Machine learning based prostate lesion identification method, system, device, and medium

Abstract

The application provides a prostate lesion identification method, system, device and medium based on machine learning, comprising: acquiring historical mp-MRI images and intraoperative TRUS images of a patient; acquiring lesion spatial information of an MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images; performing puncture on the prostate of the patient based on the lesion spatial information to acquire pathological information of a puncture point; performing spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs; training a lesion identification model based on the pathological-image feature pairs to perform prostate lesion identification based on mp-MRI images through the trained lesion identification model. The application locates suspicious areas of prostate lesions with the aid of fusion images of historical mp-MRI images and intraoperative TRUS images, and performs machine learning on the historical mp-MRI images positioned and labeled according to the pathological information of the puncture point, so as to quickly and accurately identify prostate lesions.

Description

Technical Field

[0001] This application belongs to the field of computer technology, and in particular relates to a method, system, device and medium for identifying prostate lesions based on machine learning. Background Technology

[0002] Unlike other solid tumors, prostate cancer lesions are multifocal and sporadic, making lesion identification and localization difficult. Currently, multiparametric magnetic resonance imaging (MRI) is the most common and effective imaging examination for diagnosing prostate cancer; however, its accuracy is only 75.25%, and its specificity is only 68.22%.

[0003] Artificial intelligence (AI) is an emerging technological science that uses computers and related technologies to simulate, extend, and expand human intelligence. It has developed rapidly in recent years and is now widely applied in all aspects of work and life. In recent years, with the further improvement of computer data processing capabilities and the gradual deepening of interdisciplinary research in medicine and engineering, AI has shown outstanding performance in the medical field, especially in image recognition and data recognition, including the reconstruction, processing (segmentation, denoising), analysis, and model prediction of various medical images. Machine learning (ML), as a branch of AI, can guide computers to learn from data through different algorithms and continuously improve its efficiency based on learning experience. Currently, the combination of ML and radiomics is widely used in the medical field to mine and extract massive amounts of image data, thereby achieving automatic segmentation, feature extraction, and model building of medical images, ultimately making the most accurate diagnoses based on various types of image data.

[0004] Therefore, how to achieve rapid and accurate identification of prostate lesions based on machine learning is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This application provides a machine learning-based method, system, device, and medium for identifying prostate lesions, which can solve the technical problem of how to achieve fast and accurate identification of prostate lesions based on machine learning.

[0006] In a first aspect, this application provides a machine learning-based method for identifying prostate lesions, the method comprising:

[0007] Acquire the patient's historical mp-MRI images and intraoperative TRUS images;

[0008] Based on the historical mp-MRI images and the intraoperative TRUS images, obtain the lesion spatial information of the MRI-US fused images;

[0009] Based on the spatial information of the lesion, the patient's prostate is punctured to obtain pathological information at the puncture point;

[0010] Based on the pathological information of the puncture point, spatial mapping is performed on the historical mp-MRI images to obtain pathological-image feature pairs;

[0011] The lesion recognition model is trained based on the pathological-image features, and the trained lesion recognition model is used to identify prostate lesions based on mp-MRI images.

[0012] In one implementation of the first aspect, obtaining lesion spatial information of the MRI-US fusion image based on the historical mp-MRI image and the TRUS image includes:

[0013] The historical mp-MRI images and the TRUS images are segmented based on a prostate segmentation network to obtain a first mask image and a second mask image respectively.

[0014] Optimal transformation parameters are obtained based on the first mask image and the second mask image, and the first mask image is mapped onto the second mask image based on the optimal transformation parameters to obtain the lesion spatial information of the MRI-US fusion image.

[0015] In one implementation of the first aspect, obtaining the optimal transform parameters based on the first mask image and the second mask image includes:

[0016] Resample the first mask image and the second mask image;

[0017] First registration data and second registration data are obtained based on the resampled first mask image and second mask image, respectively, and coarse registration is performed based on the first registration data and second registration data; the first registration data includes a first center point and a first principal axis corresponding to the first mask image, and the second registration data includes a second center point and a second principal axis corresponding to the second mask image;

[0018] The first and second mask images after coarse registration are subjected to nearest point matching, and the transformation parameters are iteratively calculated based on the nearest point matching results until the nearest point matching results meet the error threshold to obtain the optimal transformation parameters; the optimal transformation parameters include rotation parameters, translation parameters and local deformation parameters.

