Artificial intelligence based system and method for automated measurements for pre-treatment planning - Patents.com

Abstract

A computer-implemented method for automatically measuring structural features of an anatomical tissue using an automated measurement system includes receiving a medical image of the anatomical tissue. The method includes detecting landmarks of the anatomical tissue based on the medical image using an artificial intelligence (AI)-based model and constructing an anatomical model of the anatomical tissue based on the detected landmarks. The method includes performing quantitative measurements of selected structural features of the anatomical model and generating a report including the quantitative measurements.

Description

[Technical field]

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 326,049, filed March 31, 2022, which is incorporated by reference in its entirety.

[0002] The present disclosure relates generally to systems and methods for automated measurements for pre-treatment planning. [Background technology]

[0003] Structural analysis of a patient's anatomy is essential for minimizing complications and clinical success of many medical procedures and surgeries. Transcatheter Aortic Valve Replacement (TAVR) is one such procedure that helps treat aortic stenosis in a non-invasive manner through extensive pre-procedural planning via advanced medical imaging techniques. Pre-procedural evaluation for TAVR allows clinicians to determine the optimal valve type, size, and placement strategy that will best prevent a variety of adverse effects, including elevated transvalvular pressure gradients, paravalvular leaks, aortic root rupture at the time of implantation, coronary artery occlusion, and others. However, pre-TAVR evaluation typically involves extensive manual procedures of making measurements of numerous anatomical structures from cardiac computed tomography (CT) images, which is time-consuming and subject to technician-to-technique variability. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] PCT / US2022 / 072240(WO2022 / 241425A1) Summary of the Invention

[0005] A computer-implemented method for automatically measuring structural features of an anatomical tissue using an automated measurement system includes receiving a medical image of the anatomical tissue. The method includes detecting landmarks of the anatomical tissue based on the medical image using an artificial intelligence (AI) based model and constructing an anatomical model of the anatomical tissue based on the detected landmarks. The method includes performing quantitative measurements of selected structural features of the anatomical model and generating a report including the quantitative measurements.

[0006] A system for automatically measuring structural features of an anatomical tissue includes a database and one or more automated measurement systems coupled to the database. Each of the one or more automated measurement systems includes a non-transitory computer readable medium having stored thereon a computer program having at least one code section for automatically measuring structural features of an anatomical tissue, the at least one code section being executable by at least one processor to cause the automated measurement system to perform steps including receiving a medical image of the anatomical tissue and detecting landmarks of the anatomical tissue based on the medical image using an artificial intelligence (AI) based model. The steps include constructing an anatomical model of the anatomical tissue based on the detected landmarks. The steps include making quantitative measurements of selected structural features of the anatomical model and generating a report including the quantitative measurements.

[0007] Another element of the systems and methods includes an AI-based model that includes two independently trained convolutional neural networks.

[0008] Another element of the system and method includes detecting landmarks by inputting the medical image into a center point detection model to generate a cropped medical image and inputting the cropped medical image into an aortic landmark model to predict anatomical landmarks.

[0009] Another element of the system and method includes a step further including displaying a visualization of the landmarks overlaid on the medical image.

[0010] Another element of the system and method includes steps further including incorporating human input into the AI-based model and training the AI-based model based on the human input.

[0011] Another element of the system and method involves incorporating human input by receiving input from a user to edit the locations of landmarks.

[0012] Another element of the system and method includes incorporating human input by displaying the anatomical structure model and quantitative measurements on a user interactive dashboard and allowing a user to modify the outline of the anatomical structure model or re-measure the quantitative measurements using the user interactive dashboard.

[0013] Another element of the system and method includes steps including recording the patient's human inputs and quantitative measurements in a database and cross-checking and tracking the human inputs and quantitative measurements for patients recorded at different times.

[0014] Another element of the system and method includes that the system is configured to allow multiple users to access the database and retrieve intermediate and final results from the steps performed by the automated measurement system.

