Breast tissue characterization method and device, medium and product

Abstract

The invention discloses a mammary gland tissue characterization method and device, a medium and a product, and relates to the field of mammary gland tissue analys.The method comprises the steps that RF signals, envelope signals and B-mode signals corresponding to sections of a mammary gland cross section, a sagittal plane and a coronal plane are determined according to mammary gland three-dimensional data; generating a corresponding B-mode image according to the determined B-mode signal of each tangent plane; inputting the B mode images of all sections into an ROI detection model to obtain a tumor ROI region; the ROI detection model comprises a tumor B model edge detection model, a tracking module and an ROI generation module; based on the tumor ROI region, extracting RF signals, envelope signals and B-mode signals corresponding to the cross section, sagittal plane and coronal plane regions of interest from the mammary gland three-dimensional data; and inputting the extracted signal into a tumor tissue analysis model to obtain a tumor tissue characterization result. According to the invention, the searching efficiency and the positioning accuracy of the breast tumor can be improved, and the quantitative analysis of the tumor tissue characteristics is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of breast tissue analysis, in particular to a breast tissue localization method, device, medium and product. BACKGROUND

[0002] The evolution of breast cancer is a multi-stage and multi-step process. Most breast cancer patients are initially due to abnormal proliferation of cells in the gland and milk duct, and atypical hyperplasia occurs. This process belongs to the early formation process of breast cancer. If the tumor is not found in time, it will develop into carcinoma in situ. If the cancer cells are not treated in time, they will protrude from the duct wall and even metastasize through blood and lymph, gradually evolving into advanced breast cancer. Therefore, locating breast tumors and characterizing tumor tissues in the early stage can help assist clinical diagnosis and improve the survival rate and quality of life of breast cancer patients.

[0003] Currently, non-invasive methods for locating and characterizing breast tumors mainly include mammography, breast magnetic resonance imaging and breast ultrasound imaging. However, mammography has radiation hazards and low sensitivity to dense glands. Breast MRI imaging equipment is expensive, has high costs, a long training period for image physicians, and certain limitations for patients (for example, patients with metal implants in the body, claustrophobia, and allergy to contrast agents are not suitable). Ultrasound has the advantages of no ionizing radiation, low cost, simple operation, and accurate lesion positioning, and is widely used in the study of automatic positioning and tissue characterization of breast tumors.

[0004] However, there are two most important problems in traditional ultrasound for locating and characterizing breast tumors. First, the accurate position of the tumor needs to be determined by repeated observation by a clinician, and due to differences in experience between doctors and visual fatigue of the human eye, misdiagnosis and missed diagnosis may occur. Second, after collecting breast data, the analysis of the tissue characteristics of the breast tumor completely depends on the subjective experience of the clinician, and lacks objective quantitative analysis of the tumor. Therefore, how to improve the efficiency of breast tumor search and the accuracy of tumor positioning, and how to quantitatively analyze the tissue characteristics of the tumor are challenging and unsolved problems. SUMMARY

[0005] The purpose of the present application is to provide a breast tissue localization method, device, medium and product to improve the efficiency of breast tumor search and the accuracy of tumor positioning, and to realize quantitative analysis of the tissue characteristics of the tumor.

[0006] To achieve the above purpose, the present application provides the following solutions.

[0007] A breast tissue localization method, the method comprising the following steps.

[0008] Obtaining breast three-dimensional data.

[0009] determining signals of each section of the breast according to the three-dimensional data of the breast; the sections include transverse sections, sagittal sections and coronal sections; the signals include RF signals, envelope signals and B-mode signals.

[0010] generating a corresponding B-mode image according to the determined B-mode signal of each section.

[0011] inputting the B-mode images of all sections into an ROI detection model to obtain a tumor ROI region; wherein the ROI detection model includes a tumor B-mode edge detection model, a tracking module and an ROI generation module; the tumor B-mode edge detection model is used to input the B-mode images of all sections and output a tumor B-mode edge; the tracking module is used to track a tumor envelope tracking edge according to the input tumor B-mode edge, and track a tumor RF tracking edge according to the tumor envelope tracking edge; the ROI generation module is used to determine the tumor ROI region according to the tumor B-mode edge, the tumor envelope tracking edge and the tumor RF tracking edge.

[0012] extracting target signals from the three-dimensional data of the breast; the target signals include RF signals, envelope signals and B-mode signals corresponding to a transverse section of interest, a sagittal section of interest and a coronal section of interest respectively; wherein the transverse section of interest, the sagittal section of interest and the coronal section of interest are determined according to the tumor ROI region; the transverse section of interest includes a region of the tumor on the transverse section, the sagittal section of interest includes a region of the tumor on the sagittal section, and the coronal section of interest includes a region of the tumor on the coronal section.

[0013] inputting the target signals into a tumor tissue analysis model, and outputting a final tumor tissue characterization result from the tumor tissue analysis model; the tumor tissue characterization result includes tumor morphology, tumor aspect ratio, tumor edge, tumor internal echo, tumor rear echo characteristic, calcification characteristic, infiltrative characteristic and metastasis characteristic; wherein the calcification characteristic is used to represent whether there is calcification inside the tumor, the infiltrative characteristic is used to represent whether the tumor has infiltrative, and the metastasis characteristic is used to represent whether the tumor has metastasis.

[0014] wherein the tumor tissue analysis model includes a quantitative characterization module, a trained neural network and a trained systematic learning network; the quantitative characterization module is used to extract specific features of the tumor according to the target signals; the trained neural network is used to obtain a preliminary tumor tissue characterization result according to the target signals; and the trained systematic learning network is used to output the final tumor tissue characterization result according to the specific features and the preliminary tumor tissue characterization result.

[0015] Optionally, the method for determining the ROI detection model comprises: obtaining first training data and first labels; the first training data comprises B-mode images of each cross section of a breast determined from three-dimensional breast data as a sample; and the first labels comprise real tumor B-mode edges.

