A lesion tracking and classification method and system for liver tumor ultrasound contrast imaging

Abstract

The present invention discloses a lesion tracking and classification method and system for liver tumor ultrasound contrast imaging images, wherein the lesion tracking and classification method for liver tumor ultrasound contrast imaging images involves: S1. collecting ultrasound contrast imaging radio frequency data streams, and preprocessing the collected contrast imaging radio frequency data streams to obtain processed two-dimensional radio frequency data; S2. performing anisotropic diffusion filtering on the obtained two-dimensional radio frequency data; S3. performing edge identification and tracing on the filtered two-dimensional radio frequency data to obtain a region of interest; S4. calculating the time intensity curve of the region of interest of the two-dimensional radio frequency data to obtain characteristic parameters during the arrival of the contrast agent; S5. inputting the obtained characteristic parameters into an LSTM neural network model for training to obtain a trained tumor classification model; S6. inputting the collected two-dimensional radio frequency data into the tumor classification model to obtain a tumor classification result.

Description

Technical Field

[0001] The present invention relates to the field of medical ultrasound technology, and in particular to a method and system for tracking and classifying liver tumor lesions using ultrasound contrast imaging. Background Art

[0002] Contrast-enhanced ultrasound (CEUS), also known as acoustic contrast imaging, utilizes contrast agents to enhance backscattered echoes, significantly improving tissue resolution, sensitivity, and specificity. Due to the real-time nature of CEUS, the entire process of contrast agent arrival and disappearance can be continuously observed and stored for quantitative analysis. Currently, the primary analysis method involves the physician defining a region of interest (ROI), with the system analyzing the temporal changes in contrast agent intensity within the ROI to generate a time-intensity curve (TIC). This quantitative analysis method can assist physicians in determining the benign or malignant nature of liver tumors, reducing the influence of subjective judgment.

[0003] However, existing technologies often have the following shortcomings:

[0004] 1. During the lesion tracking process, the movement of the probe and the patient's breathing are often affected, resulting in poor tracking results, which in turn affects the accuracy of the TIC curve and causes the characteristic parameters obtained from the curve to drift;

[0005] 2. During the examination, doctors often use their own experience and TIC curve to judge the benign or malignant nature of the current tumor lesion. Different doctors may come to different conclusions, which is still highly subjective. Summary of the Invention

[0006] The purpose of the present invention is to address the shortcomings of the existing technology and provide a lesion tracking and classification method and system for liver tumor ultrasound contrast imaging images. The method and system can be well integrated with the existing ultrasound system to meet the needs of clinicians to quickly and accurately analyze the benign and malignant nature of tumors through contrast imaging TIC curves. This process does not require human intervention, is simple to operate, and is easy to implement, thereby improving the objectivity and reliability of the analysis results.

[0007] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0008] A lesion tracking and classification method for liver tumor ultrasound contrast imaging, comprising:

[0009] S1. Acquire ultrasound contrast radiofrequency data stream and preprocess the acquired contrast radiofrequency data stream to obtain processed two-dimensional radiofrequency data;

[0010] S2. performing anisotropic diffusion filtering on the obtained two-dimensional RF data;

[0011] S3. performing edge recognition and tracing of the tumor region on the filtered two-dimensional RF data to obtain a region of interest;

[0012] S4. Calculate the time intensity curve of the region of interest of the two-dimensional radiofrequency data to obtain characteristic parameters of the contrast agent during its arrival;

[0013] S5. Input the obtained feature parameters into the LSTM neural network model for training to obtain a trained tumor classification model;

[0014] S6. Input the collected two-dimensional radiofrequency data into a tumor classification model to obtain a tumor classification result.

[0015] Furthermore, the step S1 is specifically as follows:

[0016] S11. Acquire the radiofrequency data stream signal during ultrasound contrast imaging;

[0017] S12. Detecting and demodulating the acquired RF data stream signal to obtain an envelope signal of the RF data;

[0018] S13. Perform logarithmic compression on the obtained envelope signal to obtain a preprocessed two-dimensional radio frequency signal.