[0015] Another element of the system and method includes a step further comprising generating a report that includes the quantitative measurements overlaid on the medical image.

[0016] Another element of the system and method includes steps further including generating a report including aortic valve sizing and type recommendations based on the quantitative measurements.

[0017] Another element of the system and method includes that the selected structural features include the valve annulus, sinuses of Valsalva, sinotubular junction (STJ), ascending aorta, left ventricular outflow junction (LVOT), aortic valve angulation, and / or coronary artery height.

[0018] Another element of the system and method includes receiving medical images, including receiving computed tomography (CT) images, magnetic resonance imaging (MRI) images, ultrasound images, and / or positron emission tomography (PET) images. [Brief description of the drawings]

[0019] [Figure 1] FIG. 13 is an illustration of a flow chart of automated segmentation and measurement of aortic anatomical features. [Diagram 2] FIG. 13 is an example diagram showing the locations of 15 landmark points detected and used for segmentation of the aortic root structure. [Diagram 3] FIG. 1 is a flow chart illustration of an artificial intelligence (AI) model-based pipeline for detecting aortic landmarks from computed tomography (CT) images. [Figure 4] FIG. 11 is a snapshot of an exemplary user interactive dashboard visualization of AI predicted aortic landmarks overlaid on a CT image. [Diagram 5] FIG. 13. Exemplary visualization of the 3D reconstruction of the aortic root generated by the landmark-guided segmentation algorithm. [Figure 6]FIG. 13 is an exemplary visualization of the annulus plane overlaid on a 3D reconstruction of the aortic root. [Figure 7] FIG. 13 is an exemplary visualization of the centerline detected along a 3D reconstruction of the aortic root. [Figure 8] FIG. 13 is a snapshot of an exemplary user interactive dashboard visualization of aortic root segmentation and measurement. [Figure 9] FIG. 1 shows a Bland-Altman plot of automated measurements of annular area compared to clinician measurements across 28 individual cases. [Figure 10A] FIG. 13 illustrates a portion of an exemplary report generated by the systems and methods disclosed herein outlining the automatically generated measurements along with sizing charts for two types of prosthetic valves based on manufacturer recommendations in the instructions for use. [Figure 10B] FIG. 13 illustrates a portion of an exemplary report generated by the systems and methods disclosed herein outlining the automatically generated measurements along with sizing charts for two types of prosthetic valves based on manufacturer recommendations in the instructions for use. [Figure 10C] FIG. 13 illustrates a portion of an exemplary report generated by the systems and methods disclosed herein outlining the automatically generated measurements along with sizing charts for two types of prosthetic valves based on manufacturer recommendations in the instructions for use. [Figure 11] FIG. 1 shows an example of a system for automated measurement of anatomical structures for pre-treatment planning. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] The present disclosure is directed to systems and methods for automatically generating measurements of aortic structure for pre-TAVR assessment, leading to enhanced efficiency of the measurement process and robustness of the anatomical measurements, ultimately enhancing the effectiveness of TAVR pre-procedure planning.

[0021] FIG. 1 shows an exemplary flow chart of a method 100 disclosed herein for an automatic reconstruction and measurement process. The method 100 includes three main steps: artificial intelligence-based landmark detection (step 102), landmark-guided aortic root segmentation (step 104), and automatic measurement (step 106). In step 102, an artificial intelligence (AI)-based pipeline includes two convolutional neural networks that are used for detection of several landmarks (e.g., 15 landmarks as an example). In step 104, a landmark-guided segmentation of the aortic root is constructed. In step 106, measurements of the aortic anatomy that are essential for pre-TAVR evaluation are automatically performed. In step 108, a report containing the measurement results and / or recommendations (for pre-procedure planning) is generated and / or output.

