A method, device and processing equipment for detecting cerebral microbleeds in an image based on SWI

Abstract

The application provides a cerebral microbleed image detection method and device based on SWI and processing equipment, which is based on SWI and provides a set of supporting solutions for more efficient and high-precision automatic identification requirements, thereby meeting the identification requirements of CMB and subsequent diagnosis requirements. The method comprises the following steps: acquiring T1 modality images and SWI images; inputting the T1 modality images and the SWI images into a pre-configured CMB detection module to combine the T1 modality images and the SWI images for CMB detection and obtain a CMB detection result; acquiring MR plain scan images; inputting the MR plain scan images into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan images and obtain a whole lesion segmentation result; the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction; combining the CMB detection result and the whole lesion segmentation result to calculate corresponding imaging features, and the imaging features include brain region distribution, quantity, shape and size.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image detection, in particular to a cerebral microbleed image detection method and device based on SWI and processing equipment. BACKGROUND

[0002] Cerebral microbleeds (CMB) is an imaging manifestation of cerebral small vessel disease, which refers to the leakage of red blood cells due to damage to small blood vessels in the brain, and the formation of hemosiderin deposits around blood vessels after phagocytosis by macrophages.

[0003] For CMB detection, in actual operation, it is usually based on the Susceptibility Weighted Imaging (SWI) sequence combined with clinical experience.

[0004] However, the present inventors have found that the current clinical diagnosis of CMB still relies largely on manual work, which not only consumes time and effort, but also is prone to missed detection. Therefore, it is necessary to develop an efficient and high-precision automatic recognition scheme to better meet the actual needs. SUMMARY

[0005] The present application provides a cerebral microbleed image detection method, device and processing equipment based on SWI, which provides a complete supporting scheme for more efficient and high-precision automatic recognition needs based on SWI, thereby meeting the recognition needs of CMB and subsequent diagnosis needs.

[0006] In a first aspect, the present application provides a cerebral microbleed image detection method based on SWI, the method comprising:

[0007] obtaining T1 modality images and SWI images;

[0008] inputting the T1 modality images and the SWI images into a pre-configured CMB detection module to perform CMB detection in combination with the T1 modality images and the SWI images, and obtaining a CMB detection result;

[0009] obtaining MR plain scan images, wherein the MR plain scan images include multi-modality phase images of T1, T2 and FLAIR;

[0010] inputting the MR plain scan images into a pre-configured full lesion segmentation module to perform full lesion segmentation based on the MR plain scan images, and obtaining a full lesion segmentation result, wherein the lesions involved in the full lesion segmentation include WMH, PVS, LA and infarction;

[0011] The imaging features are calculated in combination with the CMB detection result and the whole lesion segmentation result, wherein the imaging features include brain region distribution, quantity, shape and size.

[0012] In combination with the first aspect of the application, in a first possible implementation manner of the first aspect of the application, the processing content of the CMB detection module includes:

[0013] The T1 modality image is input into the brain region segmentation network to obtain a brain region segmentation result.

[0014] The SWI image is input into the CMB segmentation network to obtain a CMB segmentation result.

[0015] The T1 modality image and the brain region segmentation result are registered to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and the brain region segmentation result thereof.

[0016] In combination with the first possible implementation manner of the first aspect of the application, in a second possible implementation manner of the first aspect of the application, the MR plain scan image is input into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain a whole lesion segmentation result, including:

[0017] The MR plain scan image is input into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain an initial whole lesion segmentation result.

[0018] The MR plain scan image and the initial whole lesion segmentation result are registered to the SWI image space through a rigid registration algorithm to obtain a whole lesion segmentation result of the SWI modality.

[0019] In combination with the first possible implementation manner of the first aspect of the application, in a third possible implementation manner of the first aspect of the application, before the T1 modality image is input into the brain region segmentation network to obtain a brain region segmentation result, the method further includes:

[0020] The T1 modality image is subjected to N4 bias field correction.