[0016] constructing a target model; the target model comprises a first tumor edge detection neural network, a tracking module, a tracing module, and an ROI generation module; an output of the first tumor edge detection neural network is connected with an input of the tracking module; outputs of the first tumor edge detection neural network, the tracking module, and the tracing module are all connected with the ROI generation module; the tracing module is used to trace a corresponding tumor envelope tracing edge according to an input tumor RF prediction edge, and trace a tumor B-mode tracing edge according to the obtained tumor envelope tracing edge; the tumor RF prediction edge is determined by a tumor RF edge detection model according to RF images corresponding to B-mode images in the training data.

[0017] training the first tumor edge detection neural network multiple times using the first training data and the first labels, with the goal of minimizing the total loss, to obtain a tumor B-mode edge detection model; the tumor envelope prediction edge is determined by a tumor envelope edge detection model according to envelope images corresponding to the B-mode images in the training data.

[0018] In any training process, the total loss is determined according to error values; the error values comprise errors between tumor B-mode edges output by the first tumor edge detection neural network obtained after a previous training and real tumor B-mode edges, errors between tumor envelope tracking edges generated by the tracking module and tumor envelope prediction edges, errors between tumor RF tracking edges generated by the tracking module and tumor RF prediction edges, errors between tumor envelope tracing edges generated by the tracing module and tumor envelope prediction edges, and errors between tumor B-mode tracing edges generated by the tracing module and real tumor B-mode edges.

[0019] connecting an output of the tumor B-mode edge detection model with an input of the tracking module, and connecting outputs of the tumor B-mode edge detection model and the tracking module with the ROI generation module, to obtain the ROI detection model.

[0020] Optionally, the error values further comprise errors between tumor B-mode edges output by the first tumor edge detection neural network obtained after a previous training and artificial feedback tumor edges; the artificial feedback tumor edges are determined according to artificial feedback information for the tumor ROI region.

[0021] Optionally, the quantification module comprises first to third quantification modules; wherein: the first quantification module is configured to extract a first specificity feature of the tumor according to the RF signal in the target signal.

[0022] The second quantification module is configured to extract a second specificity feature of the tumor according to the envelope signal in the target signal.

[0023] The third quantification module is configured to extract a third specificity feature of the tumor according to the B-mode signal in the target signal.

[0024] Optionally, the trained neural network comprises first to third trained neural networks; wherein: the first trained neural network is configured to obtain a preliminary tumor tissue characterization result corresponding to the RF signal according to the RF signal in the target signal.

[0025] The second trained neural network is configured to obtain a preliminary tumor tissue characterization result corresponding to the envelope signal according to the envelope signal in the target signal.

[0026] The third trained neural network is configured to obtain a preliminary tumor tissue characterization result corresponding to the B-mode signal according to the B-mode signal in the target signal.

[0027] Optionally, the trained system learning network comprises, in sequence, a feature screening module, a first layer of trained primary learners, and a second layer of trained meta-learners.

[0028] The feature screening module is configured to screen the specificity features and the preliminary tumor tissue characterization results using the mutual information method to obtain screening results.

[0029] The first layer of trained primary learners is configured to input the screening results and output primary tumor tissue characterization results.

[0030] The second layer of trained meta-learners is configured to input the primary tumor tissue characterization results and output final tumor tissue characterization results.

[0031] Optionally, determining the signals of each section of the breast comprises: extracting the RF signal of each section of the breast from the three-dimensional breast data.

[0032] The RF signal of each section is pre-amplified, the amplified signal is demodulated using an IQ demodulation algorithm, and the demodulated signal is detected using a Hilbert transform to obtain the envelope signal of each section of the breast.

[0033] The envelope signal of each section is logarithmically compressed to obtain the B-mode signal of each section of the breast.

[0034] Optionally, the breast three-dimensional data is acquired, specifically including: scanning the breast by using a multi-modal breast volume ultrasound all-in-one machine to obtain the breast three-dimensional data.

[0035] Optionally, target signals are extracted from the breast three-dimensional data, specifically including: receiving artificial calibration information for the tumor ROI region.

[0036] A final tumor ROI region is determined according to the artificial calibration information; the final tumor ROI region includes the transverse cross-section ROI, the sagittal cross-section ROI and the coronal cross-section ROI.

[0037] From the breast three-dimensional data, RF signals, envelope signals and B-mode signals corresponding to the transverse cross-section ROI, RF signals, envelope signals and B-mode signals corresponding to the sagittal cross-section ROI, and RF signals, envelope signals and B-mode signals corresponding to the coronal cross-section ROI are extracted.

[0038] Optionally, the first specificity feature includes a complexity-related feature, a texture-related feature, a frequency domain-related feature and a blood flow Doppler-related feature.

[0039] The second specificity feature includes a shape parameter and a scale parameter in a distribution model.

[0040] The third specificity feature includes an elastography-related parameter, a texture-related feature, a morphological-related feature and an image grayscale-related feature.

[0041] The present application also provides a computer device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the breast tissue characterization method.

[0042] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the breast tissue characterization method.

[0043] The present application also provides a computer program product, which includes a computer program, and the computer program is executed by a processor to realize the breast tissue characterization method.

[0044] According to the specific embodiments of the present application, the following technical effects are disclosed: according to the embodiments of the present application, the B-mode images of the breast in three cross sections of the transverse plane, the sagittal plane and the coronal plane are extracted based on the three-dimensional data of the breast, the ROI detection model is used to predict the tumor ROI region according to the B-mode images, compared with the method of the human eye judgment of the clinician, the searching efficiency and the positioning accuracy of the breast tumor are improved; and the specific features are extracted by using the quantification and characterization module in the tumor tissue analysis model, the signal quantification and characterization of the tumor ROI region corresponding to each cross section is realized, and the quantitative analysis of the tumor tissue characteristics is realized by combining the trained neural network and the trained systematic learning network. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 The flowchart of the breast tissue characterization method provided for the embodiment 1 of the present application.

[0046] Figure 2 The construction schematic diagram of the ROI detection model.

[0047] Figure 3 The training process schematic diagram of the ROI detection model.

[0048] Figure 4 The trace learning schematic diagram.

[0049] Figure 5 The interest area schematic diagram.

[0050] Figure 6 The interest area signal extraction schematic diagram.

[0051] Figure 7 The construction schematic diagram of the tumor tissue analysis model.

[0052] Figure 8 The systematic learning schematic diagram.