[0022] AI-based detection of anatomical landmarks In step 102, to start the automated measurement process from imaging data (e.g., computed tomography or CT scan data, magnetic resonance imaging or MRI, ultrasound, positron emission tomography or PET, etc.), the images are input into an artificial intelligence (AI)-based pipeline that detects several distinct aortic landmarks. The total number of landmarks can be any suitable number. FIG. 2 shows a set of CT scan images 200 showing the locations of landmark points detected and used for segmentation of the aortic root structure. In the illustrative example, there are 15 landmarks. CP1-3 and CP4-6 are the six commissure points. These two sets of points aim to resolve the finite coaptation height between the leaflets at the commissure. For example, CP1 and CP4 are at the lower and upper ends of this finite commissure coaptation line. Landmark C is at the central coaptation point of the three leaflets. L1-3 are three surface points on the leaflets, while H1-3 are the three leaflet hinges. Finally, two points at the left coronary ostia (LCO) and right coronary ostia (RCO) are detected for segmentation of the coronary arteries.

[0023] As shown in the flow chart 300 of FIG. 3, the AI-based pipeline consists of two independently trained convolutional neural networks. The first of the two convolutional neural networks is a first model (center point detection model) 302 for detecting the center of the aortic root. Trained on over 100 pairs of cardiac CT images and a manually created binary mask of the aortic root center, the center point detection model 302 outputs a probability heat map, which represents the probability that each of the voxels harbors the aortic root based on the image features that the center point detection model 302 captures from the training examples. The final predicted location of the aortic root center is determined by calculating the center of mass of the heat map. The predicted location of the aortic root center is used to crop the original CT image into a smaller volume, which is then fed into the second convolutional neural network for finer detection of the 15 aortic landmarks required to segment the anatomical features of the aortic root.

[0024] Working in synergy with the center point detection model 302, the second convolutional neural network detects 15 aortic landmarks from the image cropped around the center of the aortic root. Cropping the image around the center point allows for finer detection of aortic landmarks since the second model (aortic landmark model) 304 becomes focused on a smaller region of the image that does not house irrelevant information. The aortic landmark model 304 is trained and validated on the same cases as the center point detection model 302. The ground truth provided to train the algorithm / model are the manually labeled coordinates of the 15 aortic landmarks. The aortic landmark model 304 detects image features from the training examples that help distinguish the locations of aortic landmarks from other irrelevant locations.

[0025] Although the system can perform measurements fully automatically without any manual input from the input CT images, intermediate and final outputs of the system, such as predicted landmark locations and aortic root segmentation, may be visualized and reviewed by the user through an interactive graphical user interface (GUI) dashboard. Figure 4 displays an example of predicted landmark visualization through an interactive dashboard 400. A snapshot 402 of the user interactive dashboard visualization of the AI ​​predicted aortic landmarks is overlaid on the CT image.

[0026] Users can also visualize the landmarks and manually edit the landmark locations (if necessary) to ensure the predicted landmarks are accurate. The manually refined landmarks can then be used to train AI models, allowing for streamlined, constantly improving model performance.

[0027] As shown in Figure 4, the user can also visualize the landmarks to ensure the predicted landmarks are accurate and manually edit the landmark locations (if necessary). The visualization interface shown allows the user to move the crosshairs to the desired location to move any of the aortic landmarks. The manually refined landmarks can then be used to train the AI ​​model, allowing for model performance as a streamlined, constantly improving human-in-the-loop system. In this way, continued use of the system will improve the accuracy of the AI-based model over time through further training.

[0028] "Example 1" Example 1 describes an AI model architecture implemented to automatically detect anatomical landmarks in step 102. The architecture of a first convolutional neural network (first model, center point detection model) features a variant of U-Net utilized in computer vision tasks. The architecture features a series of convolutional layers followed by a series of transposed convolutional (ascending convolutional) layers. The relevant image features required to distinguish the center point of the aortic root from other anatomical regions are learned from the convolutional layers, while the transposed convolutional layers utilize the image features to output a volumetric heatmap of the probability that each voxel contains the center of the aortic root. Such an architecture is used for an initial coarse detection of the center of the aortic root, while a finer detection of important aortic landmarks, including the center point, is performed in a second neural network (second model, aortic landmark model).