[0019] In one implementation of the first aspect, spatial mapping is performed on the historical mp-MRI images based on the pathological information of the puncture point to obtain pathological-image feature pairs, including:

[0020] The puncture point is located in the T2W sequence image based on its physical coordinates, and the radius of the puncture point is determined based on the physical information of the puncture needle.

[0021] The puncture point region in the T2W sequence image is determined based on the puncture point and the radius of the puncture point.

[0022] The puncture point regions of the DWI sequence image and the puncture point regions of the ADC sequence image are determined based on the puncture point regions in the T2W sequence image.

[0023] Based on the puncture point region and pathological information in the T2W sequence image, the puncture point region and pathological information in the DWI sequence image, and the puncture point region and pathological information in the ADC sequence image, T2W pathological-image feature pairs, DWI pathological-image feature pairs, and ADC pathological-image feature pairs are obtained respectively.

[0024] In one implementation of the first aspect, training the lesion recognition model based on the pathological-image features includes:

[0025] The T2W pathological-image feature pair, the DWI pathological-image feature pair, and the ADC pathological-image feature pair are input into the lesion recognition model to extract T2W channel features, DWI channel features, and ADC channel features based on the lesion recognition model.

[0026] Based on the DWI channel features and / or the ADC channel features, obtain the mapping loss of the puncture point region and the weighted loss of the T2W channel features;

[0027] The lesion identification model is trained based on the mapping loss and the weighted loss until it converges.

[0028] In one implementation of the first aspect, the method includes:

[0029] The weighted loss for obtaining T2W channel features based on the DWI channel features and / or the ADC channel features includes:

[0030] Obtain mapping features based on the DWI channel features and / or the ADC channel features;

[0031] The T2W channel features are weighted based on the mapping features to obtain the weighted T2W channel features.

[0032] The weighted loss is obtained based on the weighted T2W channel features and the T2W pathological-image features;

[0033] The acquisition of mapping features based on the DWI channel features and / or the ADC channel features includes:

[0034] Multiply the puncture point region of the DWI sequence image with the DWI channel features to obtain the mapping features; and / or

[0035] The puncture point region of the ADC sequence image is multiplied with the ADC channel features to obtain the mapping features.

[0036] In one implementation of the first aspect, obtaining the mapping loss of the puncture point region based on the DWI channel features and / or the ADC channel features includes:

[0037] The mapping loss is obtained based on the DWI channel features and the puncture point region of the DWI sequence image; and / or

[0038] The mapping loss is obtained based on the ADC channel features and the puncture point region of the ADC sequence image.

[0039] Secondly, this application provides a machine learning-based prostate lesion identification system, the system comprising:

[0040] The image module is used to acquire the patient's historical mp-MRI images and intraoperative TRUS images;

[0041] The positioning module is used to obtain lesion spatial information of the MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images;

[0042] The puncture module is used to puncture the patient's prostate based on the lesion spatial information in order to obtain puncture pathology information;

[0043] The mapping module is used to perform spatial mapping on the historical mp-MRI images based on the puncture pathology information to obtain pathology-image feature pairs;

[0044] The training module is used to train a lesion recognition model based on the pathological-image features, so as to identify prostate lesions based on mp-MRI images using the trained lesion recognition model.

[0045] Thirdly, this application provides an electronic device, including: one or more processors; and one or more memories, wherein the memories store computer-readable code that, when run by the one or more processors, implements the machine learning-based prostate lesion identification method as described above.

[0046] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the machine learning-based prostate lesion identification method as described above.

[0047] As described above, the machine learning-based prostate lesion identification method, system, device, and medium of this application have the following beneficial effects:

[0048] This application utilizes the patient's historical mp-MRI images and intraoperative TRUS images to obtain MRI-US fused images, and automatically delineates the prostate contour and lesion area on the MRI-US fused images to obtain lesion spatial information. Subsequently, this application performs prostate biopsy and pathological examination based on the lesion spatial information, and spatially maps the biopsy pathology information onto historical mp-MRI images, thereby training a lesion recognition model. This trained lesion recognition model enables rapid and accurate prostate identification based on mp-MRI images. This application improves the accuracy of prostate cancer identification based on machine learning, enabling less or even no biopsy, and reducing the risk of biopsy complications. Furthermore, this application is easy to promote, improving the prostate diagnosis level of primary hospitals and realizing the downward flow of medical resources. Attached Figure Description

[0049] Figure 1 The diagram shown is an application illustration of the machine learning-based prostate lesion identification method described in this application.