[0021] In combination with the first possible implementation manner of the first aspect of the application, in a fourth possible implementation manner of the first aspect of the application, the U-Net framework is adopted for inputting into the brain region segmentation network and the CMB segmentation network, in the form of an encoder-decoder cascade, to learn global features of an image through a continuous downsampling module, decode the learned semantic features to an input image space to obtain a segmentation result through an upsampling module, and fuse the position information at a bottom layer with semantic information at a deep layer through a skip connection module to obtain a more accurate segmentation result.

[0022] In a fifth possible implementation manner of the first aspect of the present application, in combination with the CMB detection result and the whole lesion segmentation result, after the corresponding imaging features are calculated, the method further comprises:

[0023] inputting the imaging features into a pre-configured CMB cause detection module to perform CMB cause detection based on the imaging features, and obtaining a CMB cause detection result.

[0024] In a sixth possible implementation manner of the first aspect of the present application, the processing content of the CMB cause detection module comprises:

[0025] using an ordinal outcome variable nomogram algorithm to construct a linear relationship between the preset imaging features and the preset CMB causes, wherein the linear relationship simultaneously fits multiple dichotomous logistic models, and has:

[0026] Prob(y0)=βX+α0

[0027] Prob(y0+y1)=βX+α1

[0028] Prob(y0+y1+y2)=βX+α2

[0029] Prob(y0+y1+y2+y3)=βX+α3

[0030] ………

[0031] wherein Prob represents probability, y represents cause, X represents feature, β represents the linear relationship between the constructed feature and cause, and a represents different intercepts.

[0032] In a second aspect, the present application provides a SWI-based cerebral microbleeding image detection device, which comprises:

[0033] an acquisition unit configured to acquire T1 modality images and SWI images;

[0034] a CMB detection unit configured to input the T1 modality images and the SWI images into a pre-configured CMB detection module to perform CMB detection in combination with the T1 modality images and the SWI images, and obtain a CMB detection result;

[0035] the acquisition unit is further configured to acquire MR plain scan images, wherein the MR plain scan images comprise multi-modality phase images of T1, T2 and FLAIR;

[0036] a whole lesion segmentation unit configured to input the MR plain scan images into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan images, and obtain a whole lesion segmentation result, wherein the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction;

[0037] a computing unit, configured to calculate corresponding imaging features in combination with the CMB detection result and the whole lesion segmentation result, wherein the imaging features include brain region distribution, number, shape and size.

[0038] With reference to the second aspect, in a first possible implementation manner of the second aspect, the processing content of the CMB detection module includes:

[0039] inputting the T1 modality image into a brain region segmentation network to obtain a brain region segmentation result;

[0040] inputting the SWI image into a CMB segmentation network to obtain a CMB segmentation result;

[0041] registering the T1 modality image and the brain region segmentation result to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and the brain region segmentation result thereof.

[0042] With reference to the first possible implementation manner of the second aspect, in a second possible implementation manner of the second aspect, the whole lesion segmentation unit is specifically configured to:

[0043] inputting the MR plain scan image into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain an initial whole lesion segmentation result;

[0044] registering the MR plain scan image and the initial whole lesion segmentation result to the SWI image space through a rigid registration algorithm to obtain a whole lesion segmentation result of the SWI modality.

[0045] With reference to the first possible implementation manner of the second aspect, in a third possible implementation manner of the second aspect, the device further includes a correction unit configured to:

[0046] performing N4 bias field correction on the T1 modality image.

[0047] With reference to the first possible implementation manner of the second aspect, in a fourth possible implementation manner of the second aspect, both the input brain region segmentation network and the CMB segmentation network adopt a U-Net framework in a form of encoder-decoder cascade, learn global features of an image through a continuous downsampling module, decode the learned semantic features to an input image space through an upsampling module to obtain a segmentation result, and fuse the position information of a bottom layer with the semantic information of a deep layer through a skip connection module to obtain a more accurate segmentation result.

[0048] With reference to the second aspect, in a fifth possible implementation manner of the second aspect, the device further includes a CMB etiology detection unit configured to:

[0049] The imaging features are input into a pre-configured CMB cause detection module to perform CMB cause detection based on the imaging features, and a CMB cause detection result is obtained.