[0029] The architecture of the second AI model (the aortic landmark model) features a variant of the ResNet model for computer vision tasks. The ResNet model accommodates residual connections that prevent exploding gradients during training, thus allowing for greater depth in the layers of the architecture. In combination with the patch-based method, two separate outputs are used in the head of the architecture, one vector for classification and the other for regression. In the classification head, each input is classified as containing one, multiple, or none of the landmarks of interest. In the regression head, the displacement of each landmark from the center of the input is regressed.

[0030] During data processing, all input images are thresholded between a fixed intensity range and resampled to the same resolution (1.5 x 1.5 x 1.5 mm / pixel for the center point model and 0.5 x 0.5 x 0.5 x 0.5 x 0.5 x 0.5 mm / pixel for the aortic landmark model). Images are cropped around the detected center point of the aortic root such that they are the same size and the cardiac anatomy around the aortic root is included within the image. As part of the preprocessing step to optimize the training process, the intensity of the input images is normalized to range between -1 and 1 and standardized to have mean zero and unit standard deviation.

[0031] To train the center-point landmark (heatmap-based) model, the preprocessed input images in their original form are input to the neural network. A mask volume with a sphere of fixed radius at the aortic center is provided for each image. The size of the mask volumes is the same as their corresponding input images. The network outputs a heatmap volume that represents the probability that a given voxel harbors an aortic landmark center. Finally, the centroid of the heatmap is determined to be the center point of the aortic root. The Euclidean distance between the predicted and ground truth coordinates is determined as the error of the model.

[0032] To train the aortic landmark (patch-based) model, the preprocessed input image is divided into patches of 100x100x100 pixels and randomly fed into the neural network as real input. The coordinates of the 15 manually labeled aortic landmarks are provided. The neural network classifies the patch as containing or not containing the landmark of interest and regresses the vector from the center of the patch to the actual coordinate of the landmark. During inference, uniform sampling is performed so that patches of the same size are sampled uniformly across the input image volume. As in center point detection, the Euclidean distance between the predicted coordinates of each of the 15 aortic landmarks and the ground truth coordinates is determined as the error and averaged.

[0033] For optimization of the central landmark (heatmap-based) model, the Adam optimizer is used with a learning rate of 0.001. The model is trained for 100 epochs with early stopping. The loss function used is binary cross entropy using Equation 1, where y i and

number

number

[0034] For optimization of the aortic landmark (patch-based) model, the AdamW optimizer is used with an initial learning rate of 0.001. The model is trained for 250 epochs with early stopping. The loss function utilized is a combination of binary cross entropy and absolute error loss (L1 loss) using Equation 2, where the variables with subscript c represent the coefficients of the classification loss, the variables with subscript r represent the coefficients of the regression loss, and y and

number

number

[0035] 3. Landmark-guided Aortic Root Segmentation In step 104, the landmark-guided automatic segmentation generates a 3D reconstruction of the aortic root from the CT images by utilizing the detected aortic landmarks as structural information of the aortic root. The 3D reconstruction of the aortic root model can be achieved via the systems and methods discussed in PCT / US2022 / 072240, filed May 11, 2022, and published as WO2022 / 241425A1. In the present disclosure, the landmark-guided aortic root segmentation in step 104 can include an automatic segmentation algorithm that utilizes a combination of adaptive thresholding, image gradient-based, and region growing techniques to segment the aortic root wall and the coronary arteries as individual sections. The segmented sections are then combined together to form a complete 3D reconstruction of the aortic root. FIG. 5 shows an example of a fully assembled reconstruction of the aortic root generated by the segmentation algorithm in step 104.