[0050] Figure 2 The diagram shown is a flowchart of one embodiment of the machine learning-based prostate lesion identification method described in this application.

[0051] Figure 3 The diagram shown is a flowchart of one embodiment of the machine learning-based prostate lesion identification method described in this application.

[0052] Figure 4 The diagram shown is a flowchart of one embodiment of the machine learning-based prostate lesion identification method described in this application.

[0053] Figure 5 The diagram shown is a flowchart of one embodiment of the machine learning-based prostate lesion identification method described in this application.

[0054] Figure 6 The diagram shown is a structural schematic of one embodiment of the machine learning-based prostate lesion identification system described in this application.

[0055] Figure 7 The diagram shown is a structural schematic of the electronic device described in one embodiment of this application. Detailed Implementation

[0056] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0057] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0058] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0059] The following will explain the technical terms used in the embodiments of this application.

[0060] MRI: Magnetic resonance imaging is a radiation-free medical imaging technique that uses strong magnetic fields and radio waves to generate high-contrast, multi-parameter tomographic images of the human body. It can clearly display soft tissue, nerve, and vascular structures and is widely used in disease diagnosis and scientific research.

[0061] mp-MRI: Multiparametric Magnetic Resonance Imaging, is an imaging technique that combines multiple MRI sequences (such as T2W, DWI, etc.). By comprehensively analyzing different parameters (such as tissue relaxation time, diffusion characteristics, hemodynamics), it significantly improves the detection rate and diagnostic accuracy of diseases such as prostate cancer.

[0062] TRUS imaging: Transrectal Ultrasound is an examination technique that uses a high-frequency ultrasound probe inserted through the rectum to obtain high-resolution real-time images of the prostate or pelvic organs at close range.

[0063] MRI-Ultrasound Fusion Imaging.

[0064] T2W sequence images: T2-weighted imaging is a high-contrast image in MRI that highlights the differences in tissue water content and T2 relaxation time.

[0065] DWI sequence images: Diffusion-weighted imaging is an MRI technique that reflects the microstructure of tissues by detecting the random diffusion of water molecules.

[0066] ADC sequence image: Apparent Diffusion Coefficient, is a quantitative derivative image of DWI, which visually displays the microstructure of tissues by calculating the diffusion rate of water molecules.

[0067] The technical solutions in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0068] Figure 1 The diagram shows a flowchart of the machine learning-based prostate lesion identification method provided in an embodiment of this application. Figure 1 As shown, the machine learning-based prostate lesion identification method includes steps S1 to S5.

[0069] S1. Obtain the patient's historical mp-MRI images and intraoperative TRUS images.

[0070] Specifically, the patient's historical mp-MRI images and intraoperative TRUS images were obtained, and the prostate contour and suspicious lesion extent on the mp-MRI images were manually annotated, as were the prostate contour and suspicious lesion extent on the intraoperative TRUS images.

[0071] S2. Based on the historical mp-MRI images and the intraoperative TRUS images, obtain the lesion spatial information of the MRI-US fusion image.

[0072] Specifically, the historical mp-MRI images annotated in step S1 and the intraoperative TRUS images are fused to obtain MRI-US fused images, thereby better clarifying the patient's prostate contour and the extent of suspicious lesions. Figure 2 The diagram shows a flowchart of the machine learning-based prostate lesion identification method provided in an embodiment of this application. Figure 2 As shown, obtaining lesion spatial information from the MRI-US fusion image based on the historical mp-MRI image and the TRUS image includes:

[0073] S21. The historical mp-MRI image and the TRUS image are segmented based on a prostate segmentation network to obtain a first mask image and a second mask image accordingly. In image processing, a mask image is a binary image (pixel values ​​are usually 0 or 1) of the same size as the original image. It is used to specify which regions in the original image are regions of interest (regions with pixel values ​​of 1) and non-regions of interest (regions with pixel values ​​of 0).

