Respirator parameter prediction method and system based on digital twinning

Abstract

The invention belongs to the technical field of medical signal characterization and digital twinning, and relates to a breathing machine parameter prediction method and system based on digital twinning. The prediction method comprises the following steps: for each examination, respectively obtaining a plurality of lung trachea static three-dimensional model frames and lung contour static three-dimensional model frames, and fusing the frames in a one-to-one correspondence manner; performing tracheal tube correction on the fused three-dimensional model frame to obtain a lung standard static three-dimensional model; quantifying blood vessel morphological parameters, hemodynamic parameters and trachea parameters in the lung standard static three-dimensional model to obtain quantized values of the multiple parameters; the quantized values of the multiple parameters are imported into each static three-dimensional model frame, a lung static three-dimensional model with built-in parameters is obtained, the multiple lung dynamic three-dimensional models are fused into a dynamic digital twin system according to the time sequence, and parameter correction is conducted on the dynamic digital twin system; and calculating the lung health condition score and combining with external information to obtain respirator prediction parameters. The prediction parameters provided by the invention have relatively high reliability.

Description

Technical Field

[0001] The present invention relates to the fields of medical image processing and digital twin technology, and particularly to a method and system for predicting ventilator parameters based on digital twin. Background Art

[0002] The digital twin technology simulates the operating state of a device through a complex model, and the verification of the model accuracy is a difficult problem. In practical applications, various methods such as statistical analysis and experimental verification are usually required to ensure the reliability and accuracy of the model, and a model update mechanism also needs to be established to adapt to the update and improvement of the device.

[0003] The prior art combines the digital twin technology with a ventilator and measures airway pressure, volume, elasticity, hemodynamics, and pulmonary physical activities based on the digital twin model to automatically judge the patient's physical state, such as respiratory failure treatment, cardiopulmonary resuscitation, intraoperative and postoperative respiratory support, and first aid resuscitation. Combining the patient's current pulmonary health status, smoking history, and air quality in the living environment, parameters such as ventilator pressure, oxygen volume, and gas supply speed are given. Thereby providing a better analysis framework for managing treatment effects, simulating the entry of gas into the lungs, the lungs absorbing oxygen from the gas, etc., enabling doctors to evaluate patients in more detail, thus bringing better clinical effects, shorter intubation times, and more effective resource management.

[0004] However, the digital twin model in the prior art does not consider the real situations such as the real color, texture, and tracheal lumen space of the patient's lungs, does not consider the physical state and pulmonary health status of the patient who needs to use the ventilator, and does not consider the long-term effects of smoking history and air quality in the living environment on the lungs. Without the need for manual intervention, the reliability and safety of the working parameters provided by the digital twin model to the ventilator are relatively poor. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for predicting ventilator parameters based on digital twin, comprehensively considering the real situations such as the real color, texture, and tracheal lumen space of the patient's lungs, the physical state and pulmonary health status of the patient who needs to use the ventilator, and the long-term effects of smoking history and air quality in the living environment on the lungs, so as to obtain relatively reliable ventilator parameters.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a method for predicting ventilator parameters based on digital twin, including the following steps:

[0008] S10. Configure a complete pulmonary trachea examination, obtain pulmonary trachea images for each staying area of the endoscope, decompose, organize, and perform three-dimensional reconstruction on the pulmonary trachea images to obtain m static three-dimensional model frames of the pulmonary trachea;

[0009] S11. Obtain m fused images of pulmonary CT and MRI;

[0010] S12. Based on the trained medical image segmentation model, obtain the pulmonary contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image and perform three-dimensional reconstruction to obtain m static three-dimensional model frames of the pulmonary contour;

[0011] S13. Fuse the m static three-dimensional model frames of the pulmonary trachea and the m static three-dimensional model frames of the pulmonary contour in one-to-one correspondence, and correct the trachea pipeline for the fused three-dimensional model frames to obtain a standard static three-dimensional model of the lung, where the standard static three-dimensional model of the lung includes m static three-dimensional model frames;

[0012] S14. Quantify the blood vessel morphology parameters, hemodynamic parameters, and trachea parameters in the standard static three-dimensional model of the lung to obtain the quantification values of multiple parameters;

[0013] S15. Import the quantification values of multiple parameters into each static three-dimensional model frame to obtain a static three-dimensional model of the lung with built-in parameters;

[0014] S16. Perform 3D fusion on the static three-dimensional model of the lung with built-in parameters to obtain a dynamic three-dimensional model of the lung;

[0015] S17. Repeat S10~S16 to obtain multiple dynamic three-dimensional models of the lung from multiple complete pulmonary trachea examinations, and fuse the multiple dynamic three-dimensional models of the lung in chronological order into a dynamic digital twin system;

[0016] S18. Embed the trained medical image segmentation model in S12 and the quantification values of multiple parameters in S14 into the dynamic digital twin system and correct its parameters;

[0017] S19. Calculate the lung health status score based on the corrected parameters; introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the lung health status score to obtain the ventilator prediction parameters.