[0036] Automated measurement of aortic structure At step 106, measurements of the aortic structure (e.g., the aortic structure generated at step 104) are automatically performed for pre-procedural planning. Given the 3D structure of the aortic root from the landmark-guided segmentation algorithm, the system disclosed herein automatically performs accurate measurements of specific aortic structures, including the annulus, sinuses of Valsalva, sinotubular junction (STJ), ascending aorta, left ventricular outflow tract (LVOT), aortic valve angulation, coronary artery height, and / or other distinguishing features.

[0037] Measurements are made by quantifying geometric features such as area, perimeter, and length (diameter and height) of the segmentation at specific locations along the aortic root. The location of the measured aortic structure is inferred based on the detected aortic landmarks. For example, FIG. 6 shows a plane 600 along which the aortic annulus is measured, defined by three hinge points (e.g., three hinge points of the three aortic cusps that define the annulus plane).

[0038] For example, Figure 7 shows a centerline 700 detected along the aortic root. Aortic root centerline detection is based on 3D reconstruction to determine the approximate location of the STJ and ascending aorta, which are typically angled relative to the aortic annulus.

[0039] The segmentation and measurements of the aortic root, as well as the visualization of the AI ​​predicted landmarks, are visualized through a user interactive dashboard 400. Both the segmentation and measurements may be refined by the user by modifying the segmentation outline or re-measuring any of the measurements via the GUI. Figure 8 shows an example of the visualization of the segmentation and measurements of the aortic structure.

[0040] The measurement output of the system may be gated with an electrocardiogram (EKG) signal or respiratory cycle for images containing multiple phases across the heart or breath. In this way, a user may obtain measurements for any particular phase in the cardiac cycle. Additionally, measurements may be tracked across cardiac phases for temporal analysis.

[0041] Comparison of automated and clinical measurements To determine the accuracy of the automated measurement process, a comparison with clinician measurements is shown in Figure 9. In an illustrative example, 28 independent cases (from 28 different patients) are run through the automated measurement process (as in Figure 1) and their output annulus area measurements are compared with those of the clinician. As shown by the Bland-Altman plot, the two groups show good agreement with a mean percent error of -0.43%.

[0042] Two types of prosthetic valves, balloon-expandable and self-expandable, are sized based on both automated and clinician measurements using the manufacturer's recommendations in the instructions for use. Table 1 shows how the recommended valve sizing based on automated and clinical measurements shows good agreement, as there are no oversizes / undersizes based on automated measurements compared to clinical manual measurements. [Table 1]

[0043] For example, based on clinical manual measurements made on aortic root structural CT scans (dataset 1), clinicians found a 29 mm 2 ) or the Evolut valve (manufactured by Medtronic) sized at 34 mm circumference and 34 mm mean diameter. The automated measurement approach disclosed herein shows good agreement.

[0044] For example, based on clinical manual measurements made on aortic root structural CT scans (dataset 2), clinicians reported a 23 mm 2 A Sapien valve (manufactured by Edwards Lifesciences LLC) sized at 26 mm circumference and 26 mm mean diameter, or an Evolut valve (manufactured by Medtronic) sized at 26 mm circumference and 26 mm mean diameter, may be recommended. The automated measurement approach disclosed herein has shown good agreement.

[0045] It should be noted that the Sapien and Evolut valves are given as examples only, and the automated measurement system and method disclosed herein is capable of making recommendations for suitable types and sizes of heart valves from other manufacturers.

[0046] The automated measurement system and method disclosed herein can also generate and output a report including measurements and recommendations (e.g., appropriate valve sizes for different types of prosthetic valves based on manufacturer guidelines). Notably, the method 100 in FIG. 1 can include a step (step 108) for generating and / or outputting a report 1000 of measurements and / or recommendations. The report 1000 can be in any suitable format. FIGS. 10A-10C show an example of a report 1000 generated by the automated measurement system and method disclosed herein. The report 1000 can include measurements of essential aortic anatomy (valve annulus, sinuses of Valsalva, sinotubular junction (STJ), ascending aorta, left ventricular outflow tract (LVOT), aortic valve angulation, and coronary artery height, etc.) along with images illustrating the automated measurements overlaid on the original CT image and a recommended valve sizing for the desired prosthetic valve type.