[0074] Figure 3 The diagram shown is a schematic representation of the prostate segmentation network provided in an embodiment of this application. Figure 3 As shown, historical mp-MRI images and TRUS images are input into a prostate segmentation network. The input image data is processed using a random masking method on this network. Non-zero regions in the image data are input into the encoder for feature extraction. Masked regions are added to the extracted features and then input into the decoder to restore the original image values. This process yields the first mask image (corresponding to the historical mp-MRI image) and the second mask image (corresponding to the TRUS image).

[0075] in, Figure 4 The diagram shown is a schematic representation of the decoder structure in the prostate segmentation network provided in an embodiment of this application. In some embodiments, the prostate segmentation network is obtained by: initializing and loading the encoder weights of an unsupervised network, and training the unsupervised network using prostate segmentation data to obtain the prostate segmentation network.

[0076] S22. Obtain optimal transformation parameters based on the first mask image and the second mask image, and map the first mask image onto the second mask image based on the optimal transformation parameters to obtain the lesion spatial information of the MRI-US fusion image.

[0077] Figure 5 The diagram shows a flowchart of the machine learning-based prostate lesion identification method provided in an embodiment of this application. Figure 5 As shown, obtaining the optimal transformation parameters based on the first mask image and the second mask image includes:

[0078] S221. Resample the first mask image and the second mask image.

[0079] In some embodiments, the first mask image and the second mask image are resampled to unify their physical resolution. This resampling can be performed using bilinear interpolation or bicubic interpolation.

[0080] S222. Based on the resampled first mask image and second mask image, obtain first registration data and second registration data respectively, and perform coarse registration based on the first registration data and second registration data.

[0081] In some embodiments, the first registration data includes a first center point and a first principal axis corresponding to the first mask image, and the second registration data includes a second center point and a second principal axis corresponding to the second mask image. The center point can be understood as the geometric center of the target region (the region with a pixel value of 1) in the mask image, which can be determined by calculating the average of the coordinates of all pixels in the target region. The principal axis is a straight line along the longest extension direction of the target region, and this line passes through the center point of the target region. In some embodiments, the coordinate set of all pixels in the target region can be calculated first, and the covariance matrix of these points can be calculated to describe the distribution of the point set in two-dimensional space. Then, eigenvalue decomposition is performed on the covariance matrix. The direction of the principal axis is determined by the direction of the eigenvector corresponding to the largest eigenvalue.

[0082] After obtaining the first center point and first principal axis of the first mask image, and the second center point and second principal axis of the second mask image, coarse registration is performed based on the obtained data, thereby providing a basis for subsequent fine registration.

[0083] S223. Perform nearest point matching on the coarsely registered first mask image and second mask image, and iteratively calculate the transformation parameters based on the nearest point matching results until the nearest point matching results meet the error threshold to obtain the optimal transformation parameters.

[0084] The optimal transformation parameters include rotation parameters, translation parameters, and local deformation parameters.

[0085] In some embodiments, Euclidean distance can be used to calculate feature points in the first and second mask images for nearest-neighbor matching. After matching is completed, rotation parameters, translation parameters, and local deformation parameters are calculated, and the results are iteratively updated until the average error of nearest-neighbor matching is less than a preset error threshold. Then, the iteration is terminated and the final optimal transformation parameters are obtained.

[0086] After obtaining the optimal transformation parameters, that is, establishing the transformation mapping relationship between the historical mp-MRI image and the TRUS image registration, the suspicious lesion areas marked on the historical mp-MRI image (first mask image) can be mapped onto the TRUS image, thus obtaining the lesion spatial information of the MRI-US fused image.

[0087] S3. Based on the spatial information of the lesion, perform a biopsy on the patient's prostate to obtain pathological information at the puncture point.

[0088] In some embodiments, after obtaining the lesion spatial information of the MRI-US fusion image, a surgical robot can be used to perform a prostate biopsy based on the lesion spatial information on the fusion image. That is, a puncture needle is inserted into the prostate to take tissue samples and obtain the corresponding pathological information, i.e., to determine whether the puncture result is positive or negative.

[0089] Furthermore, if cancer cells are found in the tissue sampled by puncture, the puncture result can be considered positive, and the puncture point is regarded as a puncture pathology point, with its puncture pathology information being positive; if no cancer cells are found in the sampled tissue, the puncture pathology information obtained at that puncture point is negative.

[0090] S4. Based on the pathological information of the puncture point, spatial mapping is performed on the historical mp-MRI images to obtain pathological-image feature pairs.