[0018] As a possible implementation, S10 includes the following sub-steps:

[0019] S100. Decompose the staying duration of the endoscope in each staying area to obtain n complete respiratory cycles, respectively perform image frame decomposition on each respiratory cycle, and obtain the respiratory cycle with the largest number of image frames, where this respiratory cycle contains m pulmonary trachea images;

[0020] S101. Use the interpolation method to supplement the image frames of n - 1 respiratory cycles with insufficient number of other image frames to m, so that each of the n respiratory cycles contains m frames of lung trachea images;

[0021] S102. Arrange the m frames of lung trachea images in each respiratory cycle in chronological order to obtain n frame sequences. Select the l-th frame image from each frame sequence, where 1 ≤ l ≤ m, to get n frame images. Use a quality evaluation model to screen the best-quality frame from the n frame images to form a new image sequence;

[0022] S103. Repeat S100 - S102 to obtain the image sequences of each staying area of the endoscope, and each image sequence includes m frames of images;

[0023] S104. Select the s-th frame image from all the image sequences for three-dimensional reconstruction, where 1 ≤ s ≤ m, to obtain m static three-dimensional model frames of the lung trachea.

[0024] As a possible implementation, S11 includes the following sub-steps:

[0025] S110. Calculate the average time T of n complete respiratory cycles,

[0026] S111. In a complete respiratory cycle, at intervals of T / m time, maintain the current state for t time to complete a group of CT / MRI image tomography scans, and obtain m CT images and m MRI images;

[0027] S112. Based on a deep learning model, pair and fuse the m CT images and m MRI images to obtain m fused images.

[0028] As a possible implementation, the medical image segmentation model is one of Unet, Unet++, or Mask-Rcnn.

[0029] As a possible implementation, the trachea pipeline correction includes trachea pipeline morphology correction and trachea pipeline color correction;

[0030] The trachea pipeline morphology correction is: replace the trachea pipeline in the l-th static three-dimensional model frame of the lung contour with the trachea pipeline contour in the l-th static three-dimensional model frame of the lung trachea, where 1 ≤ l ≤ m;

[0031] The trachea pipeline color correction is: attach the inner wall color of the trachea pipeline in the l-th static three-dimensional model frame of the lung trachea to the inner wall of the trachea pipeline in the l-th static three-dimensional model frame of the lung contour, where 1 ≤ l ≤ m.

[0032] As a possible implementation, there are 8 parameters, including: pulmonary vascular tortuosity, pulmonary artery pressure, vascular occlusion degree, tracheal pressure, tracheal elasticity, pulmonary color, pulmonary texture, and pulmonary vitality;

[0033] The quantization values of the 8 parameters are: pulmonary vascular tortuosity coefficient, average pulmonary artery pressure within one respiratory cycle, vascular occlusion quantization value, average tracheal pressure within one respiratory cycle, tracheal elasticity quantization value, pulmonary color quantization value, pulmonary texture quantization value, and pulmonary vitality quantization value.

[0034] As a possible implementation, the parameter correction in S18 includes:

[0035] S180. Repeat step S14 K times to obtain K quantization values generated by each of the 8 parameters respectively;

[0036] S181. Repeat step S15 K times to obtain K quantization values generated by each of the 9 built-in parameters of the pulmonary static three-dimensional model respectively. The 9 built-in parameters include 8 parameters and 1 system parameter; Execute steps S182 to S184 to correct the 9 built-in parameters;

[0037] S182. Use the K quantization values generated by the pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, pulmonary color, pulmonary texture, and pulmonary vitality among the 8 parameters in step S180 to correct the K quantization values generated by the pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, pulmonary color, pulmonary texture, and pulmonary vitality among the 9 built-in parameters in step S181, and obtain the corrected pulmonary vascular tortuosity coefficient, corrected vascular occlusion quantization value, corrected average tracheal pressure within one respiratory cycle, corrected tracheal elasticity quantization value, corrected pulmonary color quantization value, corrected pulmonary texture quantization value, and corrected pulmonary vitality quantization value;

[0038] S183. Perform polynomial fitting on the K quantization values generated by the pulmonary artery pressure to obtain the corrected average pulmonary artery pressure within one respiratory cycle;

[0039] S184. The system parameter is corrected in real time by the dynamic digital twin system.

[0040] As a possible implementation, the following method is used to calculate the pulmonary health status score:

[0041]

[0042] Among them, γ represents the quantization value of the corrected 9 built-in parameters, and λ i represents the weight corresponding to the quantization value of the corrected 9 built-in parameters.

[0043] As a possible implementation, the external information includes: years of smoking, air quality at the permanent residence, and physical condition; the ventilator prediction parameters include: supply pressure, oxygen supply amount, and supply speed.