[0047] 10A, report 1000 includes patient information (e.g., name, sex, age, native valve, native valve type, etc.) and recommended valve sizing details determined using the automated measurement systems and methods disclosed herein, including valve make / manufacturer, model, and dimensions (e.g., size by area, size by average diameter, size by circumference, etc.).

[0048] 10A-10C, a report 1000 includes measurements determined using the disclosed automated measurement system and method. Measurements include automatically measured sizing (e.g., area, circumference, major diameter, minor diameter, area-based diameter, circumference-based diameter) of annulus systole and diastole, LVOT, sinus, STJ, and ascending aorta. Measurements include automatically measured sizing (e.g., minimum and maximum height) of left and right coronary arteries (LCA and RCA) and aortic angle. Automated measurement marks 1010 are overlaid on the corresponding computed tomography, or CT scan, image.

[0049] Additionally, report 1000 may include a copy of the recommended valve manufacturer's Instructions for Use (IFU) that correspond to the recommended valve sizing determined by the automated measurement systems and methods disclosed herein. The IFU typically includes an image and a manufacturer's valve sizing chart.

[0050] The automated measurement system may be utilized for multiple users. The system may record patient information and measurement results in a database where the system may cross-check and track every input information about the patient to provide a personalized analysis. Accordingly, changes in the anatomical measurements of any particular patient may be tracked over a long-term examination or plan, a feature applicable to departments such as pediatrics or obstetrics and gynecology.

[0051] Similarly, the system may also be accessed by multiple users simultaneously. Users with their own login accounts can securely access intermediate and final outputs of the case. For example, a clinician may refine the aortic landmarks or measurement output of a particular case. The edited landmark coordinates can then be used for further AI training once verified.

[0052] 11 shows an illustrative computer architecture for an automated measurement system 1100 capable of implementing the automated measurement methods described herein. The computer architecture shown in FIG. 11 illustrates an exemplary computer system configuration in which the automated measurement system 1100 may be utilized to perform any aspect of the analyses and / or components presented herein, or any components in communication therewith.

[0053] The automated measurement system 1100 has a suitable computing environment (e.g., processors, memory, algorithms, controllers, communications networks, user interfaces, input devices, output devices, displays, etc.) to receive data / information, analyze, determine results / recommendations, and output.

[0054] In one embodiment, the automated measurement system 1100 can include one or more computers in communication with each other to perform tasks collaboratively. In one embodiment, the automated measurement system 1100 can include a cloud computing environment. Cloud computing can include providing computing services over a network connection using dynamically scalable computing resources.

[0055] In its most basic configuration, the automated measurement system 1100 typically includes at least one processing unit 1110 and a system memory 1120 (eg, non-transitory and / or transitory computer-readable medium).

[0056] The automated measurement system 1100 includes an input device 1130, such as a keyboard, keypad, switch, dial, mouse, track ball, touch screen, voice recognition device, card reader, paper tape reader, or other well-known input device.

[0057] The automated measurement system 1100 includes output devices 1140, which may also include a printer, a video monitor, a liquid crystal display (LCD), a touch screen display, a display, a speaker, etc. Additional devices may be connected to the bus to facilitate communication of data between the components of the automated measurement system 1100. All these devices are well known in the art and need not be discussed at length here.

[0058] The various components may communicate via wireless and / or hardwired or other desired available communication means, systems, and hardware. The automated measurement system 1100 may be connected to a cloud database 1150. Two or more automated measurement systems 1100 may be connected to the cloud database 1150, and data / information communicated and collected within the network may be used to enhance and improve accuracy of predictions and measurements.

