Portable Ultrasonic Cardiovascular Assessment System and Method

Abstract

The present invention discloses a portable ultrasonic cardiovascular assessment system and method, comprising: (a) an ultrasonic probe with an in-built FPGA module for real-time execution of ROI positioning and compressed LBP feature extraction; (b) a host processor configured to receive the compressed LBP features and perform dynamic jigsaw recombination; (c) a rhythm adaptive switching module that automatically enables non-linear EF calculation when the coefficient of variation of the RR interval is detected to be >15%; through technologies such as myocardial texture feature extraction, non-linear cardiac cycle recombination, and multi-scale time-frequency analysis, the present invention improves the adaptability and accuracy of EF calculation, especially suitable for patients with arrhythmia. In addition, this portable ultrasonic device can efficiently and accurately perform EF measurement, meeting the requirements of clinical rapid diagnosis and telemedicine.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing and cardiovascular function assessment, and specifically relates to a portable ultrasonic cardiovascular assessment system and method. Background Art

[0002] Cardiovascular diseases (CVDs) are one of the leading causes of death and disability globally. Among them, the left ventricular ejection fraction (EF) is an important indicator for evaluating cardiac function. In clinical practice, EF is usually evaluated by echocardiography to determine the heart's pumping ability and assist in the diagnosis and management of heart failure, coronary heart disease, and other cardiovascular diseases.

[0003] Currently, the calculation of EF in echocardiography mainly relies on the following methods:

[0004] Two-dimensional Simpson's method: This method is calculated based on the volume changes of the left ventricle at the end of systole and diastole and is applicable to two-dimensional ultrasound images. However, this method has the following problems: It relies on the boundary tracking of the left ventricle and is easily affected by the operator's experience, resulting in poor repeatability. In patients with arrhythmia, the morphological changes in consecutive cardiac cycles are large, making it difficult to accurately determine the end points of systole and diastole, and the calculation error is large.

[0005] M-mode ultrasound EF measurement method: This method calculates the EF value by measuring the change in the anteroposterior diameter of the left ventricle and is applicable to patients with relatively regular left ventricular function. However, in the presence of myocardial motion abnormalities (such as segmental motion abnormalities, arrhythmia, etc.), the accuracy of this method drops significantly.

[0006] EF calculation based on tissue Doppler imaging (TDI): In recent years, TDI has been used for EF calculation, and cardiac function is estimated based on parameters such as myocardial velocity and strain rate. However, this method is limited by the imaging quality of ultrasound equipment and noise interference, and has high requirements for image processing and computing capabilities, and is not suitable for portable ultrasound devices. Summary of the Invention

[0007] The purpose of the present invention is to provide a portable ultrasonic cardiovascular assessment system and method. Through technologies such as myocardial texture feature extraction, non-linear cardiac cycle reconstruction, and multi-scale time-frequency analysis, the adaptability and accuracy of EF calculation are improved, especially suitable for patients with arrhythmia. In addition, through optimized designs such as real-time calculation by the FPGA module and feature-compressed data transmission, the portable ultrasound device can efficiently and accurately measure EF, meeting the needs of clinical rapid diagnosis and telemedicine.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A method for calculating the cardiovascular ejection fraction based on dynamic reconstruction of myocardial texture, comprising the following steps:

[0010] (a) Real-time collect a sequence of cardiac ultrasound images and automatically locate the region of interest (ROI) of the left ventricular free wall;

[0011] (b) Extract the myocardial texture features of the ROI region to generate a dynamic texture waveform;

[0012] (c) Detect the systolic start and diastolic end events in the texture waveform through multi-scale time-frequency analysis;

[0013] (d) When arrhythmia is detected, non-linearly recombine the image segments of different cardiac cycles according to the events;

[0014] (e) Calculate the non-uniform ejection fraction (EF) based on the recombined dynamic jigsaw puzzle.

[0015] Among them, the myocardial texture feature in step (b) is the compressed local binary pattern (LBP), which is generated in the following way:

[0016] (i) Calculate the standard LBP coding for the ROI region in blocks;

[0017] (ii) Perform Gray coding conversion on the LBP coding;

[0018] (iii) Use parallel Hamming weight to calculate the compressed feature dimension.