[0044] In a second aspect, the present invention provides a ventilator parameter prediction system based on digital twin, including:

[0045] A three-dimensional reconstruction unit of the pulmonary trachea, configured to obtain pulmonary trachea images of each staying area of the endoscope in a complete pulmonary trachea examination once, decompose, organize, and perform three-dimensional reconstruction on the pulmonary trachea images to obtain m static three-dimensional model frames of the pulmonary trachea;

[0046] A three-dimensional reconstruction unit of the pulmonary contour, configured to obtain m fused images of pulmonary CT and MRI, obtain the pulmonary contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image based on a trained medical image segmentation model and perform three-dimensional reconstruction to obtain m static three-dimensional model frames of the pulmonary contour;

[0047] A three-dimensional model fusion and correction unit, configured to fuse the m static three-dimensional model frames of the pulmonary trachea and the m static three-dimensional model frames of the pulmonary contour correspondingly, and perform tracheal pipeline correction on the fused three-dimensional model frames to obtain a standard static three-dimensional model of the lung;

[0048] An internal parameter generation unit, configured to quantify the blood vessel morphology parameters, hemodynamic parameters, and tracheal parameters in the standard static three-dimensional model of the lung to obtain quantization values of multiple parameters, import the quantization values of multiple parameters into each static three-dimensional model frame included in the standard static three-dimensional model of the lung to obtain a static three-dimensional model of the lung with internal parameters;

[0049] A dynamic digital twin system construction unit, configured to perform 3D fusion on the static three-dimensional model of the lung with internal parameters to obtain a dynamic three-dimensional model of the lung, and fuse multiple dynamic three-dimensional models of the lung in chronological order into a dynamic digital twin system;

[0050] A parameter correction unit, configured to embed the trained medical image segmentation model and multiple quantization values into the dynamic digital twin system and perform parameter correction on it;

[0051] A ventilator parameter prediction unit, configured to calculate a pulmonary health status score based on the corrected parameters; introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the pulmonary health status score to obtain ventilator prediction parameters.

[0052] Compared with the prior art, the beneficial effects produced by the present invention are as follows:

[0053] 1. The method for predicting ventilator parameters based on digital twin proposed by the present invention establishes a dynamic digital twin system and corrects the system parameters to make it have the capabilities of intelligence, dynamics, and automatic evolution, so as to obtain stable and reliable ventilator parameters.

[0054] 2. The method for predicting ventilator parameters based on digital twin proposed by the present invention adopts an innovative parameter correction mechanism to enable the dynamic digital twin system to have more accurate parameter prediction capabilities.

[0055] 3. The method for predicting ventilator parameters based on digital twin proposed by the present invention can comprehensively consider baseline information such as the user's physical condition, lung health condition, living environment, and smoking history, and realize ventilator parameter prediction after comprehensive evaluation, with higher accuracy and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0057] Figure 1 is a flowchart of the method for predicting ventilator parameters based on digital twin provided by an embodiment of the present invention;

[0058] Figure 2 is a schematic diagram of the quantification process of the pulmonary vascular distortion coefficient in the method for predicting ventilator parameters based on digital twin provided by an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of the structure of the system for predicting ventilator parameters based on digital twin provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds and do not limit their sequence. Those skilled in the art can understand that the words such as "first" and "second" do not limit the quantity and execution order, and the words such as "first" and "second" do not necessarily limit to be different.

[0061] It should be noted that in the present invention, words such as "exemplary" or "for example" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0062] In the present invention, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The following at least one item (item) or similar expressions thereof refer to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one item (item) of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0063] The embodiments of the present invention aim to provide a ventilator parameter prediction method and system based on digital twins, which comprehensively considers the real conditions of the patient's lungs, such as the actual color, texture, and tracheal cavity space, the patient's physical condition and lung health status requiring the use of a ventilator, as well as the long-term impact of smoking history and air quality in the living environment on the lungs, to obtain ventilator parameters with higher reliability.

[0064] In the first aspect, the embodiment of the present invention provides a ventilator parameter prediction method based on digital twin, see Figure 1 , including the following steps:

[0065] S10. Perform a complete lung and tracheal examination, obtain lung and tracheal images in each endoscope stop area, decompose, organize, and 3D reconstruct the lung and tracheal images, and obtain m static 3D lung and tracheal model frames.

[0066] As a possible implementation, S10 includes the following sub-steps:

[0067] S100. Decomposing the dwell time of the endoscope in each dwell area to obtain n complete respiratory cycles, decomposing each respiratory cycle into image frames, and obtaining the respiratory cycle with the largest number of image frames, wherein the respiratory cycle includes m frames of lung and tracheal images;

[0068] As an example, when a user is doing a bronchoscopy, the lungs will inhale and exhale in a fixed cycle. One inhalation and exhalation is counted as one respiratory cycle, and the endoscope field of view will experience N respiratory cycles when it stays in the observation area. Each time the endoscope field of view stays is decomposed into n complete respiratory cycles, and the endoscope segments that are less than one respiratory cycle are discarded, so n <N。对n个完整的呼吸周期进行图像帧分解,找出图像帧数量最多的呼吸周期,假设该呼吸周期的图像帧数量为m。

[0069] S101. Use the interpolation method to supplement the image frames of n - 1 respiratory cycles with insufficient number of other image frames to m, so that each of the n respiratory cycles contains m frames of lung trachea images;

[0070] As an example, the number of image frames included in the jth respiratory cycle is m0. Since the number of image frames included in the ith respiratory cycle is the largest, so m0 < m, and m - m0 frames of images need to be supplemented to make the number of image frames in this respiratory cycle m.

[0071] The supplementation method is as follows: Convert the m0 frames of images into grayscale images, obtain the hash fingerprints of the grayscale images and calculate the Hamming distance between adjacent grayscale images, so as to obtain the similarity between two adjacent grayscale images, and sort the grayscale images in ascending order of similarity, select the top m - m0 images, and perform differences based on two adjacent images to obtain m - m0 image frames generated by the differences.