[0019] Furthermore, the step S2 is specifically as follows:

[0020] S21. Preset the k value, iteration number t, and diffusion factor λ related to the thermal conductivity;

[0021] S22. Calculate the divergence in the four directions of east, south, west and north;

[0022] S23. Calculate the thermal conductivity in the four directions of east, south, west and north.

[0023] S24. Perform iterative filtering on the two-dimensional radio frequency signal based on the calculated divergence and thermal conductivity to obtain a filtered radio frequency signal.

[0024] Furthermore, in step S22, the divergence in the four directions of east, south, west and north is calculated, which is expressed as:

[0025] ▽ North (RF x,y )=RF x,y-1 -RF x,y

[0026] ▽ South (RF x,y )=RF x,y+1 -RF x,y

[0027] ▽ East (RF x,y )=RFx-1,y -RF x,y

[0028] ▽ West (RF x,y )=RF x+1,y -RF x,y

[0029] Among them, North (RF x,y ) represents the partial derivative in the north direction; ▽ South (RF x,y ) represents the partial derivative in the south direction; ▽ East (RF x,y ) represents the partial derivative in the east direction; ▽ West (RF x,y ) represents the partial derivative in the west direction; RF represents the radio frequency signal; x and y represent two-dimensional coordinates; ▽ represents the partial derivative.

[0030] Furthermore, in step S23, the thermal conductivity in the four directions of east, south, west and north is calculated, which is expressed as:

[0031]

[0032]

[0033]

[0034]

[0035] Among them, cNorth x,y Indicates the thermal conductivity in the north direction; cSouth x,y Indicates the thermal conductivity in the south direction; cEast x,y Indicates the thermal conductivity in the east direction; cWest x,y represents the thermal conductivity in the west direction.

[0036] Furthermore, in step S24, iterative filtering is performed, which is expressed as:

[0037] RF t+1 =RF t +λ(cNorth x,y ▽ North (RF t )+cSouth x,y ▽ South (RF t )

[0038] +cEast x,y ▽ East (RF t )+cWest x,y ▽West (RF t ))

[0039] Among them, RF t+1 Represents the signal of iteration t+1; RF t Represents the signal of iteration t.

[0040] Furthermore, the step S3 is specifically as follows:

[0041] S31. Convolve the filtered two-dimensional RF data with the Laplace template to obtain the two-dimensional RF data RF after Laplace transformation laplace ;

[0042] S32. Based on RF laplace The filtered two-dimensional RF data is sharpened to obtain the sharpened two-dimensional RF data RF sharp ;

[0043] S33. Extract RF sharp image, and obtain a feature image of two-dimensional radio frequency data;

[0044] S34. Perform threshold detection on the feature image to obtain an edge template;

[0045] S35. Trace the edge template to obtain an edge tracing curve and a region of interest.

[0046] Furthermore, the step S4 is specifically as follows:

[0047] S41. Calculate the average signal strength of all signals in the region of interest in each frame of the two-dimensional RF data;

[0048] S42. Expand the average signal intensity of the region of interest in all frames of the two-dimensional radiofrequency data along the time direction, and smooth it using a cubic spline interpolation function to obtain a time-intensity curve of the contrast perfusion.

[0049] S43. Obtain characteristic parameters of the contrast agent arrival process based on the time-intensity curve.

[0050] Furthermore, the step S5 is specifically as follows:

[0051] S51. Constructing the time-intensity curve and the tumor benign and malignant judgment results into a training set for the neural network model;

[0052] S52. Input the training set into the neural network model for training, use a 7-layer one-dimensional convolutional neural network to extract feature values, and use the Sigmoid function as the activation function to accelerate training;

[0053] S53. Setting a loss function of the neural network model and optimizing the neural network model by the loss function;

[0054] S54. Iterate steps S52-S53 until the training is completed to obtain a trained tumor classification model.