[0053] Figure 9 The specific application flowchart of the tissue characterization method.

[0054] Figure 10 The internal structure diagram of the computer device. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

[0056] The purpose of the present application is to provide a breast tissue characterization method, device, medium and product, which aims to improve the searching efficiency and positioning accuracy of the breast tumor and realize the quantitative analysis of the tumor tissue characteristics.

[0057] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easier to understand, the present application will be described in further detail below with reference to the accompanying drawings and specific embodiments.

[0058] Firstly, the overall idea of the breast tissue characterization method in the present embodiment is described.

[0059] The breast tissue characterization method in the present embodiment can realize automatic segmentation of a tumor region of interest by tracing back and feeding back learning of different sections of breast three-dimensional data, and realize breast tumor tissue characterization by systematic learning of the tumor region of interest. The method mainly involves two models, one is a ROI (Region of Interest) detection model, and the other is a tumor tissue analysis model.

[0060] Among them, the ROI detection model first acquires breast three-dimensional data through a multi-modal breast volume ultrasound all-in-one machine, then extracts the transverse section, sagittal section and coronal section of the collected breast three-dimensional data, obtains a radio frequency (RF) image, an envelope image and a B-mode image for each section, takes all the images as input, performs supervised tracing back learning according to artificial labeling, and outputs the corresponding tumor B-mode edge, tumor envelope edge and tumor RF edge to obtain the ROI detection model.

[0061] The tumor tissue analysis model first extracts the RF signal, envelope signal and B-mode signal corresponding to the three sections of the ROI obtained based on the ROI detection model. Each kind of extracted signal is quantitatively characterized to obtain different specific features, and then combined with a neural network as input, supervised systematic learning is performed according to artificial labeling to obtain the tumor tissue analysis model.

[0062] In actual application, the breast three-dimensional data to be characterized obtained is input into the above trained ROI detection model and tumor tissue analysis model to obtain the final tumor tissue characterization result.

[0063] The method can accurately characterize breast tumors, quantitatively find tumor positions, and analyze tumor tissues. The method is simple, easy to operate, and automatically scans, highlights the breast tumor region of interest and edge details, accurately analyzes the changes in tumor tissue structure acoustic characteristics, and provides a new method and theoretical basis for reducing the missed diagnosis and misdiagnosis rate in clinical practice.

[0064] The following describes several embodiments for the purpose of practical application.

[0065] Embodiment 1: As shown in the following table, the breast tissue characterization method in the present embodiment includes the following steps. Figure 1

[0066] ​Step 101: Obtain breast three-dimensional data.

[0067] In one example, a multi-modal breast volume ultrasound all-in-one machine is used to automatically scan the breast tumor to obtain the breast three-dimensional data. During the scanning process, a professional clinician is not required to operate to find the tumor. The probe is only required to be placed in the breast position for about 30 seconds to obtain the breast three-dimensional data.

[0068] In another example, in order to improve the frame frequency and improve the imaging quality, during the scanning of the multi-modal breast volume ultrasound all-in-one machine, in addition to using focused ultrasound, plane wave ultrasound is also combined for scanning, that is, a breast 3D volume focused / plane wave ultrasound is used to achieve the following. Specifically, the breast is first scanned once using focused ultrasound; then, a multi-angle plane wave coherent compounding imaging method is used to emit multiple plane waves with different deflection angles to the breast, the echo signals of each plane wave emission are collected for beam synthesis, and then coherent compounding is performed; through manual judgment, the breast three-dimensional data with the best effect output by the above two scans is taken as the final breast three-dimensional data. Subsequently, the RF image can be obtained by performing a series of signal processing operations on the breast three-dimensional data (i.e., breast RF signal), for example, the signal processing operations can include: envelope detection of B mode, autocorrelation processing of Color mode, and processing of a digital signal controller (DSC), etc.

[0069] Step 102: Determine the signal of each section of the breast according to the breast three-dimensional data, and generate a corresponding B-mode image according to the B-mode signal of each section determined.

[0070] In one example, the signal of each section of the breast is determined, specifically including: (1) extracting the RF signal of each section of the breast from the breast three-dimensional data.

[0071] (2) Pre-amplify the RF signal of each section, demodulate the amplified signal using an IQ demodulation algorithm, and detect the demodulated signal using Hilbert transform to obtain the envelope signal of each section of the breast. The calculation formula of Hilbert transform is as follows.

[0072]

[0073] In the formula, u is the demodulated signal; h represents the signal used for convolution; t is time, τ is the integral variable, P is the Cauchy principal value, and H(u)(t) is the analytic signal after Hilbert transform, that is, the envelope signal.

[0074] (3) Logarithmic compression is performed on the envelope signals of each section to obtain the B-mode signals of each section of the breast. The calculation formula of the logarithmic compression is as follows.

[0075] Y(t) = 20*log e (H(u)(t)) (2).

[0076] Y(t) represents the B-mode signal.

[0077] It can be seen that in this example, the obtained breast three-dimensional data is extracted into three sections of transverse section, sagittal section and coronal section, and one RF signal, one envelope signal and one B-mode signal are obtained based on each section, that is, three corresponding RF signals, three envelope signals and three B-mode signals are obtained for the three sections.

[0078] Subsequently, the corresponding RF image, envelope image and B-mode image can be generated based on the RF signal, envelope signal and B-mode signal of each section, that is, three corresponding RF images, three envelope images and three B-mode images are obtained for the three sections.

[0079] The relationship among the RF image, envelope image and B-mode image is as follows: (1) the RF image is an image generated by the RF signal without any processing; (2) the envelope image is an image generated by the envelope signal; (3) the B-mode image is an image generated by the B-mode signal.

[0080] Step 103: inputting the B-mode images of all sections into a ROI detection model to obtain a tumor ROI region.

[0081] The ROI detection model comprises a tumor B-mode edge detection model, a tracking module and an ROI generation module; the tumor B-mode edge detection model is configured to input the B-mode images of all sections and output a tumor B-mode edge; the tracking module is configured to track the tumor envelope tracking edge according to the input tumor B-mode edge, and track the tumor RF tracking edge according to the tumor envelope tracking edge; and the ROI generation module is configured to determine the tumor ROI region according to the tumor B-mode edge, the tumor envelope tracking edge and the tumor RF tracking edge.