[0072] Each respiratory cycle is processed using the above method so that each of the n respiratory cycles contains m frames of lung trachea images.

[0073] S102. Arrange the m frames of lung trachea images in each respiratory cycle in chronological order to obtain n frame sequences. Select the lth frame image from each frame sequence, where 1 ≤ l ≤ m, to obtain n frames of images. Use a quality evaluation model to screen the best-quality frame from the n frames of images to form a new image sequence;

[0074] As an example, use resnet50 to screen the best-quality frame from the n frames of images.

[0075] S103. Repeat S100 - S102 to obtain the image sequences of each staying area of the endoscope, and each image sequence includes m frames of images;

[0076] As the endoscope field of view moves forward, repeat S100 - S102 to obtain the complete endoscopic images of the trachea and bronchi. Arrange them according to the spatial position where the field of view stays, and ensure that there is a complete respiratory cycle in each field of view staying area, and a new image sequence is obtained for each field of view. Each image sequence includes m frames of images.

[0077] S104. Take the sth frame image from all the image sequences for 3D reconstruction, where 1 ≤ s ≤ m, to obtain m static 3D model frames of the lung trachea.

[0078] As an example, take the first-frame image in all image sequences, and use VTK 3D reconstruction software to construct the first static 3D model frame of the lung trachea. Take the second-frame image in all image sequences, and use VTK 3D reconstruction software to construct the second static 3D model frame of the lung trachea… until all m-frame images are constructed, obtaining m static 3D model frames of the lung trachea, denoted as

[0079] S11. Obtain m fused images of lung CT and MRI;

[0080] As a possible implementation, S11 includes the following sub-steps:

[0081] S110. Calculate the average time T of n complete respiratory cycles,

[0082] S111. In a complete respiratory cycle, every time, maintain the current state for t time, complete a set of CT / MRI image tomography scans, and obtain m CT images and m MRI images;

[0083] As an example, assume that the i-th image tomography scan group is being collected at this moment, 1 ≤ i ≤ m. When the user completes the collection at the current moment, if in the exhalation phase, then when starting to inhale and then exhale from the 0 moment, the time of the respiratory cycle comes to Then maintain for t time again to complete the (i + 1)-th group of image tomography scans. In this way, m groups of images are obtained, that is, m CT images and m MRI images.

[0084] S112. Pair and fuse the m CT images and m MRI images based on a deep learning model to obtain m fused images.

[0085] As an example, use the IFCNN convolutional neural network model for pairing and fusion. Specifically, the first CT image and the first MRI image are fused, the second CT image and the second MRI image are fused…, the m-th CT image and the m-th MRI image are fused.

[0086] S12. Based on the trained medical image segmentation model, obtain the lung contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image and perform 3D reconstruction to obtain m static 3D model frames of the lung contour;

[0087] As an example, the medical image segmentation model is one of Unet, Unet++, or Mask-Rcnn. After obtaining the lung contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image, use VTK 3D reconstruction software to construct the static 3D model of the lung contour, denoted as The three-dimensional model is marked with vascular regions, tracheal regions, and tumor lesion regions.

[0088] S13. Fuse each of the m static three-dimensional model frames of the lung trachea with the corresponding one of the m static three-dimensional model frames of the lung contour;

[0089] As an example, the m static three-dimensional model frames of the lung trachea are The m static three-dimensional model frames of the lung contour are Fusing each corresponding one means that 3D Q1 is fused with 3D F1 3D Q2 is fused with 3D F2 3D Qm is fused with 3D Fm The fused three-dimensional model frame is denoted as

[0090] Perform tracheal pipeline correction on the fused three-dimensional model frame to obtain a standard static three-dimensional model of the lung. The standard static three-dimensional model of the lung includes m static three-dimensional model frames; Denote the corrected standard static three-dimensional model of the lung as

[0091] As a possible implementation, tracheal pipeline correction includes tracheal pipeline shape correction and tracheal pipeline color correction;

[0092] The tracheal pipeline shape correction is: replace the tracheal pipeline in the l-th static three-dimensional model frame of the lung contour with the tracheal pipeline contour in the l-th static three-dimensional model frame of the lung trachea, where 1 ≤ l ≤ m;

[0093] The tracheal pipeline color correction is: attach the inner wall color of the tracheal pipeline in the l-th static three-dimensional model frame of the lung trachea to the inner wall of the tracheal pipeline in the l-th static three-dimensional model frame of the lung contour, where 1 ≤ l ≤ m.

[0094] S14. Quantify the vascular morphology parameters, hemodynamic parameters, and tracheal parameters in the standard static three-dimensional model of the lung to obtain quantization values of multiple parameters;

[0095] As a possible implementation, the multiple parameters are 8, including: lung vascular tortuosity, pulmonary artery pressure, vascular occlusion degree, tracheal pressure, tracheal elasticity, lung color, lung texture, and lung vitality;

[0096] The quantization values of the 8 parameters are: lung vascular tortuosity coefficient, mean pulmonary artery pressure within one respiratory cycle, vascular occlusion quantization value, mean tracheal pressure within one respiratory cycle, tracheal elasticity quantization value, lung color quantization value, lung texture quantization value, and lung vitality quantization value.