[0091] After obtaining the pathological information of the puncture point, the physical coordinate information of the puncture point is used to perform spatial mapping on the historical mp-MRI images, and the corresponding pathological information is marked to obtain pathological-image feature pairs. Figure 6 The diagram shows a flowchart of the machine learning-based prostate lesion identification method provided in an embodiment of this application. Figure 6 As shown, obtaining pathological-image feature pairs includes steps S41 to S44.

[0092] S41. Locate the puncture point in the T2W sequence image based on the physical coordinates of the puncture point, and determine the radius of the puncture point based on the physical information of the puncture needle.

[0093] S42. Determine the puncture point region in the T2W sequence image based on the puncture point and the radius of the puncture point.

[0094] Specifically, during biopsy sampling, the puncture point actually has three-dimensional physical coordinates. These physical coordinates are then used to locate the puncture point in the T2W sequence image, i.e., obtaining the pixel coordinates of the puncture point in the T2W sequence image. Furthermore, because the sampling groove of the puncture needle has a certain diameter, when the puncture needle performs sampling, it actually passes through a narrow, elongated area through the puncture point, thus forming the puncture point region.

[0095] S43. Based on the puncture point region in the T2W sequence image, determine the puncture point region of the DWI sequence image and the puncture point region of the ADC sequence image respectively.

[0096] In some embodiments, a correspondence between T2WA sequence images and DWI sequence images, and between the pixel coordinates of DWI sequence images, can be established based on the scanning parameters and positioning markers of the MRI device. That is, after obtaining the puncture point region in the T2W sequence image, the puncture point region in the DWI sequence image and the puncture point region in the ADC sequence image can be determined accordingly.

[0097] S44. Based on the puncture point region and pathological information in the T2W sequence image, the puncture point region and pathological information in the DWI sequence image, and the puncture point region and pathological information in the ADC sequence image, T2W pathological-image feature pairs, DWI pathological-image feature pairs, and ADC pathological-image feature pairs are obtained respectively.

[0098] Specifically, after acquiring the puncture point region in different MRI image sequences, pathological-image feature pairs are constructed based on the corresponding pathological information of the puncture point region. That is, after acquiring the puncture point region in T2W, DWI, and ADC sequence images, T2W pathological-image feature pairs, DWI pathological-image feature pairs, and ADC pathological-image feature pairs are constructed based on the pathological information of the puncture point.

[0099] For example, when the puncture point is a pathological site, the pathological information of the puncture point region in the sequence image should be positive, and a 1 can be assigned to that pixel region. When the pathological information of the puncture point is negative, a 0 can be assigned to that puncture point region.

[0100] Furthermore, before acquiring T2W pathological-image feature pairs, DWI pathological-image feature pairs, and ADC pathological-image feature pairs, the DWI sequence images and ADC sequence images should be resampled to T2W sequence images, and size normalization and filtering should be performed to ensure the uniformity of subsequent data input.

[0101] S5. Based on the pathological-image features, train the lesion recognition model to identify prostate lesions based on mp-MRI images using the trained lesion recognition model.

[0102] In order to identify prostate lesion areas directly using machine learning with minimal or no biopsy, this application will train a lesion identification model to predict prostate lesion areas based on new mp-MRI images. Figure 7 The diagram shows a flowchart of the machine learning-based prostate lesion identification method provided in an embodiment of this application. Figure 7 As shown, training the lesion recognition model based on the pathological-image features includes steps S51 to S53.

[0103] S51. Input the T2W pathological-image feature pair, the DWI pathological-image feature pair, and the ADC pathological-image feature pair into the lesion recognition model to extract T2W channel features, DWI channel features, and ADC channel features based on the lesion recognition model.

[0104] Specifically, extracting T2W, DWI, and ADC channel features through the lesion identification model can enhance specific information, reduce the redundancy of irrelevant information, improve the accuracy of identification and classification, and enhance the model's generalization ability.

[0105] S52. Obtain the mapping loss of the puncture point region and the weighted loss of the T2W channel features based on the DWI channel features and / or the ADC channel features.

[0106] Specifically, obtaining the weighted loss of T2W channel features based on the DWI channel features and / or the ADC channel features includes: obtaining mapping features based on the DWI channel features and / or the ADC channel features; weighting the T2W channel features based on the mapping features to obtain weighted T2W channel features; and obtaining the weighted loss based on the weighted T2W channel features and the T2W pathology-image feature pair. By weighting the T2W channel features using mapping features, the pathological region information in the T2W feature map can be highlighted more clearly. Therefore, the deviation between the weighted T2W channel features and the pathological information in the T2W pathology-image feature pair is the weighted loss.