[0082] In one example, the method for determining the ROI detection model specifically comprises:

[0083] (1) obtaining first training data and first labels; the first training data comprises B-mode images of sections of a breast determined according to breast three-dimensional data as a sample; and the first labels comprise true tumor B-mode edges.

[0084] (2) constructing a target model; the target model comprises: a first tumor edge detection neural network, a tracking module, a tracing module and an ROI generation module; the output of the first tumor edge detection neural network is connected with the input of the tracking module; the outputs of the first tumor edge detection neural network, the tracking module and the tracing module are all connected with the ROI generation module; the tracing module is used for tracing a corresponding tumor envelope tracing edge according to an input tumor RF prediction edge, and tracing a tumor B-mode tracing edge according to the obtained tumor envelope tracing edge; the tumor RF prediction edge is determined by a tumor RF edge detection model according to an RF image corresponding to a B-mode image in the training data. The first tumor edge detection neural network can comprise: a convolution layer, a pooling layer and a full connection layer.

[0085] (3) using the first training data and the first label to train the first tumor edge detection neural network multiple times with the goal of minimizing the total loss, to obtain a tumor B-mode edge detection model; the tumor envelope prediction edge is determined by a tumor envelope edge detection model according to an envelope image corresponding to a B-mode image in the training data.

[0086] Wherein, the total loss in each training process is determined according to error values; the error values comprise: the error between the tumor B-mode edge output by the first tumor edge detection neural network obtained after the previous training and the real tumor B-mode edge, the error between the tumor envelope tracking edge generated by the tracking module and the tumor envelope prediction edge, the error between the tumor RF tracking edge generated by the tracking module and the tumor RF prediction edge, the error between the tumor envelope tracing edge generated by the tracing module and the tumor envelope prediction edge, and the error between the tumor B-mode tracing edge generated by the tracing module and the real tumor B-mode edge.

[0087] The output of the tumor B-mode edge detection model is connected with the input of the tracking module, and the outputs of the tumor B-mode edge detection model and the tracking module are both connected with the ROI generation module, to obtain the ROI detection model.

[0088] The determination method of the tumor RF edge detection model used above specifically comprises:

[0089] Obtaining second training data and second labels; the second training data comprises: RF images of each cross section of a breast determined according to breast three-dimensional data as samples; the second labels comprise: real tumor RF edges.

[0090] The second tumor edge detection neural network is trained multiple times using the second training data and the second label, with the objective of minimizing RF edge loss, to obtain a tumor RF edge detection model. The second tumor edge detection neural network can have the same network structure as the first tumor edge detection neural network.

[0091] The RF edge loss is determined according to an error between a tumor RF predicted edge output by the tumor edge detection neural network and a real tumor RF edge.

[0092] The determination method of the tumor envelope edge detection model includes the following steps.

[0093] Third training data and a third label are obtained. The third training data includes envelope images of each section of the breast determined according to breast three-dimensional data as a sample. The third label includes a real tumor envelope edge.

[0094] A third tumor edge detection neural network is trained multiple times using the third training data and the third label, with the objective of minimizing envelope edge loss, to obtain a tumor envelope edge detection model. The third tumor edge detection neural network can have the same network structure as the first tumor edge detection neural network.

[0095] The envelope edge loss is determined according to an error between a tumor envelope predicted edge output by the tumor edge detection neural network and a real tumor envelope edge.

[0096] In another example, the error value further includes an error between a tumor B-mode edge output by the first tumor edge detection neural network obtained after the previous training and an artificial feedback tumor edge. The artificial feedback tumor edge is determined according to artificial feedback information for the tumor ROI region.

[0097] The following will be described in detail Figure 2 and Figure 3 The training process of the ROI detection model will be further described.

[0098] The B-mode images of the three sections, the envelope images of the three sections, and the RF images of the three sections are taken as inputs, and supervised retroactive learning is performed in combination with artificial calibration, as shown in Figure 2 and Figure 3 The B-mode images of the three sections are input into the convolutional layer, the pooling layer, and the fully connected layer of the first tumor edge detection neural network to obtain a tumor B-mode edge. Based on the tumor B-mode edge, the tumor envelope predicted edge output by the trained second tumor edge detection neural network, and the tumor RF predicted edge output by the trained third tumor edge detection neural network, the tumor envelope edge and the tumor RF edge are obtained through retroactive learning. Subsequently, the tumor ROI region is determined through the ROI generation module.

[0099] The network results of the first, second, and third tumor edge detection neural networks are the same. Taking the first tumor edge detection neural network as an example, it includes n convolutional layers (Conv1, Conv2, ..., Conv-n), n pooling layers (Pool1, Pool2, ..., Pool-n), and a fully connected layer. The connection relationships of these layers are as follows: Figure 3 As shown.

[0100] Tracing learning comprises two parts: tracking and tracing. Tracking refers to the forward propagation of information in a signal from top to bottom, following a specific pattern or source of information. For example, based on the tumor B-model edge, the tumor envelope edge is matched with the tumor envelope edge to obtain the tumor envelope tracking edge, and then the tumor envelope tracking edge is matched with the tumor RF edge to obtain the tumor RF tracking edge. Tracing, on the other hand, refers to the reverse propagation of information in a signal from bottom to top, i.e., the ability to cyclically feedback information. For example, from the tumor RF edge, the tumor envelope edge is matched with the tumor envelope edge to obtain the tumor envelope tracing edge, and then the tumor envelope tracing edge is matched with the tumor B-model edge to obtain the tumor B-model tracing edge. Figure 4 As shown. The iterative process in retrospective learning can be represented by formula (3).

[0101] Q2(s,a)←Q1(s,a)+α[R+γmax a' Q(s',a')-Q(s,a)](3).

[0102] In the formula, Q2(s,a) represents the new value of signal a in state s, Q1(s,a) represents the original value of signal a in state s, R is the reward obtained based on a, and max a' Q(s',a') is the maximum value of all signals in the next state s', where α and γ are parameters, 0≤α≤1, 0≤γ≤1.