[0097] See Figure 2, as an example, the quantification process of pulmonary vascular distortion is as follows: Select the frame 3D of the standard static three-dimensional model of the lungs in the maximum inspiration state from the standard static three-dimensional model of the lungs FQLmax , project the vascular contours in 3D FQLmax from the chest-back direction to obtain a vascular distribution projection image img x , and on the basis of the connected domain, separate each blood vessel in the image img x ; Use the Zhang-Suen thinning algorithm to find the center line of the blood vessel, and represent the length L of the blood vessel center line with the total number of pixels of the blood vessel center line k ; Make a normal line of the center line at a certain point on the blood vessel center line to intersect the blood vessel wall at two points, and obtain the blood vessel diameter D corresponding to this point kh , then the average blood vessel diameter D k = mean(D kh ), where k represents the kth blood vessel, h represents the hth pixel on the center line of the kth blood vessel, and mean represents the average value. Connect the end point O1 and the tail point O1 of the blood vessel, and the point where the connection line intersects the blood vessel is called the critical point. Record the number of critical points f and calculate the length L of the connection line o ; Calculate the maximum vertical offset d of the blood vessel center line relative to the connection line segment O1O2 to obtain the distortion coefficient of a single blood vessel: Furthermore, obtain the pulmonary vascular distortion coefficient:

[0098]

[0099] The average pulmonary artery pressure within one respiratory cycle: The average pulmonary artery pressure P' within one respiratory cycle is measured by right heart catheterization.

[0100] Vascular occlusion quantification value: Based on the previously marked vascular region and tumor lesion region, obtain the total vascular area S, calculate the total area C of the overlapping region between the tumor lesion region and the vascular region in the fused image, then the vascular occlusion quantification value is

[0101] The average tracheal pressure within one respiratory cycle: The average tracheal pressure Y' within one respiratory cycle is measured by a tracheal pressure gauge.

[0102] Tracheal elasticity quantification value: Use the Zhang-Suen thinning algorithm to find the center lines of all tracheas, and make a normal line of the center line at a point on the center line to intersect the tracheal wall at two points. The Euclidean distance between the two points is the tracheal diameter at the current position; Record the tracheal diameter value XQ at the moment of the end of inspiration and the tracheal diameter value TQ at the moment of the end of exhalation, and the tracheal elasticity quantification value

[0103] Pulmonary color quantification value: Pulmonary color quantification value

[0104] Lung texture quantization value: Record the horizontal pixel value and vertical pixel value in each fused image, calculate the reciprocal of the pixel area, and the lung texture quantization value W′ is the average value of the reciprocals of the pixel areas of m pixels.

[0105] Lung vital capacity quantization value: During a complete respiratory cycle, the three-dimensional volume of the lungs after all the gas in the lungs is exhaled is V min , and the volume of the lungs is the largest at the moment when inhalation ends. At this time, the three-dimensional volume of the lungs is V max , and from S110, the average time of a complete respiratory cycle is T, and the lung vital capacity quantization value

[0106] S15. Import the quantization values of the multiple parameters into each static three-dimensional model frame to obtain a lung static three-dimensional model with built-in parameters;

[0107] That is, import the above quantization values ξ′, P′, S′, Y′, T′, C′, W′, and H′ into each lung standard static three-dimensional model frame in the corrected lung standard static three-dimensional model of the corrected lung standard static three-dimensional model.

[0108] S16. Perform 3D fusion on the lung static three-dimensional model with built-in parameters to obtain a lung dynamic three-dimensional model;

[0109] As an example, import the lung static three-dimensional model frames with built-in parameters into unity in sequence to obtain a lung dynamic three-dimensional model.

[0110] S17. Repeat S10 - S16 to obtain multiple lung dynamic three-dimensional models for multiple complete lung trachea examinations, and fuse the multiple lung dynamic three-dimensional models in chronological order into a dynamic digital twin system;

[0111] S18. Embed the trained medical image segmentation model in S12 and the quantization values of the multiple parameters in S14 into the dynamic digital twin system and correct its parameters;

[0112] As an example, the embedding is achieved by importing the calculation process in the form of code.

[0113] As a possible implementation, the parameter correction includes:

[0114] S180. Repeat step S14 K times to obtain K quantization values generated by each of the 8 parameters respectively;

[0115] As an example, generate a list of K quantization values for each parameter as follows:

[0116] List ξ′ =[ξ′1,ξ′2,…,ξ′k ;

[0117] List P′ = [P′1, P′2, …, P′ k ;

[0118] List S′ = [S′1, S′2, …, S′ k ;

[0119] List Y′ = [Y′1, Y′2, …, Y′ k ;

[0120] List T′ = [T′1, T′2, …, T′ k ;

[0121] List C′ = [C′1, C′2, …, C′ k ;

[0122] List W′ = [W′1, W′2, …, W′ k ;

[0123] List H′ = [H′1, H′2, …, H′ k ;

[0124] S181. Repeat step S15 K times to obtain K quantization values generated by each of the nine built-in parameters of the lung static three-dimensional model. The nine built-in parameters include the eight parameters and one system parameter X′; Execute steps S182 to S184 to correct the nine built-in parameters;

[0125] As an example, the K quantization values generated by each built-in parameter and the K quantization values generated by the system parameter generate a list as follows:

