Magnetic sensitive weighted imaging detection system and method based on machine learning

Abstract

The embodiment of the invention discloses a magnetic sensitive weighted imaging detection system and method based on machine learning, and the system comprises a building module which is used for building a magnetic sensitive weighted imaging image library; wherein each image in the magnetic sensitive weighted imaging image library is marked with an ROI (Region of Interest) in advance; the feature module is used for performing feature extraction and screening based on the magnetic sensitive weighted imaging image library; the processing module is used for training by using a preset machine learning network according to the screened features so as to obtain a detection model; the detection module is used for reading a to-be-identified SWI image and transmitting the to-be-identified SWI image to the detection model for identification so as to obtain a classification result of corresponding detection; the method has the beneficial effects that the working efficiency and accuracy of detection and recognition are improved, and interference of subjective factors is also avoided; the defects that currently, the requirement for technicians is high, popularization is difficult to achieve in some areas with the low technical level, and much time is consumed are overcome.

Description

Technical Field

[0001] The present invention relates to the field of auxiliary detection technology, and in particular to a detection system and method for magnetic susceptibility weighted imaging based on machine learning. Background Art

[0002] Susceptibility Weighted Imaging (SWI) is a technique based on magnetic resonance imaging (MRI) that has significant sensitivity in detecting hemorrhagic lesions such as microbleeds, cerebral hemorrhages, and tumor hemorrhages. Susceptibility Weighted Imaging is a new imaging technique that uses the difference in magnetic sensitivity between different tissues to produce a contrast enhancement mechanism.

[0003] At present, susceptibility weighted imaging usually requires professional technicians to operate and interpret; therefore, this makes it difficult to popularize in some areas with low technical levels, resulting in high requirements for technicians and the disadvantage of being more time-consuming. Summary of the invention

[0004] In order to overcome the deficiencies in the prior art, the object of the present invention is to provide a detection system and method for magnetic susceptibility weighted imaging based on machine learning.

[0005] The technical solution provided by the present invention is:

[0006] In a first aspect, a detection system for magnetic susceptibility weighted imaging based on machine learning is provided, the detection system comprising:

[0007] A construction module is used to establish a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-labeled with a ROI region;

[0008] A feature module, used for extracting and screening features based on the susceptibility weighted imaging image library;

[0009] A processing module is used to train a preset machine learning network based on the screened features to obtain a detection model;

[0010] The detection module is used to read the SWI image to be identified and transmit it to the detection model for identification to obtain the corresponding detection classification result.

[0011] Preferably, the screening comprises:

[0012] The extracted feature values ​​were subjected to Z-score regularization, and the correlation of each feature was evaluated to exclude features with too high consistency.

[0013] Preferably, before extracting features, the feature module also converts the image corresponding to each ROI region into an image with information of multiple different scales as input, so that the feature module can extract feature information of different scales of the image.

[0014] Preferably, during the training, images of different scales correspond to machine learning networks of different structures respectively and serve as pre-training sub-models;

[0015] The sub-models are formed into an integrated model through a weighted fusion method of the model output category probability, and the integrated model is used as the detection model.

[0016] Preferably, the features extracted include first-order features, morphological features and texture features.

[0017] In a second aspect, the present invention further provides a detection method of susceptibility weighted imaging based on machine learning, which is applied to a detection system of susceptibility weighted imaging based on machine learning described in the first aspect, and the method comprises:

[0018] Establishing a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-marked with a ROI region;

[0019] Performing feature extraction and screening based on the susceptibility weighted imaging image library;

[0020] According to the screened features, a preset machine learning network is used for training to obtain a detection model;

[0021] The SWI image to be identified is read and transmitted to the detection model for identification to obtain a classification result of the corresponding detection.

[0022] Through the above technical solution, the present invention can bring the following beneficial effects:

[0023] The present invention realizes intelligent processing of SWI images to be identified by combining susceptibility weighted imaging with machine learning, thereby improving the efficiency and accuracy of detection and identification, and avoiding the interference of subjective factors; it overcomes the current defects of high requirements for technical personnel, difficulty in popularization in some areas with low technical levels, and long time consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 An overall architecture diagram of a detection system for magnetic susceptibility weighted imaging based on machine learning provided by an embodiment of the present invention;

[0025] Figure 2 A flow chart of a detection method for susceptibility weighted imaging based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are only for illustration and are not intended to limit the present invention. In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is apparent to those of ordinary skill in the art that these specific details need not be adopted to implement the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] In the present application, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0029] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.

[0030] Throughout the specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment of the present invention. Therefore, the phrases "in one embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily all refer to the same embodiment or example. Furthermore, particular features, structures, or characteristics may be combined in one or more embodiments or examples in any appropriate combinations and / or subcombinations.