[0055] Accordingly, a liver tumor lesion tracking and classification system based on ultrasound contrast imaging is also provided, comprising:

[0056] An acquisition module is used to acquire ultrasound contrast radio frequency data stream and pre-process the acquired ultrasound contrast radio frequency data stream to obtain processed two-dimensional radio frequency data;

[0057] A filtering module, used for performing anisotropic diffusion filtering on the obtained two-dimensional radio frequency data;

[0058] an identification module for performing edge identification and tracing of the tumor area on the filtered two-dimensional radio frequency data to obtain a region of interest;

[0059] A calculation module is used to calculate the time intensity curve of the region of interest of the two-dimensional radio frequency data and obtain characteristic parameters of the contrast agent during its arrival;

[0060] A training module is used to input the obtained feature parameters into the LSTM neural network model for training to obtain a trained tumor classification model;

[0061] The classification module is used to input the collected two-dimensional radio frequency data into the tumor classification model to obtain the tumor classification results.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. The present invention performs detection, demodulation and logarithmic compression based on two-dimensional radio frequency signals, and simultaneously performs smoothing filtering that preserves the lesion boundary, thereby maximizing the distinction between the lesion and surrounding tissues and improving the accuracy of lesion identification.

[0064] 2. Using artificial intelligence training and analysis models, there is no need for human intervention to manually label data sets, avoiding experience differences between different doctors. The operation is simple and easy to implement, thereby improving the objectivity and reliability of the analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a flow chart of a lesion tracking and classification method for liver tumor ultrasound contrast imaging provided in Example 1. DETAILED DESCRIPTION

[0066] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0067] The purpose of the present invention is to address the deficiencies of the prior art and to provide a method and system for tracking and classifying liver tumor lesions using ultrasound contrast-enhanced imaging.

[0068] Example 1

[0069] This embodiment provides a method for tracking and classifying liver tumor lesions using ultrasound contrast-enhanced imaging. Figure 1 As shown, including:

[0070] S1. Acquire ultrasound contrast radiofrequency data stream and preprocess the acquired contrast radiofrequency data stream to obtain processed two-dimensional radiofrequency data;

[0071] S2. performing anisotropic diffusion filtering on the obtained two-dimensional RF data;

[0072] S3. performing edge recognition and tracing of the tumor region on the filtered two-dimensional RF data to obtain a region of interest;

[0073] S4. Calculate the time intensity curve of the region of interest of the two-dimensional radiofrequency data to obtain characteristic parameters of the contrast agent during its arrival;

[0074] S5. Input the obtained feature parameters into the LSTM neural network model for training to obtain a trained tumor classification model;

[0075] S6. Input the collected two-dimensional radiofrequency data into a tumor classification model to obtain a tumor classification result.

[0076] In step S1, ultrasound contrast radio frequency data stream is collected and preprocessed to obtain processed two-dimensional radio frequency data; specifically,

[0077] S11. Acquiring a radio frequency (RF) data stream signal during ultrasound contrast imaging;

[0078] S12. Detecting and demodulating the acquired RF data stream signal to obtain an envelope signal of the RF data; wherein the detection and demodulation methods include orthogonal demodulation and Hilbert transform;

[0079] S13. Perform logarithmic compression on the obtained envelope signal to obtain a preprocessed two-dimensional RF signal; the purpose of performing logarithmic compression is to stretch the effective signal area and compress the invalid signal area.