[0126] List1 ξ″ = [ξ′1, ξ′2, …, ξ′ k ;

[0127] List P″ = [P′1, P′2, …, P′ k ;

[0128] List S″ = [S′1, S′2, …, S′ k ;

[0129] List Y″ = [Y′1, Y′2, …, Y′k ;

[0130] List T″ = [T′1, T′2, …, T′ k ;

[0131] List C″ = [C′1, C′2, …, C′ k ;

[0132] List W″ = [W′1, W′2, …, W′ k ;

[0133] List H″ = [H′1, H′2, …, H′ k ;

[0134] List X′ = [X′1, X′2, …, X′ k ;

[0135] S182. Using the K quantization values generated from the pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, pulmonary color, pulmonary texture, and pulmonary vitality among the 8 parameters described in step S180, correct the K quantization values generated from the pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, pulmonary color, pulmonary texture, and pulmonary vitality among the 9 built-in parameters described in step S181, to obtain the corrected pulmonary vascular tortuosity coefficient, the corrected vascular occlusion quantization value, the corrected average tracheal pressure within one respiratory cycle, the corrected tracheal elasticity quantization value, the corrected pulmonary color quantization value, the corrected pulmonary texture quantization value, and the corrected pulmonary vitality quantization value;

[0136] Next, taking the correction of the pulmonary vascular tortuosity as an example for illustration, first calculate the corresponding error between List ξ′ and List1 ξ″ to obtain the error-time relationship pair:

[0137] List ξ′-t = [(Δξ′1, t1), (Δξ′2, t2), …, (Δξ ′ k , t k );

[0138] Perform polynomial fitting on Δξ′ k , t k to obtain the corresponding relationship with time Δξ′ t →t, then the result of the quantization output of the vascular state of the corrected pulmonary static three-dimensional model is Mξ′ t = Zξ′ t+Δξ′ t , where Zξ′ t is the quantitative value of the blood vessel state given by the static three-dimensional model of the lungs, and Mξ′ t is the actual lung blood vessel distortion coefficient considering error correction.

[0139] Referring to the above steps, the corrected quantitative value of blood vessel occlusion MS′ t , the corrected average tracheal pressure MY′ t during one breathing cycle, the corrected tracheal elasticity quantitative value MT′ t , the corrected lung color quantitative value MC′ t , the corrected lung texture quantitative value MW′ t and the corrected lung vitality quantitative value MH′ t are obtained.

[0140] S183. Perform polynomial fitting on the K quantitative values generated by the pulmonary artery pressure to obtain the corrected average pulmonary artery pressure MP′ t during one breathing cycle;

[0141] S184. The system parameters are corrected and updated in real time by the dynamic digital twin system to obtain MX′ t .

[0142] S19. Calculate the lung health status score based on the corrected parameters; introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the lung health status score to obtain the ventilator prediction parameters.

[0143] As an example, the trained machine learning classifier includes a feature fitting sub-network and a classification sub-network; the corrected 9 parameter quantitative values Mξ′ t , MS′ t , MY′ t , MT′ t , MW′ t , MC′ t , MH′ t , MP′ t and MX′ t are input into the trained machine learning classifier (such as decision tree, random forest, SVM, etc.) for classification to obtain the lung health status score.

[0144] As a possible implementation, the following method is used to calculate the lung health status score:

[0145]

[0146] Among them, γ represents the quantitative value of the corrected 9 built-in parameters, and λ iIndicates the weights corresponding to the quantization values of the 9 built-in parameters after correction.

[0147] As a possible implementation, the external information includes: years of smoking, air quality at the permanent residence, and physical condition; the ventilator prediction parameters include: supply pressure, oxygen supply amount, and supply speed.

[0148] As an example, the external information is assigned in the following way:

[0149] The years of smoking are the actual assigned numbers;

[0150] Air quality at the permanent residence: 0 - very good, 1 - good, 2 - average, 3 - poor, 4 - very poor;

[0151] Physical condition: 0 - respiratory failure treatment, 1 - cardiopulmonary resuscitation, 2 - intraoperative and postoperative respiratory support, 3 - first aid resuscitation.

[0152] The lung health status score and the above assignment results of the external information are input into a trained parameter recommendation classifier (such as a machine learning model that can perform multiple regression like a decision tree or a random forest) for ternary regression prediction. The prediction targets include instantaneous regression values such as the ventilator supply pressure, ventilator oxygen supply amount, and ventilator supply speed. Also, according to the actual configuration parameters of the ventilator, the number of regressions output by the model and the specific prediction targets can be changed.