[0059] The automated measurement system 1100 includes software and / or hardware components and modules required to enable the functionality of the modeling, simulation, and methods disclosed in this disclosure. In some embodiments, the automated measurement system 1100 includes an artificial intelligence (AI) module or algorithm and / or a machine learning (ML) module or algorithm (e.g., stored in the system memory 1120 and / or the cloud database 1150). The AI ​​and / or ML modules / algorithms can improve the predictive power of the models, simulations, and / or methods disclosed in this disclosure. For example, by using deep learning, AI, and / or ML model training, including patient information and any applicable input data to the computational model, the predictive power of the computational model can be greatly enhanced. The AI ​​and / or ML modules / algorithms also help improve the sensitivity and specificity of the predictions as the database grows.

[0060] In some embodiments, the automated measurement system 1100 can include virtual reality (VR), augmented reality (AR), and / or mixed reality displays, headsets, glasses, or any other suitable display devices as part of the output devices 1140 and / or input devices 1130. In some embodiments, the display devices may be interactive to allow a user to select from options including with or without AR, with or without VR, or may be integrated with real-time clinical imaging to assist the clinician in interacting and making decisions.

[0061] In some embodiments, the automated measurement system 1100 can interface with an extensive language learning model to output written or verbal descriptions of anatomical features and measurements. In response, the output of the system can be automatically analyzed and summarized to provide a detailed description of the anatomical measurements and their implications for pre-treatment planning.

[0062] Although the input types of the system have been discussed with reference to computed tomography, or CT scan images, the system is capable of processing many types of medical imaging, including CT, magnetic resonance imaging (MRI), ultrasound, positron emission tomography (PET), echocardiography, and others.

[0063] Although the exemplary systems and methods have been discussed in relation to the aortic valve and TAVR, the exemplary systems and methods can be readily applied to the mitral valve and to other structural thoracic pre-procedural planning.

[0064] The exemplary systems and methods may be found in routine clinical practice for TAVR and other structural thoracic pre-operative evaluation.

Claims

1. A computer implementation method for recommending heart valves based on the automatic measurement of structural characteristics of anatomical tissue using an automated measurement system, wherein the method is: Receiving medical images of the aforementioned anatomical tissue, Using an artificial intelligence (AI)-based model that includes two independently trained neural networks, comprising a first artificial intelligence (AI) model and a second artificial intelligence (AI) model, to detect anatomical tissue landmarks based on the medical image, To construct an anatomical structure model of the anatomical tissue based on the detected landmarks, To quantitatively measure the selected structural features of the aforementioned anatomical model, To generate a report including the aforementioned quantitative measurements and recommendations for the sizing and type of the heart valve. Computer implementation methods, including those mentioned above.

2. The computer implementation method according to claim 1, wherein the two independently trained neural networks are convolutional neural networks including a first AI model configured to detect the center of an anatomical structure of a target and a second AI model which is a model for detecting landmarks of anatomical tissue.

3. Detecting the aforementioned landmark The medical image is input into the first AI model to generate a cropped medical image. The trimmed medical image, including the left coronary artery (LCA) and the right coronary artery (RCA), is input into the second AI model to predict the landmarks of the anatomical tissue. The computer implementation method according to claim 2, including the method described in claim 2.

4. The computer implementation method according to claim 1, further comprising displaying a visualization of the landmark overlaid on the medical image.

5. The computer implementation method according to claim 1, further comprising incorporating human input into the AI-based model and training the AI-based model based on the human input.

6. The computer implementation method according to claim 5, wherein incorporating the aforementioned human input includes receiving user input for editing the location of the landmark.

7. Incorporating the aforementioned human input, The anatomical structural model and the quantitative measurements are displayed on a user-interactive dashboard. The user-interactive dashboard allows the user to modify the outline of the anatomical structure model or remeasure the quantitative measurements. The computer implementation method according to claim 5, including.