[0080] In step S2, anisotropic diffusion filtering is performed on the obtained two-dimensional radio frequency data to retain the tumor edge signal in the two-dimensional radio frequency data while performing smoothing filtering; wherein the anisotropic diffusion filtering is specifically as follows:

[0081] S21. Preset the k value, iteration number t, and diffusion factor λ related to the thermal conductivity; a larger k value results in smoother results and makes it less likely that the tumor boundary will be preserved;

[0082] S22. Calculate the divergence in the four directions of east, south, west and north, that is, find the partial derivative of the signal in the four directions, expressed as:

[0083] ▽ North (RF x,y )=RF x,y-1 -RF x,y

[0084] ▽ South (RF x,y )=RF x,y+1 -RF x,y

[0085] ▽ East (RF x,y )=RF x-1,y -RF x,y

[0086] ▽ West (RF x,y )=RF x+1,y -RF x,y

[0087] Among them, North (RF x,y ) represents the partial derivative in the north direction; ▽ South (RF x,y ) represents the partial derivative in the south direction; ▽ East (RF x,y ) represents the partial derivative in the east direction; ▽ West (RF x,y ) represents the partial derivative in the west direction; RF represents the radio frequency signal; x and y represent two-dimensional coordinates; ▽ represents the partial derivative.

[0088] S23. Calculate the thermal conductivity in the four directions cNorth, cSouth, cEast, and cWest, expressed as:

[0089]

[0090]

[0091]

[0092]

[0093] Among them, cNorth x,y Indicates the thermal conductivity in the north direction; cSouth x,y Indicates the thermal conductivity in the south direction; cEast x,y Indicates the thermal conductivity in the east direction; cWest x,y represents the thermal conductivity in the west direction.

[0094] S24. Perform iterative filtering on the two-dimensional RF signal based on the calculated divergence and thermal conductivity to obtain a filtered RF signal; wherein the iterative filtering is expressed as:

[0095] RF t+1 =RF t +λ(cNorth x,y ▽ North (RF t )+cSouth x,y ▽ South (RF t )

[0096] +cEast x,y ▽ East (RF t )+cWest x,y ▽ West (RF t ))

[0097] Among them, RF t+1 Represents the signal of iteration t+1; RF t Represents the signal of iteration t.

[0098] In step S3, the edge of the tumor region is identified and traced on the filtered two-dimensional radio frequency data to obtain a region of interest; specifically:

[0099] S31. Convolve the filtered two-dimensional RF data with the Laplace template to obtain the two-dimensional RF data RF after Laplace transformation laplace ;

[0100] S32. Based on RF laplace The filtered two-dimensional RF data is sharpened to obtain the sharpened two-dimensional RF data RF sharp ;

[0101] S33. Extract RFsharp image, and obtain a feature image of two-dimensional radio frequency data;

[0102] S34. Perform threshold detection on the feature image to obtain an edge template;

[0103] S35. Trace the edge template to obtain an edge tracing curve and a region of interest (ROI).

[0104] In step S4, the time intensity curve of the region of interest of the two-dimensional radio frequency data is calculated to obtain the characteristic parameters of the contrast agent arrival process; specifically:

[0105] S41. Calculate the average signal strength Avg1...Avg of all signals in the region of interest in each frame of the two-dimensional radio frequency data. N ; Where N represents the number of frames in the data stream;

[0106] S42. The average signal intensity of the region of interest in all frames of the two-dimensional RF data is expanded along the time direction and smoothed using a cubic spline interpolation function to eliminate parameter errors caused by fluctuation interference (such as human breathing and probe movement) to obtain the time-intensity TIC curve of contrast perfusion.

[0107] S43. According to the time-intensity TIC curve, characteristic parameters of the contrast agent arrival process are obtained; wherein the characteristic parameters include:

[0108] Arrival Time (AT): The time it takes for the intensity of the contrast agent to change significantly relative to its initial state.

[0109] Time to Peak (TTP): The time from the initial state to the time when the contrast agent signal intensity reaches the maximum value;

[0110] Area under the TIC curve (AUC): the area under the time-intensity curve of the angiography process;

[0111] Peak Intensity (PI): peak intensity of angiography;

[0112] Ascending Slope (AS): The slope between the start and peak of lesion perfusion on the curve;

[0113] Descending Slope (DS): The descending slope of the curve.