[0153] In a second aspect, the present invention provides a ventilator parameter prediction system based on digital twin. Refer to Figure 3 , including:

[0154] A three-dimensional reconstruction unit for the pulmonary trachea, which is used to obtain the pulmonary trachea images of each stay area of the endoscope during a complete pulmonary trachea examination, decompose, organize, and perform three-dimensional reconstruction on the pulmonary trachea images to obtain m static three-dimensional model frames of the pulmonary trachea;

[0155] A three-dimensional reconstruction unit for the pulmonary contour, which is used to obtain m fused images of the pulmonary CT and MRI, and based on a trained medical image segmentation model, obtain the pulmonary contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image and perform three-dimensional reconstruction to obtain m static three-dimensional model frames of the pulmonary contour;

[0156] A three-dimensional model fusion and correction unit, which is used to fuse the m static three-dimensional model frames of the pulmonary trachea and the m static three-dimensional model frames of the pulmonary contour correspondingly, and correct the tracheal ducts of the fused three-dimensional model frames to obtain a standard static three-dimensional model of the lungs;

[0157] An internal parameter generation unit, configured to quantify vascular morphological parameters, hemodynamic parameters, and tracheal parameters in a standard static three-dimensional model of the lungs, obtain quantization values of multiple parameters, import the quantization values of the multiple parameters into each static three-dimensional model frame included in the standard static three-dimensional model of the lungs, and obtain a static three-dimensional model of the lungs with internal parameters;

[0158] A dynamic digital twin system construction unit, configured to perform 3D fusion on the static three-dimensional model of the lungs with internal parameters to obtain a dynamic three-dimensional model of the lungs, and fuse multiple dynamic three-dimensional models of the lungs in chronological order into a dynamic digital twin system;

[0159] A parameter correction unit, configured to embed a trained medical image segmentation model and multiple quantization values into the dynamic digital twin system and perform parameter correction on it;

[0160] A ventilator parameter prediction unit, configured to calculate a lung health status score based on the corrected parameters; introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the lung health status score to obtain ventilator prediction parameters.

[0161] Although the present invention has been described in conjunction with various embodiments herein, however, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure content, and the description of the drawings, etc. In the specification, the term "comprising" does not exclude other components or steps, and "a" or "one" does not exclude the case of multiple. A single processor or other unit can implement several functions listed in the specification. Certain measures are recited in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0162] Although the present invention has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present invention. Accordingly, the present specification and the drawings are merely exemplary descriptions of the present invention and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A method for predicting ventilator parameters based on digital twin, characterized in that, It includes the following steps: S10. Configure a complete pulmonary trachea examination, obtain pulmonary trachea images of each staying area of the endoscope, decompose, organize, and perform three-dimensional reconstruction on the pulmonary trachea images to obtain m static three-dimensional model frames of the pulmonary trachea; S11. Obtain m fused images of pulmonary CT and MRI; S12. Based on the trained medical image segmentation model, obtain the pulmonary contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image and perform three-dimensional reconstruction to obtain m static three-dimensional model frames of the pulmonary contour; S13. Fuse the m static three-dimensional model frames of the pulmonary trachea and the m static three-dimensional model frames of the pulmonary contour in one-to-one correspondence, and correct the trachea pipeline of the fused three-dimensional model frames to obtain a standard static three-dimensional model of the lung, and the standard static three-dimensional model of the lung includes m static three-dimensional model frames; S14. Quantify the blood vessel morphology parameters, hemodynamic parameters, and trachea parameters in the standard static three-dimensional model of the lung to obtain quantification values of multiple parameters; S15. Import the quantification values of the multiple parameters into each static three-dimensional model frame to obtain a static three-dimensional model of the lung with built-in parameters; S16. Perform 3D fusion on the static three-dimensional model of the lung with built-in parameters to obtain a dynamic three-dimensional model of the lung; S17. Repeat S10 to S16 to obtain multiple dynamic three-dimensional models of the lung for multiple complete pulmonary trachea examinations, and fuse the multiple dynamic three-dimensional models of the lung in chronological order into a dynamic digital twin system; S18. Embed the trained medical image segmentation model in S12 and the quantification values of the multiple parameters in S14 into the dynamic digital twin system and correct its parameters; S19. Calculate the lung health status score based on the corrected parameters; introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the lung health status score to obtain the ventilator prediction parameters.

2. The method for predicting ventilator parameters based on digital twin according to claim 1, wherein, The S10 includes the following sub-steps: S100. Decompose the staying duration of the endoscope in each staying area to obtain n complete respiratory cycles, perform image frame decomposition on each respiratory cycle respectively, and obtain the respiratory cycle with the largest number of image frames. This respiratory cycle contains m pulmonary trachea images; S101. Use the interpolation method to supplement the image frames of the n - 1 respiratory cycles with insufficient number of image frames less than m, so that each of the n respiratory cycles contains m pulmonary trachea images; S102. Arrange the m pulmonary trachea images in each respiratory cycle in chronological order to obtain n frame sequences. Select the l-th image from each frame sequence, where 1 ≤ l ≤ m, to obtain n images. Use a quality evaluation model to screen the best-quality one from the n images to form a new image sequence; S103. Repeat S100 to S102 to obtain the image sequences of each staying area of the endoscope, and each image sequence includes m images; S104. Take the s-th image in all the image sequences for three-dimensional reconstruction, where 1 ≤ s ≤ m, to obtain m static three-dimensional model frames of the pulmonary trachea.

3. The method for predicting ventilator parameters based on digital twin according to claim 2, wherein, The S11 includes the following sub-steps: S110. Calculate the average time T of n complete respiratory cycles, S111. In a complete breathing cycle, every T / m time, maintain the current state for t time, complete a set of CT / MRI image tomography scans, and obtain m CT images and m MRI images; S112. Based on a deep learning model, pair and fuse the m CT images and m MRI images to obtain m fused images.