[0103] by Figure 4 For example, let formula (3) represent the iterative process in the two tracing processes: tracing from the tumor RF edge to the tumor envelope edge, and tracing from the tumor envelope edge to the tumor B-model edge. When tracing from the tumor RF edge to the tumor envelope edge, Q2(s,a) is the accurate edge of the tumor envelope, i.e., the label of the tumor envelope edge; Q1(s,a) is the tumor RF edge before adjustment; Q(s',a') is the tumor envelope edge obtained after one adjustment of Q1(s,a); and Q(s,a) is the tumor envelope edge adjusted by Q(s',a') in the previous step. The formula for tracing from the tumor envelope edge to the tumor B-model edge is similar to formula (3) and will not be repeated here.

[0104] Next, the tumor edge obtained using the 3 sections corresponding to 9 images is used to locate the tumor, and finally the obtained tumor region and surrounding infiltrated tissue are taken as the interest region to obtain the ROI detection model, as shown in Figure 5 The obtained tumor location is fed back as the label in the retrospective learning process based on the B-mode image artificial calibration by the clinician to improve the accuracy of the edge detection result. The transfer function describing the relationship between the input and output in the feedback algorithm is shown in equation (4).

[0105] G(s) = H(s) / (1 + H(s) * F(s)) (4).

[0106] Corresponding Figure 3 In equation (4), G(s) is the transfer function of the feedback link, representing the ROI detection model; H(s) is the transfer function of the feedback path, representing the feedback of the clinician on the tumor edge; and F(s) is the transfer function of the forward path, representing the tumor ROI region obtained through retrospective learning.

[0107] Step 104: Extracting target signals from the breast three-dimensional data.

[0108] The target signals include RF signals, envelope signals and B-mode signals corresponding to the transverse plane interest region, the sagittal plane interest region and the coronal plane interest region respectively; the transverse plane interest region, the sagittal plane interest region and the coronal plane interest region are determined according to the tumor ROI region; the transverse plane interest region includes the region of the tumor on the transverse plane, the sagittal plane interest region includes the region of the tumor on the sagittal plane, and the coronal plane interest region includes the region of the tumor on the coronal plane.

[0109] In one example, the above step 104 specifically includes:

[0110] Receiving artificial calibration information for the tumor ROI region.

[0111] Determining a final tumor ROI region according to the artificial calibration information; the final tumor ROI region includes the transverse plane interest region, the sagittal plane interest region and the coronal plane interest region.

[0112] From the breast three-dimensional data, extracting RF signals, envelope signals and B-mode signals corresponding to the transverse plane interest region, RF signals, envelope signals and B-mode signals corresponding to the sagittal plane interest region, and RF signals, envelope signals and B-mode signals corresponding to the coronal plane interest region.

[0113] Specifically, after obtaining the ROI model, the RF signals, envelope signals and B-mode signals corresponding to the transverse plane, the sagittal plane and the coronal plane are extracted based on the interest region, a total of 9 signals are extracted. As shown inFigure 6 As shown in one of the tumor region of interest extraction schematic diagram, first, the RF data of the transverse, sagittal and coronal sections of the three-dimensional region of interest is obtained by projection method, then according to the RF data of the three sections, the envelope signal of the three sections is obtained by pre-gain and demodulation, and then the B-mode signal of the three sections is obtained by logarithmic compression of the envelope signal, that is, a total of 9 signals are extracted, and the 9 signals are quantitatively characterized and learned by neural network.

[0114] Step 105: inputting the target signal into the tumor tissue analysis model, and outputting the final tumor tissue characterization result from the tumor tissue analysis model.

[0115] The tumor tissue characterization result includes tumor morphology, tumor aspect ratio, tumor edge, tumor internal echo, tumor rear echo characteristic, calcification characteristic, infiltrative characteristic and metastasis characteristic; wherein the calcification characteristic is used to represent whether there is calcification inside the tumor; the infiltrative characteristic is used to represent whether the tumor has infiltrative property, and the metastasis characteristic is used to represent whether the tumor has metastasis.

[0116] The tumor tissue analysis model includes a quantitative characterization module, a trained neural network and a trained systematic learning network; the quantitative characterization module is used to extract specific features of the tumor according to the target signal; the trained neural network is used to obtain a preliminary tumor tissue characterization result according to the target signal; and the trained systematic learning network is used to output the final tumor tissue characterization result according to the specific features and the preliminary tumor tissue characterization result.

[0117] In one example, the quantitative characterization module includes first to third quantitative characterization modules. Wherein:

[0118] The first quantitative characterization module is used to extract the first specific feature of the tumor according to the RF signal in the target signal. The first specific feature may exemplarily include complexity-related features, texture-related features, frequency domain-related features and blood flow Doppler-related features.

[0119] The second quantitative characterization module is used to extract the second specific feature of the tumor according to the envelope signal in the target signal. The second specific feature may exemplarily include shape parameters and scale parameters in the distribution model.

[0120] The third quantitative characterization module is used to extract the third specific feature of the tumor according to the B-mode signal in the target signal. The third specific feature may exemplarily include elasticity imaging-related parameters, texture-related features, morphological-related features and image gray scale-related features.

[0121] The trained neural network includes first to third trained neural networks; wherein:

[0122] The first trained neural network is used to obtain a preliminary tumor tissue characterization result corresponding to the RF signal according to the RF signal in the target signal. The second trained neural network is used to obtain a preliminary tumor tissue characterization result corresponding to the envelope signal according to the envelope signal in the target signal. The third trained neural network is used to obtain a preliminary tumor tissue characterization result corresponding to the B-mode signal according to the B-mode signal in the target signal.

[0123] The trained systematized learning network specifically comprises: a feature screening module, a first layer of trained primary learners, and a second layer of trained meta-learners connected in sequence. The fitting degree of the first layer of primary learners is greater than that of the second layer of meta-learners; and the second layer of meta-learners is a generalized linear model.

[0124] The feature screening module is used to screen the input specific features and preliminary tumor tissue characterization results respectively by using the mutual information method to obtain screening results. The first layer of trained primary learners is used to input the screening results and output primary tumor tissue characterization results. The second layer of trained meta-learners is used to input the primary tumor tissue characterization results and output final tumor tissue characterization results.