[0019] Among them, the multi-scale time-frequency analysis in step (c) uses the stepped wavelet transform.

[0020] Among them, the non-linear recombination in step (d) includes:

[0021] (i) Construct a similarity matrix for the cardiac cycle segments, where the similarity calculation is:

[0022] ;

[0023] Among them: represents the similarity score between the and cardiac cycle segments, is the Pearson correlation coefficient function, is the compressed local binary pattern feature vector of the th segment, is the time stamp (unit: second) of the corresponding segment;

[0024] (ii) Use the greedy algorithm to connect the segments whose similarity exceeds the threshold.

[0025] Among them, the calculation formula for the non-uniform EF in step (e) is:

[0026] ;

[0027] Where: A max,i is the maximum area of myocardial texture in the i-th cardiac cycle (unit: pixel ; A min,i is the minimum area of myocardial texture in the i-th cardiac cycle (unit: pixel ; is the median function, used for extracting representative values in multiple cycles; k is the volume-area correction factor, and its value range is 0.85 to 0.95, calibrated through the clinical dataset.

[0028] A portable ultrasonic cardiovascular assessment system for implementing the above method, comprising:

[0029] a) An ultrasonic probe with an in-built FPGA module for real-time execution of ROI positioning and compressed LBP feature extraction;

[0030] b) A host processor configured to receive the compressed LBP features and perform dynamic puzzle recombination;

[0031] c) A rhythm adaptive switching module that automatically enables non-linear EF calculation when the coefficient of variation of the RR interval > 15% is detected.

[0032] Wherein, the FPGA module uses a look-up table (LUT) to solidify the weights of the transfer learning model, where:

[0033] i) The ROI positioning model is a lightweight version of MobileNetV2;

[0034] ii) The fully connected layer is replaced by bilinear interpolation approximate calculation.

[0035] Wherein, the data transmission between the ultrasonic probe and the host uses a feature compression protocol, including:

[0036] a) Compressing the 256×256 pixels of the original ultrasonic image into a 112-bit feature vector;

[0037] b) Using differential coding to reduce redundant data of consecutive frames.

[0038] A computer-readable storage medium stores a computer program, and when the program is executed by a processor, the steps of the above method are implemented.

[0039] An ultrasonic image processing device includes a memory, a processor, and a computer program stored on the memory. When the processor executes the program, the steps of the above method are implemented.

[0040] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are:

[0041] 1. Existing ultrasonic EF (ejection fraction) measurements are usually based on left ventricular volume calculation, while the present invention achieves a more refined assessment through dynamic reorganization of myocardial texture. The compressed local binary pattern (LBP) is used to extract myocardial texture features, which can capture local changes in myocardial contraction more accurately compared to traditional boundary tracking methods.

[0042] 2. Existing methods have large errors in patients with arrhythmia. The present invention detects systolic and diastolic events in the cardiac cycle through multi-scale time-frequency analysis, which can adapt to different rhythm changes. The non-linear reorganization method is used to reorganize different cardiac cycle segments by constructing a similarity matrix and a greedy algorithm, improving the accuracy of EF calculation in the case of arrhythmia.

[0043] 3. The stepped wavelet transform is used for multi-scale time-frequency analysis, which can capture dynamic changes in myocardial texture more effectively compared to the traditional FFT (Fast Fourier Transform). The volume-area correction factor based on the patient's body size can improve the accuracy of personalized EF calculation compared to the fixed model.

[0044] 4. Existing portable ultrasonic devices usually have limited computing power. The present invention combines an FPGA module and a host processor to improve computing efficiency: an FPGA module is built into the ultrasonic probe for real-time ROI positioning and compressed LBP feature extraction, reducing the computing burden on the host. A rhythm adaptive switching module automatically enables non-linear EF calculation when the RR interval variation > 15%, improving the applicability in the case of arrhythmia.

[0045] 5. Existing ultrasonic devices have a large amount of data, which affects transmission and storage. The present invention uses a lookup table (LUT) to solidify the weights of the transfer learning model, optimizing the computing burden. The feature compression protocol compresses ultrasonic images from 256×256 pixels to 112-bit feature vectors and uses differential coding to reduce inter-frame data redundancy, improving real-time performance.