4. The method for predicting ventilator parameters based on digital twin according to claim 1, wherein The medical image segmentation model is one of Unet, Unet++, or Mask-Rcnn.

5. The method for predicting ventilator parameters based on digital twin according to claim 1, wherein The tracheal tube correction includes tracheal tube morphology correction and tracheal tube color correction; The tracheal tube morphology correction is: replacing the tracheal tube in the l-th static three-dimensional model frame of the lung contour with the tracheal tube contour in the l-th static three-dimensional model frame of the lung trachea, 1 ≤ l ≤ m; The tracheal tube color correction is: attaching the inner wall color of the tracheal tube in the l-th static three-dimensional model frame of the lung trachea to the inner wall of the tracheal tube in the l-th static three-dimensional model frame of the lung contour, 1 ≤ l ≤ m.

6. The method for predicting ventilator parameters based on digital twin according to claim 1, wherein The multiple parameters are 8, including: pulmonary vascular tortuosity, pulmonary artery pressure, vascular occlusion degree, tracheal pressure, tracheal elasticity, lung color, lung texture, and lung vitality; The quantization values of the 8 parameters are: pulmonary vascular distortion coefficient, mean pulmonary artery pressure within one breathing cycle, vascular occlusion quantization value, mean tracheal pressure within one breathing cycle, tracheal elasticity quantization value, lung color quantization value, lung texture quantization value, and lung vitality quantization value.

7. The method for predicting ventilator parameters based on digital twin according to claim 6, wherein, The parameter correction in S18 includes: S180. Repeat step S14 K times to obtain K quantization values generated by each of the 8 parameters respectively; S181. Repeat step S15 K times to obtain K quantization values generated by each of the 9 built-in parameters of the lung static three-dimensional model respectively. The 9 built-in parameters include the 8 parameters and 1 system parameter; perform steps S182 to S184 to correct the 9 built-in parameters; S182. Use the K quantization values generated by pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, lung color, lung texture, and lung vitality among the 8 parameters in step S180 to correct the K quantization values generated by pulmonary vascular tortuosity, vascular occlusion degree, tracheal pressure, tracheal elasticity, lung color, lung texture, and lung vitality among the 9 built-in parameters in step S181, to obtain the corrected pulmonary vascular distortion coefficient, corrected vascular occlusion quantization value, corrected mean tracheal pressure within one breathing cycle, corrected tracheal elasticity quantization value, corrected lung color quantization value, corrected lung texture quantization value, and corrected lung vitality quantization value; S183. Perform polynomial fitting on the K quantization values generated by the pulmonary artery pressure to obtain the corrected mean pulmonary artery pressure within one breathing cycle; S184. The system parameter is corrected in real time by the dynamic digital twin system.

8. The method for predicting ventilator parameters based on digital twin according to claim 7, wherein The following method is used to calculate the lung health status score: Among them, γ represents the quantization value of the 9 built-in parameters after correction, and λ i represents the weight corresponding to the quantization value of the 9 built-in parameters after correction.

9. The method for predicting ventilator parameters based on digital twin according to claim 1, wherein The external information includes: years of smoking, air quality of the permanent residence, and physical condition; the ventilator prediction parameters include: supply pressure, oxygen supply amount, and supply air speed.

10. A ventilator parameter prediction system based on digital twin, characterized in that, Including: A three-dimensional reconstruction unit for the pulmonary trachea, which is used to obtain pulmonary trachea images of each staying area of the endoscope in a complete pulmonary trachea examination, decompose, organize, and three-dimensionally reconstruct the pulmonary trachea images to obtain m static three-dimensional model frames of the pulmonary trachea; A three-dimensional reconstruction unit for the pulmonary contour, which is used to obtain m fused images of pulmonary CT and MRI, obtain the pulmonary contour, blood vessel contour, trachea contour, and tumor lesion contour in each fused image based on a trained medical image segmentation model and perform three-dimensional reconstruction to obtain m static three-dimensional model frames of the pulmonary contour; A three-dimensional model fusion and correction unit, which is used to correspondingly fuse the m static three-dimensional model frames of the pulmonary trachea and the m static three-dimensional model frames of the pulmonary contour, and correct the tracheal pipeline of the fused three-dimensional model frames to obtain a standard static three-dimensional model of the lung; An internal parameter generation unit, which is used to quantify the blood vessel morphology parameters, hemodynamic parameters, and tracheal parameters in the standard static three-dimensional model of the lung, obtain the quantification values of multiple parameters, and import the quantification values of the multiple parameters into each static three-dimensional model frame included in the standard static three-dimensional model of the lung to obtain a static three-dimensional model of the lung with internal parameters; A dynamic digital twin system construction unit, which is used to perform 3D fusion on the static three-dimensional model of the lung with internal parameters to obtain a dynamic three-dimensional model of the lung, and fuse multiple dynamic three-dimensional models of the lung in chronological order into a dynamic digital twin system; A parameter correction unit, which is used to embed a trained medical image segmentation model and multiple quantification values into the dynamic digital twin system and correct its parameters; A ventilator parameter prediction unit, which calculates the pulmonary health status score based on the corrected parameters; Introduce external information and assign values to it, and perform regression prediction based on the external information assignment and the pulmonary health status score to obtain ventilator prediction parameters.

Images