[0050] In a sixth possible implementation manner of the second aspect of the present application, in combination with the second aspect of the present application, the processing content of the CMB cause detection module includes:

[0051] An ordinal outcome variable nomogram algorithm is used to construct a linear relationship between the preset imaging features and the preset CMB causes, wherein the linear relationship simultaneously fits multiple dichotomous logistic models, and has:

[0052] Prob(y0+y1+y2+y3)=βX+α3

[0053] Prob(y0+y1+y2)=βX+α2

[0054] Prob(y0+y1)=βX+α1

[0055] Prob(y0)=βX+α0

[0056] ………

[0057] Wherein, Prob represents probability, y represents cause, X represents feature, β represents the linear relationship between the constructed feature and cause, and α represents different intercepts.

[0058] In a third aspect, the present application provides a processing device, including a processor and a memory, the memory stores a computer program, and the processor executes the computer program in the memory to perform the method provided by the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0059] In a fourth aspect, the present application provides a computer readable storage medium, which stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method provided by the first aspect of the present application or any possible implementation manner of the first aspect of the present application.

[0060] From the above, the present application has the following beneficial effects:

[0061] For the recognition of CMB, in one aspect, after the T1 modality image and the SWI image are acquired, the T1 modality image and the SWI image are input into the pre-configured CMB detection module to combine the T1 modality image and the SWI image for CMB detection and obtain a CMB detection result. In another aspect, after the MR plain scan image is acquired, the MR plain scan image is input into the pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain a whole lesion segmentation result. At this time, the corresponding imaging features are calculated by combining the CMB detection result and the whole lesion segmentation result of the two aspects. In this process, the CMB detection based on the SWI image is further combined with the whole lesion segmentation of the MR plain scan image to provide more accurate imaging features for subsequent CMB diagnosis work. Thus, a set of supporting solutions for more efficient and high-precision automatic recognition needs is provided, thereby meeting the recognition needs of CMB and subsequent diagnosis needs. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0063] Figure 1 A flowchart of a SWI-based cerebral microbleeding image detection method of the present application;

[0064] Figure 2 A scene diagram of the processing content of the CMB detection module of the present application;

[0065] Figure 3 A network architecture diagram of the U-Net of the present application;

[0066] Figure 4 A scene diagram of calculating imaging features of the present application;

[0067] Figure 5 A scene diagram of CMB detection of the present application;

[0068] Figure 6 A scene diagram of linear relationship of the present application;

[0069] Figure 7 A structure diagram of a SWI-based cerebral microbleeding image detection device of the present application;

[0070] Figure 8 A structure diagram of a processing device of the present application. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative work fall within the scope of protection of the present application.

[0072] The terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or modules does not have to be limited to only those steps or modules clearly listed, but can include other steps or modules that are not clearly listed or inherent to the process, method, product, or device. The naming or numbering of the steps appearing in the present application does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering, and the named or numbered flow steps can change the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0073] The division of modules appearing in the present application is a logical division, and in actual application, there can be another division manner, for example, a plurality of modules can be combined or integrated in another system, or some features can be ignored or not executed, in addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be through some interface, the indirect coupling or communication connection between the modules can be electrical or other similar forms, which are not limited in the present application. Moreover, the modules or sub-modules described as separate components can or can not be physically separated, can or can not be physical modules, or can be distributed into a plurality of circuit modules, and part or all of the modules can be selected according to actual needs to achieve the purpose of the present application scheme.

[0074] Before introducing the SWI-based brain microhemorrhage image detection method provided by the present application, the background content involved in the present application is first introduced.

[0075] The SWI-based brain microhemorrhage image detection method, device and computer readable storage medium provided by the present application can be applied to a processing device, and provide a set of supporting solutions for more efficient and high-precision automated identification requirements based on SWI, thereby meeting the identification requirements of CMB and subsequent diagnosis requirements.

[0076] The SWI-based cerebral microbleeding image detection method mentioned in the present application can be executed by an SWI-based cerebral microbleeding image detection device, or a server, a physical host, or a user equipment (UE) or other types of processing devices integrated with the SWI-based cerebral microbleeding image detection device. The SWI-based cerebral microbleeding image detection device can be implemented in hardware or software. The UE can be a terminal device such as a smartphone, a tablet computer, a notebook computer, a desktop computer, or a personal digital assistant (PDA). The processing device can be set in a device cluster.