[0046] 6. The present invention is not limited to methods only, but also proposes an ultrasonic image processing device, including a memory, a processor, and a computer program, forming a complete software and hardware combination solution. The computer-readable storage medium can be used for subsequent software upgrades and adaptation between different devices, improving the scalability of the solution.

[0047] In summary, compared with the prior art, the present invention has significant advantages in improving the accuracy of EF calculation, adapting to arrhythmia, optimizing computing and storage efficiency, etc. At the same time, combined with the design of portable ultrasonic devices, it realizes a more efficient, accurate, and portable cardiovascular assessment. Brief Description of the Drawings

[0048] Figure 1This is the working schematic diagram of the portable ultrasonic cardiovascular assessment system of the present invention;

[0049] Figure 2 This is the flow schematic diagram of the cardiovascular ejection fraction calculation method of the present invention. Specific embodiments

[0050] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] See Figure 1 and 2 , the present invention relates to a cardiovascular ejection fraction calculation method based on dynamic myocardial texture recombination, including the following steps:

[0052] (a) Real-time collect a sequence of cardiac ultrasound images and automatically locate the region of interest (ROI) of the left ventricular free wall;

[0053] (b) Extract the myocardial texture features of the ROI region to generate a dynamic texture waveform;

[0054] (c) Detect the systolic start and diastolic end events in the texture waveform through multi-scale time-frequency analysis;

[0055] (d) When arrhythmia is detected, non-linearly recombine the image segments of different cardiac cycles according to the events;

[0056] (e) Calculate the non-uniform ejection fraction (EF) based on the recombined dynamic jigsaw puzzle.

[0057] Further, the myocardial texture feature in step (b) is the compressed local binary pattern (LBP), which is generated by the following method:

[0058] (i) Calculate the standard LBP encoding for each block of the ROI region;

[0059] (ii) Perform Gray code conversion on the LBP encoding;

[0060] (iii) Use parallel Hamming weight to calculate the compressed feature dimension.

[0061] Further, the multi-scale time-frequency analysis in step (c) uses the stepped wavelet transform.

[0062] Further, the non-linear recombination in step (d) includes:

[0063] (i) Construct a similarity matrix for the cardiac cycle segments, where the similarity calculation is:

[0064] ;

[0065] Wherein: represents the similarity score of the th cardiac cycle segment, is the Pearson correlation coefficient function, is the compressed local binary pattern feature vector of the th segment, is the timestamp of the corresponding segment (unit: second);

[0066] (ii) Use a greedy algorithm to connect segments with similarity exceeding the threshold.

[0067] Furthermore, the formula for non-uniform EF in step (e) is:

[0068] ;

[0069] Wherein: A max,i is the maximum area of myocardial texture in the i-th cardiac cycle (unit: pixel ; A min,i is the minimum area of myocardial texture in the i-th cardiac cycle (unit: pixel ; is the median function, used for extracting representative values in multiple cycles; k is the volume-area correction factor, with a value range of 0.85 to 0.95, calibrated through a clinical dataset. The volume-area correction factor (k) in the present invention is an empirical coefficient used to estimate the actual volume change of the left ventricle from the area of myocardial texture in a two-dimensional image. Since the area change in the image cannot directly reflect the true change in ventricular volume, it is necessary to multiply the area difference by the correction factor k according to clinical data statistics. This factor is obtained from the fitting relationship between the large-sample true EF and the texture area, usually taking values between 0.85 and 0.95. It is an empirical value obtained through data training and is not a single fixed value. Under the influence of different patient body types, imaging angles, etc., the k value can be individually adjusted according to the device default model or the clinical calibration database.

[0070] A portable ultrasonic cardiovascular assessment system for implementing the method, comprising:

[0071] (a) An ultrasonic probe with an in-built FPGA module for real-time execution of ROI positioning and compressed LBP feature extraction;

[0072] (b) A host processor configured to receive the compressed LBP features and perform dynamic puzzle recombination;

[0073] (c) A rhythm adaptive switching module that automatically enables non-linear EF calculation when the coefficient of variation of the RR interval > 15%.