[0031] It should be noted that, unless otherwise specified, the technical terms in this embodiment have the common meanings understood in the relevant technical field.

[0032] like Figure 1 As shown, an embodiment of the present invention provides a detection system for magnetic susceptibility weighted imaging based on machine learning, and the detection system includes:

[0033] A construction module is used to establish a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-labeled with a ROI region;

[0034] A feature module, used for extracting and screening features based on the susceptibility weighted imaging image library;

[0035] A processing module is used to train a preset machine learning network based on the screened features to obtain a detection model;

[0036] The detection module is used to read the SWI image to be identified and transmit it to the detection model for identification to obtain the corresponding detection classification result.

[0037] When applied, a corresponding database is constructed by pre-acquired various normal, hemorrhagic, benign and malignant susceptibility weighted imaging images; and then the region of interest (ROI) is pre-marked for each image.

[0038] In this embodiment, the features extracted include first-order features, morphological features and texture features.

[0039] Specifically, the first-order features use the characteristic values ​​of the ROI image to quantitatively describe the signal intensity and its distribution in the image with a histogram; the morphological features can quantify the shape and size of the ROI image, such as the circumference, sphericity, major axis length and minor axis length of the ROI image; the texture features can quantitatively characterize the heterogeneity within the relevant lesions; for example, quantify the difference between the grayscale value within a given distance and the average grayscale value within its adjacent distance, including roughness, contrast, and complexity.

[0040] In this embodiment, the extracted features are further screened to remove similar and redundant features to improve the computational efficiency of the algorithm and prevent overfitting; the screening includes:

[0041] The extracted feature values ​​were subjected to Z-score regularization, and the correlation of each feature was evaluated to exclude features with too high consistency.

[0042] When applied, the Spearman correlation test is used to evaluate the correlation of each feature. The positive or negative value of the correlation coefficient reflects whether the direction of feature change is consistent. The larger the absolute value, the higher the correlation between features.

[0043] Furthermore, during implementation, in order to avoid ignoring the rich detail information in the image due to a single viewing angle scale, on the basis of the above technical solution, the feature module converts the image corresponding to each ROI area into an image with information of multiple scales as input before performing feature extraction, so that the feature module can extract feature information of different scales of the image, thereby improving the corresponding feature extraction capability.

[0044] It can be converted into multiple scales according to the required situation; for example, four or five scales, etc., which are not limited here;

[0045] Correspondingly, in order to effectively solve the problem that a single model or network only focuses on one aspect and affects the classification performance; during the training, images of different scales correspond to a machine learning network with a different structure, and serve as a pre-training sub-model;

[0046] The sub-models are formed into an integrated model through a weighted fusion method of the model output category probability, and the integrated model is used as the detection model.

[0047] Taking four machine learning networks with four different structures corresponding to four scales as an example, four machine learning networks with different structures (such as random forest, XGBoost and deep convolutional neural network, etc.) are selected as pre-trained sub-models. Through each network, the defects of each other are complemented to extract diversified feature information and improve the recognition ability of images.

[0048] For the fully connected layers involved, the Dropout layer and regularization are used to reduce the overfitting phenomenon; thus effectively reducing the calculation parameters, speeding up the training, and improving the model classification performance;

[0049] And use transfer learning to fine-tune some convolutional layers of the model and the weights of the convolutional neural network to make it more data-specific and better learn the feature differences between the source domain and the target domain.

[0050] Since the combination strategy in the integrated model will also affect the final prediction performance, this application comprehensively selects a weighted fusion method using the model output category probability. Compared with other methods such as model output category probability, it can obtain the best classification performance and better combine the diversity of the model.

[0051] The above scheme realizes intelligent processing of SWI images to be identified by combining susceptibility weighted imaging with machine learning, thereby improving the efficiency and accuracy of detection and identification, and avoiding the interference of subjective factors; it overcomes the current defects of high requirements for technical personnel, difficulty in popularization in some areas with low technical levels, and a long time consumption.

[0052] Based on the same inventive concept, Figure 2 A detection method of magnetic susceptibility weighted imaging based on machine learning is applied to the detection system of magnetic susceptibility weighted imaging based on machine learning as described above, and the method comprises:

[0053] S101, establishing a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-labeled with a ROI region;

[0054] S102, performing feature extraction and screening based on the susceptibility weighted imaging image library;

[0055] S103, training using a preset machine learning network according to the screened features to obtain a detection model;

[0056] S104, reading the SWI image to be identified, and transmitting it to the detection model for identification, so as to obtain a classification result of the corresponding detection.

[0057] Wherein, the screening includes:

[0058] The extracted feature values ​​were subjected to Z-score regularization, and the correlation of each feature was evaluated to exclude features with too high consistency.