[0077] It should be understood that the present application is mainly configured for the image detection needs of CMB in clinical practice. Therefore, the processing device applying the SWI-based cerebral microbleeding image detection method of the present application is usually a related device in clinical practice. Of course, it is not excluded that the image detection needs of CMB can also be involved in laboratory scenarios, and therefore the SWI-based cerebral microbleeding image detection method provided by the present application can also be applied to meet the image detection needs of CMB.

[0078] Next, the SWI-based cerebral microbleeding image detection method provided by the present application will be introduced.

[0079] First, referring to Figure 1 , Figure 1 Fig. 1 shows a flowchart of the SWI-based cerebral microbleeding image detection method of the present application. The SWI-based cerebral microbleeding image detection method provided by the present application can specifically include the following steps S101 to S104:

[0080] Step S101, acquiring T1 modality images and SWI images;

[0081] For image detection of CMB, the present application configures two aspects of settings, so as to perform data fusion processing later to achieve better CMB image detection effect.

[0082] Here, the acquisition of T1 modality images and SWI images corresponds to the first aspect. The first aspect is to make an initial CMB detection to provide a data reference for higher-precision CMB detection later.

[0083] It can be understood that T1 modality images and SWI images are image data that can be involved in the detection of CMB and other diseases in clinical practice. The detection devices involved in the specific collection source are not the focus of the present application, and therefore are not described in detail.

[0084] Here step S101 is generally a ready data call processing, of course, in the specific operation can also be real-time data acquisition processing, with the actual situation / need adjustment can be.

[0085] Step S102, the T1 modality image and SWI image are input into the pre-configured CMB detection module to combine the T1 modality image and SWI image for CMB detection, and obtain the CMB detection result.

[0086] For the first aspect of CMB image detection, the present application pre-configures a CMB detection module, which can be understood as CMB detection model, CMB detection rule or CMB detection strategy, which is based on both T1 modality image and SWI image to perform preliminary detection of CMB.

[0087] Among them, the specific form of the detection result of the CMB detection module is obviously more flexible, which can achieve the effect of describing the detected CMB.

[0088] For example, the detection result of the CMB detection module can be embodied by a mask graph.

[0089] Step S103, obtaining MR plain scan image, wherein the MR plain scan image includes T1, T2 and FLAIR multi-modality phase;

[0090] In the second aspect of the CMB image detection of the present application, the present application is based on the MR plain scan image, which can also be understood as multi-modality image, which involves T1, T2 and FLAIR multi-modality phase.

[0091] For this MR plain scan image, similar to the previous one, the detection equipment involved in its specific acquisition source is not the focus of the present application, so it is not described.

[0092] Among them, it needs to be understood that the T1 modality image obtained in the previous step S101 can also be the T1 modality image in the MR plain scan image, that is, one of the images obtained by MR plain scan processing, of course, in the specific application, the acquisition of TI modality image in step S101 and step S103 here can be independent of each other, and can be adjusted according to actual needs.

[0093] Similarly, the step S103 here is generally a ready data call processing, of course, in the specific operation can also be real-time data acquisition processing, with the actual situation / need adjustment can be.

[0094] Step S104, inputting the MR plain scan image into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image, and obtaining a whole lesion segmentation result, wherein the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction;

[0095] For the second aspect of CMB image detection, the present application pre-configures a whole lesion segmentation module, which can be understood as a whole lesion segmentation detection model, whole lesion segmentation rules or whole lesion segmentation strategies, etc., which is based on the MR plain scan image to identify the image area of the lesion and perform segmentation.

[0096] Wherein, the specific form of the result of the whole lesion segmentation module is obviously more flexible, and can achieve the effect of describing the detected CMB.

[0097] For example, the whole lesion segmentation result of the whole lesion segmentation module can be embodied by a mask image.