[0074] Further, the FPGA module uses a look-up table (LUT) to solidify and migrate the weights of the transfer learning model, where:

[0075] (i) The ROI positioning model is a lightweight version of MobileNetV2;

[0076] (ii) The fully connected layer is replaced with bilinear interpolation approximation calculation.

[0077] Further, the data transmission between the ultrasonic probe and the host uses a feature compression protocol, including:

[0078] (a) Compress the 256×256 pixels of the original ultrasonic image into a 112-bit feature vector;

[0079] (b) Use differential coding to reduce redundant data in consecutive frames.

[0080] A computer-readable storage medium stores a computer program, and when the program is executed by a processor, the steps of the method are implemented.

[0081] An ultrasonic image processing device includes a memory, a processor, and a computer program stored on the memory. When the processor executes the program, the steps of the method are implemented.

[0082] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for calculating the cardiovascular ejection fraction based on the dynamic reorganization of myocardial texture, characterized in that, It includes the following steps: (a) Collect a sequence of cardiac ultrasound images in real time and automatically locate the region of interest (ROI) of the left ventricular free wall; (b) Extract the myocardial texture features of the ROI region to generate a dynamic texture waveform; (c) Detect the systolic start and diastolic end events in the texture waveform through multi-scale time-frequency analysis; (d) When arrhythmia is detected, non-linearly recombine the image segments of different cardiac cycles according to the events; (e) Calculate the non-uniform ejection fraction (EF) based on the recombined dynamic jigsaw puzzle; The formula for calculating the non-uniform EF in step (e) is: ； Where: A max,i is the maximum area of myocardial texture in the i-th cardiac cycle (unit: pixel²); A min,i is the minimum area of myocardial texture in the i-th cardiac cycle (unit: pixel²); is the median function, used for extracting representative values in multiple cycles; k is the volume-area correction factor, with a value range of 0.85 to 0.95, calibrated through a clinical dataset.

2. The method according to claim 1, wherein The myocardial texture feature in step (b) is the compressed local binary pattern (LBP), which is generated in the following way: (i) Calculate the standard LBP encoding for each block of the ROI region; (ii) Perform Gray coding conversion on the LBP encoding; (iii) Use parallel Hamming weight to calculate the compressed feature dimension.

3. The method according to claim 1, characterized in that, The multi-scale time-frequency analysis in step (c) uses the stepped wavelet transform.

4. The method according to claim 1, wherein The non-linear recombination in step (d) includes: (i) Construct a similarity matrix for the cardiac cycle segments, where the similarity calculation is: ； Wherein: represents the similarity score of the th cardiac cycle segment, is the Pearson correlation coefficient function, is the compressed local binary pattern feature vector of the th segment, is the timestamp of the corresponding segment (unit: second); (ii) Use the greedy algorithm to connect the segments with similarity exceeding the threshold.

5. A portable ultrasonic cardiovascular assessment system for implementing the method according to any one of claims 1-4, characterized in that, It includes: (a) An ultrasound probe with an in-built FPGA module for real-time performing ROI positioning and compressed LBP feature extraction; (b) A host processor configured to receive the compressed LBP features and perform dynamic jigsaw puzzle recombination; (c) A rhythm adaptive switching module that automatically enables non-linear EF calculation when the coefficient of variation of the RR interval > 15% is detected.

6. The system according to claim 5, wherein The FPGA module uses a look-up table (LUT) to solidify the weights of the transfer learning model, where: (i) The ROI positioning model is a lightweight version of MobileNetV2; (ii) The fully connected layer is replaced by bilinear interpolation approximation calculation.

7. The system according to claim 6, wherein The data transmission between the ultrasound probe and the host uses a feature compression protocol, including: (a) Compress the 256×256 pixels of the original ultrasound image into a 112-bit feature vector; (b) Use differential coding to reduce the redundant data of consecutive frames.

8. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, the steps of the method described in any one of claims 1-4 are implemented.

9. An ultrasonic image processing device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, When the processor executes the program, the steps of the method described in any one of claims 1-4 are implemented.

Images