[0125] The training process of the tumor tissue analysis model is as shown in Figure 7 .

[0126] Next, the first specific feature and the first trained neural network are described in detail.

[0127] (1) The complexity-related features include: entropy, Rayleigh entropy, cross-entropy, and cumulative residual entropy. The calculation formula of the entropy H(Y) is as formula (5), the calculation formula of the Rayleigh entropy R(Y) is as formula (6), the calculation formula of the cross-entropy H(p,q) is as formula (7), and the calculation formula of the cumulative residual entropy ε(Y) is as formula (8). c β

[0128]

[0129] In the formula, Y represents the number of intervals for dividing the amplitude range of the radio frequency signal; w(y) represents the statistical histogram of the radio frequency signal; β>0, β≠1; p is the probability distribution of the true feature; q is the probability distribution of the non-true feature; and W represents the cumulative probability distribution of the radio frequency signal.

[0130] (2) The texture-related features include: energy, correlation, and homogeneity features obtained based on a gray level co-occurrence matrix, roughness, contrast, and directionality obtained based on Tamura texture feature extraction, and spatial domain features extracted based on mathematical morphology.

[0131] ​​(3) Frequency domain related features include: energy concentration, frequency peak, spectral width, frequency domain coherence and scatterer spacing.

[0132] (4) Blood flow Doppler related features include: whether the blood flow is rich, blood flow velocity. On the one hand, input the above-mentioned features into the neural network, and the signals in the three directions and their labels (i.e. the real tumor tissue localization results) are used as a training set for supervised learning. After the model is trained, it is used as the first trained neural network, and the RF signals in different sections are input into the first trained neural network.

[0133] Next, the second specificity feature and the second trained neural network are described in detail.

[0134] For the envelope signals corresponding to the three sections, on the one hand, the shape parameter m and the scale parameter Ω in the Nakagami distribution are extracted, as shown in formula (9), and the aggregation parameter mu and the derivation parameter As a parameter feature, as shown in formula (10), on the other hand, input the above-mentioned features into the neural network, and the signals in the three directions and their labels (i.e. the real tumor tissue localization results) are used as a training set for supervised learning. After the model is trained, it is used as the second trained neural network, and the envelope signals in different sections are input into the second trained neural network.

[0135]

[0136]

[0137] In the formula, E represents the envelope of the backscattering signal; Γ(m) represents the Euler Gamma equation; ε 2 and σ 2 represent the coherent component and incoherent component in the signal respectively; f(E) represents the probability density function of R; U(E) represents the standard uniform distribution function; P k (A|ε,σ 2 , mu) represents the n-dimensional HK distribution; m represents the shape parameter; J represents the zero-order first-class Bessel function; mμ represents the aggregation parameter; A represents the envelope amplitude of the backscattering signal.

[0138] Next, the third specificity feature and the third trained neural network are described in detail.

[0139] For the B-mode signals corresponding to the three sections, on the one hand, the elasticity imaging related parameters, texture related features, morphological related features, and image gray related features are extracted; the specific features contained in these categories are listed in detail as follows: (1) the elasticity imaging related parameters are mainly the Young's modulus value E, as shown in formula (11); (2) the texture related features are the same as the texture features extracted from the RF signals; (3) the morphological related features mainly include tumor size, tumor shape, and tumor texture; and (4) the image gray related features mainly include gray mean, gray variance, gray energy, and gray entropy. On the other hand, the signals in the three section directions and their labels (i.e., the real tumor tissue characterization results) are input into the neural network, and the signals are used as a training set for supervised learning. After the model is trained, it is used as a third trained neural network, and the B-mode signals of different sections are input into the third trained neural network.

[0140] E = χ / η = (F / S) / (ΔL / L0) = FL0 / SΔL (11).

[0141] In the formula, χ is the uniaxial stress; η is the strain; S is the cross-sectional area; F is the compression force or stretching force; ΔL is the length change; and L0 is the original length.

[0142] Next, the systematic learning network described above is introduced in detail.

[0143] The quantification characterization part and the neural network learning part of the 9 signals are input, and the tissue characteristics (tumor morphology, tumor aspect ratio, tumor edge, tumor internal echo, tumor rear echo characteristics, presence or absence of calcification in the tumor, whether it is infiltrative, whether there is metastasis, etc.) of different tumors are artificially labeled as labels by a clinician, and a systematic learning is performed to obtain a final tumor tissue analysis model.

[0144] The systematic learning network is as shown in Figure 8 As shown in formula (12), the mutual information method is used to perform feature screening on all the specific features extracted from all the signals and the preliminary tumor tissue characterization results output by the neural network. Next, the stacking ensemble learning method is used to first construct a first layer of primary learners, and the first layer of primary learners includes c primary learners, which are primary learner 1, primary learner 2,..., and primary learner c. Then, a second layer of meta-learners is constructed, and the results of the first layer of primary learners are used as the input of the second layer of meta-learners, to form a final systematic learning network.

[0145] The c primary learners in the first layer of primary learners use a model with high fitting degree, such as a neural network, a support vector machine, etc., to pursue sufficient learning of the training data and automatically extract effective features; the second layer of meta-learners uses a simple model, such as a logistic regression, a Lasso regression, etc. general linear model, to reduce the risk of overfitting, and outputs the tumor tissue analysis result, which quantifies the degree of tumor malignancy by judging whether the tumor is invasive and whether it is metastatic.

[0146]

[0147] In the formula, X represents a feature set; Z represents a category set; I(X, Z) represents mutual information, which is used for feature screening; p(x, z) represents the probability of a feature being x and a category being z; p(x) represents the probability of a feature being x; and p(z) represents the probability of a feature being z.

[0148] The following will be described in combination with Figure 9 The specific application process of the breast tissue characterization method will be described.

[0149] Specifically, first, a multi-modal breast volume ultrasound all-in-one machine is used to perform focused / planar wave ultrasound scanning on the breast to obtain breast three-dimensional data. Then, the RF image, envelope image and B-mode image corresponding to the transverse plane, sagittal plane and coronal plane of the breast extracted from the breast three-dimensional data are input into the trained ROI detection model as input, and the tumor region and surrounding tissue are extracted as the tumor ROI region, and then the extracted tumor ROI region is discriminated by a clinician.