[0107] In some embodiments, obtaining mapping features based on the DWI channel features and / or the ADC channel features includes: multiplying the puncture point region of the DWI sequence image with the DWI channel features to obtain mapping features; and / or multiplying the puncture point region of the ADC sequence image with the ADC channel features to obtain mapping features.

[0108] Specifically, obtaining the mapping loss of the puncture point region based on the DWI channel features and / or the ADC channel features includes: obtaining the mapping loss based on the DWI channel features and the puncture point region of the DWI sequence image; and / or obtaining the mapping loss based on the ADC channel features and the puncture point region of the ADC sequence image. That is, the mapping loss is the deviation value of the pathological information between the DWI channel features and the puncture point region of the DWI sequence image, and / or the deviation value of the pathological information between the ADC channel features and the puncture point region of the ADC sequence image.

[0109] S53. Train the lesion identification model based on the mapping loss and the weighted loss until it converges.

[0110] Specifically, the mapping loss and weighted loss are used as the update backpropagation values ​​for the lesion identification model. When the mapping loss converges, it indicates that the DWI channel features and / or ADC channel features extracted by the lesion identification model are consistent with the puncture point region in the labeled DWI and / or ADC sequence images. Similarly, when the weighted loss converges, it indicates that the T2W channel features extracted by the lesion identification model are consistent with the puncture point region in the T2W sequence images. Therefore, the trained lesion identification model can receive new mp-MRI images and directly extract the corresponding channel features to identify the lesion region of the prostate.

[0111] Figure 8 This diagram illustrates the training principle of the lesion recognition model described in an embodiment of this application. Figure 8 As shown, T2W, DWI, and ADC sequences are used as input information, and the pathological information of the puncture site region already acquired on the images is used as output information, thereby enabling supervised training of the lesion recognition model. Once the training process converges according to the above, the lesion recognition model can directly predict the lesion region of the prostate on new mp-MRI images.

[0112] Figure 9 The diagram shown illustrates the application of the lesion recognition model described in this embodiment. Figure 9 As shown, T2W, DWI, and ADC sequence images are input into the lesion recognition model to extract T2W, DWI, and ADC channel features. These channel features effectively highlight the pathological region information in the feature map. Then, based on the imaging signal intensity of the pathological region on different sequences, strong representations are determined from the DWI and ADC sequence images, while the T2W sequence image serves as a weak representation and identifies the tissue region. Therefore, when a pixel in the sequence image represents a puncture pathological point but the DWI / ADC contrast is not highly distinguishable, the feature weights favor T2W; conversely, when the DWI / ADC contrast is highly distinguishable, the feature weights favor DWI / ADC. Thus, the different channel features are fused and stitched together with corresponding weights, and a lesion region probability map is output based on the fused features and displayed in the T2W image, along with pixel coordinates and world coordinates.

[0113] The scope of protection of the machine learning-based prostate lesion identification method described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0114] This application also provides a machine learning-based prostate lesion identification system. The machine learning-based prostate lesion identification system can implement the machine learning-based prostate lesion identification method described in this application. However, the implementation device of the machine learning-based prostate lesion identification method described in this application includes, but is not limited to, the structure of the machine learning-based prostate lesion identification system listed in this embodiment. All structural modifications and substitutions of the prior art made in accordance with the principles of this application are included within the protection scope of this application.

[0115] Figure 10 The diagram shown is an architectural schematic of the machine learning-based prostate lesion identification system described in an embodiment of this application. Figure 10 As shown in the illustration, this application also provides a machine learning-based prostate lesion identification system, which includes:

[0116] Image module 41 is used to acquire the patient's historical mp-MRI images and intraoperative TRUS images;

[0117] The positioning module 42 is used to obtain lesion spatial information of the MRI-US fusion image based on the historical mp-MRI image and the intraoperative TRUS image;

[0118] The puncture module 43 is used to perform a puncture on the patient's prostate based on the lesion spatial information in order to obtain pathological information of the puncture point;

[0119] Mapping module 44 is used to perform spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to obtain pathological-image feature pairs;

[0120] Training module 45 is used to train a lesion recognition model based on the pathological-image features, so as to identify prostate lesions based on mp-MRI images using the trained lesion recognition model.