[0098] Step S105, combining the CMB detection result and the whole lesion segmentation result to calculate the corresponding imaging features, wherein the imaging features include brain region distribution, number, shape and size.

[0099] After obtaining the processing results of the two aspects, i.e. the CMB detection result and the whole lesion segmentation result, the two can be used as data sources for the calculation and processing of the imaging features of CMB, and in the calculation process, it is easy to understand that the content of the other aspect can be used for reference, so that the imaging features related to CMB can be more accurately captured.

[0100] Wherein, it should be understood that the imaging features involved here are not only the imaging features of CMB, and the lesions identified in the second aspect as mentioned above can also be related to CMB, or the detection of related lesions can also be included in the scope of CMB detection, because the imaging features calculated here are essentially for the service of CMB diagnosis, therefore, the imaging features provided here can provide data reference for subsequent related CMB diagnosis work.

[0101] And in specific operation, the calculation and processing of the imaging features can also involve the application of Brain Observer Microbleed Rating Scale (BOMBS) or Microbleed Anatomical Rating Scale (MARS) result forms, so as to standardize and standardize the imaging feature calculation results here.

[0102] From Figure 1As can be seen from the embodiments shown, for the recognition of CMB, one aspect of the present application is to input the T1 modality image and the SWI image into the pre-configured CMB detection module after obtaining the T1 modality image and the SWI image, to combine the T1 modality image and the SWI image for CMB detection, and obtain the CMB detection result, and another aspect is to input the MR plain scan image into the pre-configured whole lesion segmentation module after obtaining the MR plain scan image, to perform whole lesion segmentation based on the MR plain scan image, and obtain the whole lesion segmentation result, at this time, the corresponding imaging features are calculated by combining the CMB detection result and the whole lesion segmentation result of the two aspects, in this process, the CMB detection performed on the SWI image is further combined with the whole lesion segmentation of the MR plain scan image to provide more accurate imaging features for subsequent CMB diagnosis work, thereby providing a set of supporting solutions for more efficient and high-precision automatic recognition needs, thereby meeting the recognition needs of CMB and subsequent diagnosis needs.

[0103] Continue to describe the above Figure 1 The various steps of the embodiments shown and their possible implementation in practical applications are described in detail.

[0104] As an exemplary implementation, the CMB detection process performed in step S102 above, i.e., the processing content of the CMB detection module, can specifically include:

[0105] inputting the T1 modality image into the brain region segmentation network to obtain the brain region segmentation result;

[0106] inputting the SWI image into the CMB segmentation network to obtain the CMB segmentation result;

[0107] registering the T1 modality image and the brain region segmentation result to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and the brain region segmentation result thereof.

[0108] As can be seen, two networks configured by the present application for CMB detection are involved here, i.e., the brain region segmentation network and the CMB segmentation network, which are respectively used for segmentation processing of the T1 modality image (T1 modality) and the SWI image obtained in step S101, and the unification of the SWI modality is completed through registration later.

[0109] At this time, it is worth noting that the CMB detection result here is the CMB segmentation result and the brain region segmentation result of the SWI modality, which specifically provides two data references. As before, the specific forms of the segmentation results of the two networks are relatively flexible, and can be embodied by mask graphs.

[0110] Specifically, for the processing of the embodiments here, reference can also be made to Figure 2A scenario diagram of the processing content of the CMB detection module of the present application is shown for a more visual understanding.

[0111] Wherein, from Figure 2 It can also be seen that before the T1 modality image is input into the brain region segmentation network to obtain the brain region segmentation result, the method of the present application can further include the following steps:

[0112] The N4 bias field correction is performed on the T1 modality image.

[0113] It can be understood that the N4 bias field correction can also be a N4 bias field correction, which is itself a relatively mature data preprocessing means, which can be used here to improve the data quality of the T1 modality image, make its image content more idealized, and make the image features more accurate and clear, so as to promote the brain region segmentation accuracy and segmentation efficiency.

[0114] At the same time, for the brain region segmentation network and the CMB segmentation network, obviously any type of neural network can be used, such as U-Net or V-Net4 neural network.