[0150] Next, the tumor needs to be characterized. Based on the tumor ROI region, the RF signal, envelope signal and B-mode signal corresponding to the three section directions are extracted, and these signals are simultaneously input into the trained tumor tissue analysis model to obtain the breast tumor tissue characterization result. The tumor tissue characterization result is mainly used to describe the differences in acoustic characteristics between different tumors and analyze the tissue structure of different tumors.

[0151] The above embodiment specifically introduces the training of the tissue characterization model and the actual application of the tissue characterization method.

[0152] Among them, the training of the tissue characterization model is divided into the training of the ROI detection model and the training of the tumor tissue analysis model. The ROI detection model mainly automatically extracts the tumor region and surrounding tissue by tracing learning; and the tumor tissue analysis model mainly extracts the RF signal, envelope signal and B-mode signal of the three section directions based on the tumor ROI region, quantitatively characterizes and neural network learns all signals, and finally obtains the specific features extracted by quantitatively characterizing all signals and the results after neural network learning through systematic learning.

[0153] The tissue characterization method is applied again, mainly by inputting the collected data into the trained ROI detection model to obtain the tumor ROI region, which greatly reduces the dependence on the experience of the operator, improves the efficiency of tumor searching and positioning, and relieves the demand for ultrasound doctors. Based on the tumor ROI region, the extracted signal is input into the tumor tissue analysis model to obtain the final breast tumor tissue characterization result, so as to objectively and quantitatively analyze the breast tumor tissue, which has very important research value and application prospect for early screening, diagnosis and treatment in clinic.

[0154] The advantages of the breast tissue characterization method of the above embodiment are described in detail below.

[0155] Due to the long scanning time and low efficiency of using traditional two-dimensional ultrasound for breast tumor positioning and tissue characterization, a large number of ultrasound doctors need to spend a lot of time to do this work, and the ultrasound imaging quality obtained by scanning (i.e. the accuracy of the position of the tumor and the result of the tissue characterization) depends on the subjective experience of the clinician. Although the recently launched breast volume ultrasound has solved some problems, due to the loss of part of the information in the generation of B-ultrasound and the fact that the two-dimensional image quality (transverse and sagittal) is not necessarily better than that of traditional two-dimensional ultrasound, the delineation of the position of the breast tumor and the tissue characterization of the tumor tissue still depends on the subjective experience of the clinical ultrasound doctor, lacking accurate quantitative and objective description. Although some methods for extracting and characterizing the region of interest based on the quantitative characteristics of ultrasound have been proposed, there are still certain limitations in using only one or a few characteristics and methods to analyze the tumor tissue.

[0156] The embodiment uses a multimodal breast volume ultrasound all-in-one machine to perform focused / planar wave ultrasound scanning on the breast, locates and edge detects the tumor in the breast through a trace learning network, adjusts the extracted edge using the feedback of the clinician, quantitatively characterizes and neural network learns different signals corresponding to different section directions, and systematically learns the two methods of all section corresponding to all signals, fully utilizes the advantages of each section and each signal, effectively solves the problems of long scanning time, strong subjectivity, insufficient image quality and incomplete information, realizes automatic delineation of the tumor region and tissue characterization of the tumor, greatly reduces the dependence on the experience of the operator and the ultrasound doctor, reduces the workload of the ultrasound doctor, improves the accuracy and efficiency of the tumor region delineation, and provides a new method for tumor tissue characterization. The method has the advantages of automatic and rapid scanning, automatic tumor delineation, automatic analysis of tumor tissue, and saving of medical resources.

[0157] Embodiment 2

[0158] A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the computer program to implement the breast tissue mapping method in embodiment 1.

[0159] Embodiment 3

[0160] A computer readable storage medium having stored thereon a computer program, the computer program being executed by a processor to implement the breast tissue mapping method in embodiment 1.

[0161] Embodiment 4

[0162] A computer program product comprising a computer program, the computer program being executed by a processor to implement the breast tissue mapping method in embodiment 1.

[0163] Embodiment 5

[0164] A computer device, the internal structure diagram of which can be as shown in Figure 10 The computer device comprises a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is used to store to-be-processed transactions. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the breast tissue mapping method in embodiment 1.

[0165] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, etc., without being limited thereto.

[0166] Any combination of the technical features of the above embodiments can be made, and in order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0167] The principles and implementation modes of the present application are described by applying specific examples herein, and the above-mentioned embodiments are only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In view of the above, the content of the present application should not be understood as a limitation.

Claims

1. A method for characterizing breast tissue, characterized in that, The method includes: Obtain three-dimensional data of the breast; Based on three-dimensional breast data, the signals of each section of the breast are determined; the sections include transverse, sagittal, and coronal planes; the signals include RF signals, envelope signals, and B-mode signals. Based on the B-mode signal of each determined cross section, generate the corresponding B-mode image; The B-model images of all sections are input into the ROI detection model to obtain the tumor ROI region. The ROI detection model includes a tumor B-model edge detection model, a tracking module, and an ROI generation module. The tumor B-model edge detection model is used to input the B-model images of all sections and output the tumor B-model edges. The tracking module is used to obtain the tumor envelope tracking edge based on the input tumor B-model edges, and to obtain the tumor radiofrequency (RF) tracking edge based on the tumor envelope tracking edge. The ROI generation module is used to determine the tumor ROI region based on the tumor B-model edges, the tumor envelope tracking edge, and the tumor RF tracking edge. Target signals are extracted from three-dimensional breast data. These target signals include RF signals, envelope signals, and B-mode signals corresponding to the transverse, sagittal, and coronal regions of interest (ROIs). The transverse, sagittal, and coronal ROIs are determined based on the tumor region of interest. The transverse ROI includes the area of ​​the tumor on the transverse plane, the sagittal ROI includes the area of ​​the tumor on the sagittal plane, and the coronal ROI includes the area of ​​the tumor on the coronal plane. The target signal is input into a tumor tissue analysis model, which outputs the final tumor tissue characterization results. The tumor tissue characterization results include: tumor morphology, tumor aspect ratio, tumor margin, internal echogenicity of the tumor, posterior echogenicity of the tumor, calcification characteristics, invasive characteristics, and metastatic characteristics. Among them, calcification characteristics are used to characterize whether there is calcification inside the tumor; invasive characteristics are used to characterize whether the tumor is invasive; and metastatic characteristics are used to characterize whether the tumor has metastasized. The tumor tissue analysis model includes: a quantitative characterization module, a trained neural network, and a trained systematic learning network; the quantitative characterization module is used to extract specific features of the tumor based on the target signal; the trained neural network is used to obtain preliminary tumor tissue characterization results based on the target signal; and the trained systematic learning network is used to output the final tumor tissue characterization results based on the specific features and the preliminary tumor tissue characterization results.