[0121] The structure and principle of the image module 41, the positioning module 42, the puncture module 43, the mapping module 44, and the training module 45 correspond one-to-one with the steps in the above-mentioned machine learning-based prostate lesion identification method, so they will not be described again here.

[0122] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0123] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of this application, depending on actual needs. For example, the functional modules / units in the various embodiments of this application may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0124] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0125] This application also provides an electronic device, including: one or more processors; and one or more memories, wherein the memories store computer-readable code that, when run by the one or more processors, executes the machine learning-based prostate lesion identification method as described in this application.

[0126] The electronic device can be used by means of Figure 11 The architecture of the exemplary computing device shown is used for implementation. Figure 11As shown, this exemplary computing device may include a bus 910, one or more GPUs 920, a read-only memory (ROM) 930, a random access memory (RAM) 940, a communication port 950 connected to a network, an input / output component 960, a hard disk 970, etc. The storage devices in the computing device 900, such as the ROM 930 or the hard disk 970, may store various data or files used for computer processing and / or communication, as well as program instructions executed by the GPU. The computing device 900 may also include a user interface 980. Of course, Figure 7 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 6 One or more components in the computing device shown.

[0127] This application also provides a computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the machine learning-based prostate lesion identification method as described in this application. Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be executed by computer-readable instructions stored on the computer-readable storage medium. The computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM).

[0128] This application also provides a computer program product or computer program that includes computer-readable instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer-readable instructions from the computer-readable storage medium and execute the instructions, causing the computer device to perform the machine learning-based prostate lesion identification method described in the various embodiments above.

[0129] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0130] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method of prostate lesion identification based on machine learning, characterized by, The method comprises: acquiring historical mp-MRI images and intraoperative TRUS images of a patient; acquiring lesion spatial information of an MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images; performing puncture on the prostate of the patient based on the lesion spatial information to acquire pathological information of a puncture point; performing spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs; training a lesion recognition model based on the pathological-image feature pairs to perform prostate lesion recognition based on mp-MRI images through the trained lesion recognition model; wherein the spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs comprises: locating the puncture point in a T2W sequence image based on the physical coordinates of the puncture point and determining the radius of the puncture point based on the physical information of the puncture needle; determining the puncture point region in the T2W sequence image based on the puncture point and the radius of the puncture point; determining the puncture point region in the DWI sequence image and the puncture point region in the ADC sequence image based on the puncture point region in the T2W sequence image respectively; correspondingly acquiring T2W pathological-image feature pairs, DWI pathological-image feature pairs and ADC pathological-image feature pairs based on the puncture point region in the T2W sequence image and the pathological information, the puncture point region in the DWI sequence image and the pathological information, and the puncture point region in the ADC sequence image and the pathological information respectively. 2.The machine learning-based prostate lesion identification method of claim 1, wherein, The method comprises: acquiring historical mp-MRI images and intraoperative TRUS images of a patient; acquiring lesion spatial information of an MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images; 3.The machine learning based prostate lesion identification method of claim 2, wherein, performing puncture on the prostate of the patient based on the lesion spatial information to acquire pathological information of a puncture point; performing spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs; training a lesion recognition model based on the pathological-image feature pairs to perform prostate lesion recognition based on mp-MRI images through the trained lesion recognition model; wherein the spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs comprises: locating the puncture point in a T2W sequence image based on the physical coordinates of the puncture point and determining the radius of the puncture point based on the physical information of the puncture needle; determining the puncture point region in the T2W sequence image based on the puncture point and the radius of the puncture point; determining the puncture point region in the DWI sequence image and the puncture point region in the ADC sequence image based on the puncture point region in the T2W sequence image respectively; correspondingly acquiring T2W pathological-image feature pairs, DWI pathological-image feature pairs and ADC pathological-image feature pairs based on the puncture point region in the T2W sequence image and the pathological information, the puncture point region in the DWI sequence image and the pathological information, and the puncture point region in the ADC sequence image and the pathological information respectively. The method comprises: acquiring historical mp-MRI images and intraoperative TRUS images of a patient; acquiring lesion spatial information of an MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images; performing puncture on the prostate of the patient based on the lesion spatial information to acquire pathological information of a puncture point; performing spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs; training a lesion recognition model based on the pathological-image feature pairs to perform prostate lesion recognition based on mp-MRI images through the trained lesion recognition model; wherein the spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to acquire pathological-image feature pairs comprises: locating the puncture point in a T2W sequence image based on the physical coordinates of the puncture point and determining the radius of the puncture point based on the physical information of the puncture needle; determining the puncture point region in the T2W sequence image based on the puncture point and the radius of the puncture point; determining the puncture point region in the DWI sequence image and the puncture point region in the ADC sequence image based on the puncture point region in the T2W sequence image respectively; correspondingly acquiring T2W pathological-image feature pairs, DWI pathological-image feature pairs and ADC pathological-image feature pairs based on the puncture point region in the T2W sequence image and the pathological information, the puncture point region in the DWI sequence image and the pathological information, and the puncture point region in the ADC sequence image and the pathological information respectively. 4.The machine learning based prostate lesion identification method of claim 1, wherein, training a lesion recognition model based on the pathology-image feature pairs comprises: inputting the T2W pathology-image feature pairs, the DWI pathology-image feature pairs and the ADC pathology-image feature pairs into the lesion recognition model to extract T2W channel features, DWI channel features and ADC channel features based on the lesion recognition model; obtaining a mapping loss of a puncture point region and a weighted loss of T2W channel features based on the DWI channel features and / or the ADC channel features; training the lesion recognition model based on the mapping loss and the weighted loss until convergence. 5.The machine learning based prostate lesion identification method of claim 4, wherein, obtaining a weighted loss of T2W channel features based on the DWI channel features and / or the ADC channel features comprises: obtaining a mapping feature based on the DWI channel features and / or the ADC channel features; performing weighted processing on the T2W channel features based on the mapping feature to obtain weighted T2W channel features; obtaining the weighted loss based on the weighted T2W channel features and the T2W pathology-image feature pairs; wherein obtaining a mapping feature based on the DWI channel features and / or the ADC channel features comprises: performing multiplication operation on the puncture point region of the DWI sequence image and the DWI channel features to obtain a mapping feature; and / or performing multiplication operation on the puncture point region of the ADC sequence image and the ADC channel features to obtain a mapping feature. 6.The machine learning based prostate lesion identification method of claim 4, wherein, obtaining a mapping loss of a puncture point region based on the DWI channel features and / or the ADC channel features comprises: obtaining the mapping loss based on the DWI channel features and the puncture point region of the DWI sequence image; and / or obtaining the mapping loss based on the ADC channel features and the puncture point region of the ADC sequence image.