[0115] Taking U-Net as an example, in order to realize the extraction of the image features such as the position and size of CMB, the present application uses two segmentation networks to realize the segmentation of brain region and CMB, and the U-Net network is used for the segmentation algorithm, that is, the U-Net framework is used for the input brain region segmentation network and the CMB segmentation network.

[0116] Referring to Figure 3 A network architecture diagram of the U-Net of the present application is shown, and the network architecture of the U-Net is in the form of encoder-decoder cascade, learns the global features of the image through consecutive down-sampling modules, decodes the learned semantic features to the input image space to obtain the segmentation result using up-sampling modules, and uses the skip connection module to fuse the position information of the bottom layer with the semantic information of the deep layer to obtain more accurate segmentation results.

[0117] Wherein, the U-Net is composed of basic convolution layer (convolution), BN layer (batch normalization), ReLU activation function layer and softmax layer, the network framework can efficiently and accurately learn the image features involved in CMB image detection, and well realize the goal of automatic segmentation of brain region and recognition of CMB.

[0118] In addition, for the full lesion segmentation detection processing performed by the step S104, that is, the processing content of the full lesion segmentation module, as another exemplary implementation manner, it can specifically include:

[0119] input the MR plain scan image into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain an initial whole lesion segmentation result;

[0120] The MR plain scan image and the initial whole lesion segmentation result are registered to the SWI image space through a rigid registration algorithm to obtain a whole lesion segmentation result of the SWI modality.

[0121] It can be seen that the whole lesion segmentation module configured by the present application is involved here, which is used to perform whole lesion segmentation on the MR plain scan image obtained in step S103, and the registration is performed later to complete the unification of the SWI modality.

[0122] It is worth noting at this time that the whole lesion segmentation result here also involves registration of the SWI modality, that is, registration to the SWI image space with the MR plain scan image (MR plain scan modality), which corresponds to the processing content of the CMB detection module in the front, facilitating subsequent fusion processing of data based on the SWI modality.

[0123] As before, the specific form of the whole lesion segmentation result is flexible and can be represented by a mask graph.

[0124] After the segmentation processing in the above two aspects (two exemplary implementations), the imaging features (including distribution, number, shape, and size) of CMB, WMH, PVS, LA, and infarction lesions can be calculated based on the CMB segmentation result and the brain region segmentation result of the SWI modality, and the whole lesion segmentation module of the SWI modality.

[0125] Specifically, for the content here, the CMB imaging features can also be calculated based on the CMB segmentation result and the brain region segmentation result of the SWI modality, and the whole lesion segmentation module of the SWI modality. Figure 4 The scene schematic diagram of the present application for calculating imaging features is shown.

[0126] And from the CMB imaging features calculated by the present application, the CMB imaging features can be calculated based on the CMB segmentation result and the brain region segmentation result of the SWI modality, and the whole lesion segmentation module of the SWI modality. Figure 4 Finally, it can be seen that after the imaging features are calculated, the CMB and its related imaging features (associated image markers) recognized by the image can be output to complete a CMB image detection.

[0127] In addition, as mentioned earlier, the image detection of CMB-related imaging features by the present application can serve the subsequent CMB diagnosis work, in which case, after the image detection of CMB-related imaging features is automatically completed, the present application can perform a calculation of CMB-related causes based on the calculated imaging features.

[0128] It can be understood that the specific diagnosis idea can be:

[0129] Firstly, evaluate whether the CMB is lobar CMB, deep CMB or mixed type;

[0130] If it is limited in the lobar, consider the pathogenesis is cerebral amyloid angiopathy (CAA), the hemorrhagic markers of CAA are lobar hemorrhage or cortical surface iron deposition, the non-hemorrhagic markers of CAA include posterior white matter hyperintensity (WMH), perivascular space (PVS) in the center of the semiovale and cortical microinfarction;

[0131] If it is limited in the deep basal ganglia region, consider the pathogenesis is arteriosclerosis, the most common cause is hypertension, the hemorrhagic marker of arteriosclerosis is deep brain hemorrhage, the non-hemorrhagic marker of arteriosclerosis is deep lacuna, WMH in basal ganglia region and PVS in basal ganglia region;

[0132] If it is mixed type, all the above causes may occur, and there is also the possibility of hereditary cerebral small vessel disease, such as autosomal dominant arteriopathy with subcortical infarction and leukoencephalopathy (CADASIL) and the like.