2. The method for characterizing breast tissue according to claim 1, characterized in that, The method for determining the ROI detection model specifically includes: Acquire first training data and first labels; the first training data includes: B-model images of various sections of the breast determined based on three-dimensional breast data as samples; the first label includes: the edge of the actual tumor B-model; A target model is constructed, comprising: a first tumor edge detection neural network, a tracking module, a tracing module, and an ROI generation module; the output of the first tumor edge detection neural network is connected to the input of the tracking module; the outputs of the first tumor edge detection neural network, the tracking module, and the tracing module are all connected to the ROI generation module; the tracing module is used to trace the corresponding tumor envelope tracing edge based on the input tumor RF predicted edge, and to trace the tumor B-model tracing edge based on the obtained tumor envelope tracing edge; the tumor RF predicted edge is determined by the tumor RF edge detection model based on the RF image corresponding to the B-model image in the training data; Using the first training data and the first label, the first tumor edge detection neural network is trained multiple times with the goal of minimizing the total loss to obtain a tumor B-mode edge detection model; the tumor envelope predicted edge is determined by the tumor envelope edge detection model based on the envelope image corresponding to the B-mode image in the training data; The total loss in any training process is determined based on the error value; the error value includes: the error between the tumor B-model edge output by the first tumor edge detection neural network obtained after the previous training and the real tumor B-model edge, the error between the tumor envelope tracking edge generated by the tracking module and the tumor envelope prediction edge, the error between the tumor RF tracking edge generated by the tracking module and the tumor RF prediction edge, the error between the tumor envelope tracing edge generated by the tracing module and the tumor envelope prediction edge, and the error between the tumor B-model tracing edge generated by the tracing module and the real tumor B-model edge; The output of the tumor B-mode edge detection model is connected to the input of the tracking module, and the outputs of both the tumor B-mode edge detection model and the tracking module are connected to the ROI generation module to obtain the ROI detection model.

3. The method for characterizing breast tissue according to claim 2, characterized in that, The error value also includes the error between the tumor B-model edge output by the first tumor edge detection neural network obtained after the previous training and the manually fed-in tumor edge; the manually fed-in tumor edge is determined based on the manually fed-in information for the tumor ROI region.

4. The method for characterizing breast tissue according to claim 1, characterized in that, The quantization and characterization module includes first to third quantization and characterization modules; wherein: The first quantitative characterization module is used to extract the first specific feature of the tumor based on the RF signal in the target signal; The second quantitative characterization module is used to extract the second specific feature of the tumor based on the envelope signal in the target signal; The third quantization and characterization module is used to extract the third specific feature of the tumor based on the B-mode signal in the target signal.

5. The method for characterizing breast tissue according to claim 1, characterized in that, The trained neural networks include the first through third trained neural networks; among which: The first trained neural network is used to obtain preliminary tumor tissue characterization results corresponding to the RF signal in the target signal; The second trained neural network is used to obtain preliminary tumor tissue characterization results corresponding to the envelope signal based on the envelope signal in the target signal; The third trained neural network is used to obtain preliminary tumor tissue characterization results corresponding to the B-mode signal based on the B-mode signal in the target signal.

6. The method for characterizing breast tissue according to claim 1, characterized in that, The trained systematic learning network specifically includes: a feature selection module connected in sequence, a trained first-layer primary learner, and a trained second-layer meta-learner; The feature filtering module is used to filter the input specific features and preliminary tumor tissue characterization results using the mutual information method to obtain the filtering results. The trained first-layer primary learner is used to input the screening results and output the primary tumor tissue characterization results; The trained second-layer meta-learner is used to input the primary tumor tissue characterization results and output the final tumor tissue characterization results.

7. The method for characterizing breast tissue according to claim 1, characterized in that, The signals for different sections of the breast are determined, specifically including: RF signals of each section of the breast were extracted from the three-dimensional breast data; The RF signals of each section are pre-gain amplified, the amplified signals are demodulated using the IQ demodulation algorithm, and the demodulated signals are detected using Hilbert transform to obtain the envelope signal of each section of the breast. Logarithmic compression was performed on the envelope signals of each section to obtain the B-mode signal of each section of the breast.

8. The method for characterizing breast tissue according to claim 1, characterized in that, Obtaining three-dimensional breast data specifically includes: The breast was scanned using a multimodal breast volume ultrasound machine to obtain three-dimensional breast data.

9. The method for characterizing breast tissue according to claim 1, characterized in that, Extracting target signals from three-dimensional breast data specifically includes: Receive manually calibrated information for the tumor ROI region; The final tumor ROI region is determined based on the artificially calibrated information; the final tumor ROI region includes the transverse region of interest, the sagittal region of interest, and the coronal region of interest; From the three-dimensional breast data, the RF signal, envelope signal, and B-mode signal corresponding to the transverse region of interest, the RF signal, envelope signal, and B-mode signal corresponding to the sagittal region of interest, and the RF signal, envelope signal, and B-mode signal corresponding to the coronal region of interest are extracted.

10. The method for characterizing breast tissue according to claim 4, characterized in that, The first specific features include: complexity-related features, texture-related features, frequency domain-related features, and blood flow Doppler-related features; The second specific feature includes: shape parameters and scale parameters in the distribution model; The third specific feature includes: elastography-related parameters, texture-related features, morphology-related features, and image grayscale-related features.

11. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the breast tissue characterization method according to any one of claims 1-10.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the breast tissue characterization method according to any one of claims 1-10.

13. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the breast tissue characterization method according to any one of claims 1-10.

Images