[0114] In step S5, the obtained feature parameters are input into the LSTM neural network model for training to obtain a trained tumor classification model; specifically:

[0115] S51. The TIC curve data expanded over time and the benign and malignant tumor judgment results are constructed as a training set for the neural network model; wherein the benign and malignant tumor judgment results are: 0 for benign and 1 for malignant;

[0116] S52. Use Keras to build an LSTM neural network model. Input the training set into the LSTM neural network model for training. Use a 7-layer one-dimensional convolutional neural network to extract feature values. The convolutional layer uses the Sigmoid function as the activation function to accelerate training. Compress the segmentation plane to the interval [0, 1] to determine whether the data is a positive or negative sample.

[0117] S53. Set the loss function of the neural network model and optimize the neural network model through the loss function; the loss function Loss is expressed as:

[0118]

[0119] Among them, z represents the actual value; z' represents the classification result.

[0120] S54. Start model training. Assume that the training cycle is 40 periods, and iterate steps S52-S53 until the training is completed to obtain a trained tumor classification model.

[0121] In step S6, the collected two-dimensional radio frequency data is input into the tumor classification model to obtain the tumor classification result.

[0122] The obtained tumor classification model is used to classify tumors into benign and malignant types, and based on the collected two-dimensional radio frequency data signals, the corresponding lesions are divided into benign and malignant types, thereby assisting doctors in diagnosis.

[0123] Compared with the prior art, this embodiment has the following beneficial effects:

[0124] 1. This embodiment performs detection, demodulation, and logarithmic compression on a two-dimensional radio frequency signal, while simultaneously performing smoothing filtering that preserves the lesion boundary, thereby maximizing the degree of distinction between the lesion and surrounding tissue and improving the accuracy of lesion identification.

[0125] 2. Using artificial intelligence training and analysis models, there is no need for manual data set labeling, which avoids differences in experience between different doctors. The operation is simple and easy to implement, thereby improving the objectivity and reliability of the analysis results.

[0126] 3. This embodiment uses the characteristic parameters of the curve calculated from the angiography TIC curve and inputs them into the LSTM neural network model to automatically classify benign and malignant liver tumors, assist doctors in diagnosis, and improve the objectivity and accuracy of the diagnostic results.

[0127] Example 2

[0128] This embodiment provides a lesion tracking and classification system for liver tumor ultrasound contrast imaging, including:

[0129] An acquisition module is used to acquire ultrasound contrast radio frequency data stream and pre-process the acquired ultrasound contrast radio frequency data stream to obtain processed two-dimensional radio frequency data;

[0130] A filtering module, used for performing anisotropic diffusion filtering on the obtained two-dimensional radio frequency data;

[0131] an identification module for performing edge identification and tracing of the tumor area on the filtered two-dimensional radio frequency data to obtain a region of interest;

[0132] A calculation module is used to calculate the time intensity curve of the region of interest of the two-dimensional radio frequency data and obtain characteristic parameters of the contrast agent during its arrival;

[0133] A training module is used to input the obtained feature parameters into the LSTM neural network model for training to obtain a trained tumor classification model;

[0134] The classification module is used to input the collected two-dimensional radio frequency data into the tumor classification model to obtain the tumor classification results.

[0135] It should be noted that this embodiment provides a lesion tracking and classification system for liver tumor ultrasound contrast imaging, which is similar to the first embodiment and will not be described in detail here.

[0136] Compared with the prior art, this embodiment has the following beneficial effects:

[0137] 1. This embodiment performs detection, demodulation, and logarithmic compression on a two-dimensional radio frequency signal, while simultaneously performing smoothing filtering that preserves the lesion boundary, thereby maximizing the degree of distinction between the lesion and surrounding tissue and improving the accuracy of lesion identification.

[0138] 2. Using artificial intelligence training and analysis models, there is no need for human intervention to manually label data sets, avoiding experience differences between different doctors. The operation is simple and easy to implement, thereby improving the objectivity and reliability of the analysis results.