8. The aforementioned human input and the aforementioned quantitative measurements are recorded in a database, Cross-check and track the aforementioned human input and quantitative measurements recorded at different times. The computer implementation method according to claim 5, further comprising:

9. The computer implementation method according to claim 1, further comprising generating the report including the quantitative measurements overlaid on the medical image.

10. The computer implementation method according to claim 1, wherein the selected structural features include the minimum and maximum heights of the coronary arteries and the angular direction of the aorta.

11. The computer implementation method according to claim 1, wherein the selected structural features include a valve annulus, Valsalva sinus, sinus junction (STJ), ascending aorta, left ventricular outflow tract (LVOT), aortic valve angulation, and / or coronary artery height.

12. The computer implementation method according to claim 1, wherein receiving the medical images includes receiving computed tomography (CT) images, magnetic resonance imaging (MRI) images, ultrasound diagnostic images, and / or positron emission tomography (PET) images.

13. database and One or more automated measurement systems connected to the aforementioned database A system that recommends heart valves based on the automatic measurement of structural characteristics of anatomical tissue, comprising: Each of the one or more automated measurement systems is A non-temporary computer-readable medium storing a computer program having at least one code section for automatically measuring the structural characteristics of anatomical tissue. Equipped with, The aforementioned at least one code section is executed by at least one processor, Receiving medical images of the aforementioned anatomical tissue, Using an artificial intelligence (AI)-based model that includes two independently trained neural networks, comprising a first artificial intelligence (AI) model and a second artificial intelligence (AI) model, to detect anatomical tissue landmarks based on the medical image, To construct an anatomical structure model of the anatomical tissue based on the detected landmarks, To quantitatively measure the selected structural features of the aforementioned anatomical model, To generate a report including the aforementioned quantitative measurements and recommendations for the sizing and type of the heart valve. The automated measurement system is made to perform the steps including the following: system.

14. The system according to claim 13, wherein the two independently trained neural networks are convolutional neural networks comprising a first AI model configured to detect the center of an anatomical structure of a target and a second AI model which is a model for detecting landmarks of anatomical tissue.

15. Detecting the aforementioned landmark The medical image including the left coronary artery (LCA) and the right coronary artery (RCA) is input into the first AI model to generate a cropped medical image, The cropped medical image is input into the second AI model to predict the landmarks of the anatomical tissue. The system according to claim 14, including the system described in claim 14.

16. The system according to claim 13, wherein the step further includes displaying a visualization of the landmark overlaid on the medical image.

17. The system according to claim 13, wherein the steps further include incorporating human input into the AI-based model and training the AI-based model based on the human input.

18. The system according to claim 17, wherein incorporating the aforementioned human input includes receiving user input for editing the location of the landmark.

19. Incorporating the aforementioned human input, The anatomical structural model and the quantitative measurements are displayed on a user-interactive dashboard. The user-interactive dashboard allows the user to modify the outline of the anatomical structure model or remeasure the quantitative measurements. The system according to claim 17, including.

20. The above step is, The patient's personal input and quantitative measurements are recorded in the database, Cross-check and track the human input and quantitative measurements of the patient recorded at different times. The system according to claim 17, further comprising:

21. The system according to claim 13, configured to allow a large number of users to access the database and retrieve intermediate and final results from the steps performed by the automated measurement system.

22. The system according to claim 13, wherein the step further comprises generating the report including the quantitative measurements overlaid on the medical image.

23. The system according to claim 13, wherein the selected structural features include the minimum and maximum heights of the coronary arteries and the angular direction of the aorta.

24. The system according to claim 13, wherein the selected structural features include a valve annulus, Valsalva sinus, sinus junction (STJ), ascending aorta, left ventricular outflow tract (LVOT), aortic valve angulation, and / or coronary artery height.

25. The system according to claim 13, wherein receiving the medical images includes receiving computed tomography (CT) images, magnetic resonance imaging (MRI) images, ultrasound diagnostics, and / or positron emission tomography (PET) images.

Images