7. A machine learning based prostate lesion identification system, characterized by, The system comprises: an image module configured to obtain historical mp-MRI images and intraoperative TRUS images of a patient; a positioning module configured to obtain lesion spatial information of an MRI-US fusion image based on the historical mp-MRI images and the intraoperative TRUS images; a puncture module configured to puncture the prostate of the patient based on the lesion spatial information to obtain pathological information of a puncture point; a mapping module configured to perform spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to obtain pathology-image feature pairs; a training module configured to train a lesion recognition model based on the pathology-image feature pairs, so that the prostate lesion recognition is performed based on mp-MRI images by using the trained lesion recognition model; wherein the spatial mapping on the historical mp-MRI images based on the pathological information of the puncture point to obtain pathology-image feature pairs comprises: locating the puncture point in the T2W sequence image based on the physical coordinates of the puncture point, and determining the radius of the puncture point based on the physical information of the puncture needle; determining the puncture point region in the T2W sequence image based on the puncture point and the radius of the puncture point; determine a puncture point region of the DWI sequence image and a puncture point region of the ADC sequence image based on the puncture point region in the T2W sequence image respectively; correspondingly acquire a T2W pathology-image feature pair, a DWI pathology-image feature pair and an ADC pathology-image feature pair based on the puncture point region and the pathology information in the T2W sequence image, the puncture point region and the pathology information in the DWI sequence image, and the puncture point region and the pathology information in the ADC sequence image respectively.

8. An electronic device, comprising: comprise: one or more processors; and one or more memories, wherein the memories have stored therein computer readable code which, when executed by the one or more processors, implements the machine learning based prostate lesion identification method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored therein instructions which, when executed by a processor, cause the processor to perform the machine learning based prostate lesion identification method according to any one of claims 1 to 6.

Images