[0059] Furthermore, the method further comprises:

[0060] Before feature extraction, the image corresponding to each ROI area is converted into an image with information of multiple different scales as input, so that the feature module can extract feature information of different scales of the image.

[0061] In this embodiment, during the training, images of different scales correspond to machine learning networks of different structures and serve as pre-training sub-models;

[0062] The sub-models are formed into an integrated model through a weighted fusion method of the model output category probability, and the integrated model is used as the detection model.

[0063] Taking four machine learning networks with four different structures corresponding to four scales as an example, four machine learning networks with different structures (such as random forest, XGBoost and deep convolutional neural network, etc.) are selected as pre-trained sub-models. Through each network, the defects of each other are complemented to extract diversified feature information and improve the recognition ability of images.

[0064] For the fully connected layers involved, the Dropout layer and regularization are used to reduce the overfitting phenomenon; thus effectively reducing the calculation parameters, speeding up the training, and improving the model classification performance;

[0065] And use transfer learning to fine-tune some convolutional layers of the model and the weights of the convolutional neural network to make it more data-specific and better learn the feature differences between the source domain and the target domain.

[0066] It should be noted that for a more specific description of the workflow of the method embodiment, please refer to the aforementioned system embodiment part, which will not be repeated here.

[0067] The above scheme realizes intelligent processing of SWI images to be identified by combining the involved susceptibility weighted imaging with machine learning, thereby improving the efficiency and accuracy of detection and identification, and avoiding the interference of subjective factors; it overcomes the current defects of high requirements for technical personnel, difficulty in popularization in some areas with low technical levels, and a long time consumption.

[0068] In the several embodiments provided in the present application, it should be understood that the disclosed detection system of magnetic susceptibility weighted imaging based on machine learning can be implemented in other ways. For example, the embodiments described above are only schematic. For example, the division of the modules can be divided in other ways in actual implementation, such as multiple units or components can be combined or integrated into another system or device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0069] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A detection system for magnetic susceptibility weighted imaging based on machine learning, characterized in that: The detection system comprises: A construction module is used to establish a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-labeled with a ROI region; A feature module, used for extracting and screening features based on the susceptibility weighted imaging image library; A processing module is used to train a preset machine learning network based on the screened features to obtain a detection model; The detection module is used to read the SWI image to be identified and transmit it to the detection model for identification to obtain the corresponding detection classification result.

2. The detection system of magnetic susceptibility weighted imaging based on machine learning according to claim 1, characterized in that: The screening includes: The extracted feature values ​​were subjected to Z-score regularization, and the correlation of each feature was evaluated to exclude features with too high consistency.

3. The detection system of magnetic susceptibility weighted imaging based on machine learning according to claim 2, characterized in that: Before extracting features, the feature module converts the images corresponding to the ROI regions into images with information of multiple different scales as input, so that the feature module can extract feature information of different scales of the image.

4. The detection system of magnetic susceptibility weighted imaging based on machine learning according to claim 3, characterized in that: During the training, images of different scales correspond to machine learning networks of different structures and serve as pre-training sub-models; The sub-models are formed into an integrated model through a weighted fusion method of the model output category probability, and the integrated model is used as the detection model.

5. A detection system for magnetic susceptibility weighted imaging based on machine learning according to any one of claims 1 to 4, characterized in that: The features extracted include first-order features, morphological features and texture features.

6. A detection method of magnetic susceptibility weighted imaging based on machine learning, characterized in that: The method applied to the detection system of magnetic susceptibility weighted imaging based on machine learning as described in claim 1 comprises: Establishing a susceptibility weighted imaging image library; wherein each image in the susceptibility weighted imaging image library is pre-marked with a ROI region; Performing feature extraction and screening based on the susceptibility weighted imaging image library; According to the screened features, a preset machine learning network is used for training to obtain a detection model; The SWI image to be identified is read and transmitted to the detection model for identification to obtain a classification result of the corresponding detection.

7. The detection method of magnetic susceptibility weighted imaging based on machine learning according to claim 6, characterized in that: The screening includes: The extracted feature values ​​were subjected to Z-score regularization, and the correlation of each feature was evaluated to exclude features with too high consistency.

8. The detection method of magnetic susceptibility weighted imaging based on machine learning according to claim 7, characterized in that: The method further comprises: Before feature extraction, the image corresponding to each ROI area is converted into an image with information of multiple different scales as input, so that the feature module can extract feature information of different scales of the image.

9. The detection method of magnetic susceptibility weighted imaging based on machine learning according to claim 8, characterized in that: During the training, images of different scales correspond to machine learning networks of different structures and serve as pre-training sub-models; The sub-models are formed into an integrated model through a weighted fusion method of the model output category probability, and the integrated model is used as the detection model.

Images