[0139] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A liver tumor lesion tracking and classification system based on ultrasound contrast-enhanced imaging, characterized in that: include: An acquisition module is used to acquire ultrasound contrast radio frequency data stream and pre-process the acquired ultrasound contrast radio frequency data stream to obtain processed two-dimensional radio frequency data; A filtering module, used for performing anisotropic diffusion filtering on the obtained two-dimensional radio frequency data; an identification module for performing edge identification and tracing of the tumor area on the filtered two-dimensional radio frequency data to obtain a region of interest; A calculation module is used to calculate the time intensity curve of the region of interest of the two-dimensional radio frequency data and obtain characteristic parameters of the contrast agent during its arrival; A training module is used to input the obtained feature parameters into the LSTM neural network model for training to obtain a trained tumor classification model; A classification module is used to input the collected two-dimensional radiofrequency data into a tumor classification model to obtain a tumor classification result; The filtering module is specifically used for: Preset the k value, iteration number t and diffusion factor λ related to thermal conductivity; Calculate the divergence in the four directions of east, south, west and north; Calculate the thermal conductivity in the four directions of east, south, west and north; Iteratively filtering the two-dimensional radio frequency signal according to the calculated divergence and thermal conductivity to obtain a filtered radio frequency signal; The divergence in the four directions of east, south, west and north is calculated as follows: in, represents the partial derivative in the north direction; represents the partial derivative in the south direction; represents the partial derivative in the east direction; represents the partial derivative in the west direction; RF represents the radio frequency signal; x and y represent two-dimensional coordinates; represents partial derivative; The thermal conductivity coefficients in the four directions of east, south, west and north are calculated as follows: Among them, cNorth x,y Indicates the thermal conductivity in the north direction; cSouth x,y Indicates the thermal conductivity in the south direction; cEast x,y Indicates the thermal conductivity in the east direction; cWest x,y represents the thermal conductivity in the west direction; The iterative filtering is expressed as: Among them, RF t+1 Represents the signal of iteration t+1; RF t Represents the signal of iteration t.

2. The liver tumor lesion tracking and classification system based on ultrasound contrast-enhanced imaging according to claim 1, characterized in that: The acquisition module is specifically used for: Acquiring radio frequency data stream signals during ultrasound contrast imaging; Detecting and demodulating the acquired radio frequency data stream signal to obtain an envelope signal of the radio frequency data; Logarithmically compress the obtained envelope signal to obtain a preprocessed two-dimensional radio frequency signal.

3. The liver tumor lesion tracking and classification system based on ultrasound contrast-enhanced imaging according to claim 1, characterized in that: The identification module is specifically used for: Convolve the filtered two-dimensional RF data with the Laplace template to obtain the two-dimensional RF data RF after Laplace transformation laplace ; RF-based laplace The filtered two-dimensional RF data is sharpened to obtain the sharpened two-dimensional RF data RF sharp ; Extract RF sharp image, and obtain a feature image of two-dimensional radio frequency data; Perform threshold detection on the feature image to obtain the edge template; The edge template is traced to obtain the edge tracing curve and the region of interest.

4. The liver tumor lesion tracking and classification system based on ultrasound contrast-enhanced imaging according to claim 1, characterized in that: The calculation module is specifically used for: Calculate the average signal intensity of all signals in the region of interest in each frame of two-dimensional RF data; The average signal intensity of the region of interest in all frames of the two-dimensional RF data is expanded along the time direction and smoothed using a cubic spline interpolation function to obtain the time-intensity curve of contrast perfusion; According to the time-intensity curve, the characteristic parameters of the contrast agent arrival process are obtained.

5. The liver tumor lesion tracking and classification system based on ultrasound contrast-enhanced imaging according to claim 1, characterized in that: The training module is specifically used for: The time-intensity curve and the benign or malignant tumor judgment results are constructed as the training set of the neural network model; The training set is input into the neural network model for training. A 7-layer one-dimensional convolutional neural network is used to extract feature values, and the Sigmoid function is used as the activation function to accelerate training. Set the loss function of the neural network model and optimize the neural network model through the loss function; The above process is iterated until the training is completed to obtain a trained tumor classification model.

Images