bifurcation identification module, a lumen area calculation module, a reference frame automatic identification module, and an FFR fully automatic prediction module that are connected in sequence; the lumen and bifurcation identification module is used to extract contour information from the vascular lumen and bifurcation information; the lumen area calculation module is used to obtain the vascular lumen boundary and calculate the vascular lumen area based on the vascular lumen edge, morphology, and structural space information; the reference frame automatic calculation module uses a triplet loss function to build a healthy lumen identification model to increase the correlation within the class and the difference between classes, and then automatically selects the reference frame position in combination with the bifurcation position; the FFR fully automatic prediction module uses a fluid dynamics model, combined with the vascular lumen area and the reference frame area, to calculate the blood flow and flow resistance in the blood vessel, and combines the actually measured systolic and diastolic pressure data to obtain the intravascular FFR value.

3. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 2, characterized in that: The data processing unit also includes an image cropping module, which is used to segment the image information obtained by the signal conversion module and extract specific areas of the image.

4. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 1 or 2, characterized in that: A cleaning module is also arranged between the probe interface module and the detection module.

5. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 4, characterized in that: The detection module further comprises a catheter (1), an elastic component (2), and an optical fiber (3); the catheter (1) comprises an interior and an exterior; a cleaning module is arranged on the exterior of the catheter (1); the detection module is connected to a probe interface module via the exterior of the catheter (1); an ultrasonic probe (4), an optical fiber (3), and an elastic component (2) are arranged inside the catheter (1); the ultrasonic probe (4) is connected to the probe interface module via the optical fiber (3), and the elastic component (2) is arranged on the optical fiber (3).

6. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 5, characterized in that: A flushing liquid outlet (7) is provided at one end of the detection module catheter (1) which is located inside the body and away from the outside of the catheter (1).

7. The device for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 5 or 6, characterized in that: A transparent outer tube sleeve (6) is also sleeved on the outside of the inner body of the catheter (1), and the depth of the transparent outer tube sleeve (6) inside the body of the catheter (1) does not exceed the axial position of the ultrasonic probe (4) inside the body of the catheter (1).

8. A method for predicting blood flow reserve fraction based on optical coherence tomography images, characterized in that: The method comprises the following steps: Step 1: The optical signal returned by the near-infrared light from the tissue is superimposed and interfered with the optical signal reflected by the reference reflector. The generated interference signal is detected by the photodetector, which converts the interfered optical signal into an electrical signal. The amplified electrical signal is sent to the signal processing module through the probe interface module to complete the signal data acquisition. Step 2, performing signal conversion, image calibration, and image enhancement on the collected signal data to complete image data collection; Step 3, construct a multi-task learning model for lumen and bifurcation recognition; Step 4, establishing the prior constraints of the lumen and bifurcation contours; Step 5, combining the contour prior constraints in step 4, executing the multi-task learning model learning task of lumen and bifurcation recognition, obtaining the edge, morphology, and structural space information of the vascular lumen, calculating the vascular lumen area, and obtaining the bifurcation position; Step 6: construct a healthy blood vessel recognition model based on triples, identify healthy blood vessels in the pullback OCT image by increasing the differences between classes and the similarities within classes, and select a reference frame based on the lumen area and bifurcation position; the healthy blood vessel recognition model of the triplet is: L=max(d(f 健康1 -f 健康2 )-d(f 健康1 -f 非健康1 )+margin,0) The function d(·) represents the Euclidean distance, and f 健康1 、f 健康2 represents the deep features extracted from healthy blood vessels, f 非健康1 Represents the features extracted from non-healthy blood vessels, and margin is a bias term; Step 7: Using the fluid dynamics program, combined with the vascular lumen area and the reference frame area, a fully automatic FFR prediction model is constructed to calculate the intravascular FFR value. The FFR calculation model is as follows: ΔP=FV+FV 2 Among them, P 近端 represents the proximal pressure of OCT data, ΔP is the intravascular pressure loss, and μ is the viscosity of blood, which is 4.0×10 - 3 Pa·S, L represents the length of the vascular lumen, A 自动 represents the automatically selected reference frame, A s The lumen area at the stenosis, V represents the blood flow velocity, k represents the influence coefficient of the inlet and outlet on the pressure drop, which is taken as 1, and ρ represents the blood density, which is taken as ρ = 1050 kg / m 3 .

9. The method for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 8, characterized in that: The step 2 also includes cropping the collected signal data.

10. The method for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 8, characterized in that: The step 3 constructs two parallel convolutional neural networks of the lumen and bifurcation recognition multi-task learning model network structure, wherein the convolution kernel sizes used include a 5x5 convolution layer and a 3x3 convolution layer.

11. The method for predicting blood flow reserve fraction based on optical coherence tomography images according to claim 8, characterized in that: The contour and bifurcation prior constraints of step 4 are to add the contour loss function to the network loss function so that the model can learn more structural information. The specific formula is: L 总损失 =λ1(L 管腔掩膜 +L 分叉掩膜 )+λ2(L 管腔轮廓 +L 分叉轮廓 ) Where L represents the loss function, which uses the negative likelihood logarithm, that is, Where x represents the position of the pixel in the image. represents the predicted probability of the network model relative to the true label l, λ1 and λ2 represent the weight parameters of the mask and contour in multi-task learning, respectively.

12. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors execute the method for predicting blood flow reserve fraction based on optical coherence tomography images as described in any one of claims 8 to 11.