[0133] In this regard, step S105 combines the CMB detection result and the whole lesion segmentation result, calculates the corresponding imaging features, and the method of the application can further include:

[0134] Input the imaging features into the pre-configured CMB cause detection module to detect the cause of the CMB based on the imaging features, and obtain the CMB cause detection result.

[0135] It can be seen that the CMB cause detection module pre-configured herein can be understood as a CMB cause detection model, a CMB cause detection rule or a CMB cause detection strategy, which is based on the imaging features to detect the cause of the CMB.

[0136] Obviously, the specific form of the detection result of the CMB cause detection module is flexible, which can achieve the effect of describing the detected CMB cause.

[0137] In the process of predicting the cause of CMB by statistical, quantitative, fitting and other data processing means, clinical information (such as age, gender, medical history, etc.) can also be applied, not only relying on the imaging features calculated above.

[0138] For this, in addition to the above CMB image detection and CMB cause detection, reference can also be made to Figure 5 A scenario diagram of the CMB detection of the present application is shown for a more intuitive understanding, wherein the multi-modal image feature extraction module is obviously for the processing of the imaging features of step S105, which includes the full lesion segmentation module mentioned above.

[0139] For the cause prediction in the embodiment, that is, the causes involved in CMB cause detection, in actual operation, can specifically include cerebral amyloid angiopathy, hypertension, moyamoya disease, hereditary cerebral small vessel disease, infective endocarditis, critical illness and trauma, etc.

[0140] It can be seen that the common causes of CMB in clinical practice are mainly divided into two types: CAA and hypertension (HTN), and based on the high-precision imaging features obtained above, in addition to analyzing the differentiation of CMB caused by CAA and hypertension, further other relatively rare causes such as hereditary cerebral small vessel disease and infective endocarditis can be analyzed, which is helpful to capture relatively rare or easily ignored causes, thereby promoting better CMB diagnosis.

[0141] And for the processing content of the CMB cause detection module, specifically can include the following contents:

[0142] An ordered outcome variable nomogram algorithm is used to construct a linear relationship between the preset imaging features and the preset CMB causes, wherein the linear relationship simultaneously fits multiple dichotomous logistic models, and has:

[0143] Prob(y0)=βX+α0

[0144] Prob(y0+y1)=βX+α1

[0145] Prob(y0+y1+y2)=βX+α2

[0146] Prob(y0+y1+y2+y3)=βX+α3

[0147] ………

[0148] Wherein, Prob represents probability, y represents cause, X represents feature, β represents the linear relationship between the constructed feature and cause, and α represents different intercepts.

[0149] Through the linear relationship (probability prediction model) of the categories y0, y0+y1, y0+y1+y2, …, the possibility of all etiological categories of CMB can be exhausted, and thus the ordered outcome variable nomogram can be drawn according to the linear combination of the independent variables fitted by the model, so as to facilitate the calculation and presentation of the probabilities of various etiologies of CMB, and further to better assist the diagnosis of CMB.

[0150] In addition, the linear relationship (probability prediction model) of the etiology of CMB can also be combined with the following aspects. Figure 6 A scenario diagram of the linear relationship of the present application is shown, which can be used to more intuitively understand the linear relationship (probability prediction model) of the etiology of CMB.

[0151] The etiological prediction result (probability form) obtained by the linear relationship (probability prediction model) can obtain the probability score corresponding to each group of features, so as to achieve the effect of accurate quantification.

[0152] For example, after obtaining the probability scores corresponding to all groups of features, all the scores can be added, and the final total score is the probability corresponding to all etiological possibilities.

[0153] In addition, for the input features involved in the linear relationship (probability prediction model) (in addition to the image features calculated above, relevant clinical information can also be included), the present application can also involve feature screening processing to determine the features adopted in specific work.

[0154] For example, in order to analyze the correlation between the initial features adopted and the etiology of CMB, the present application can use the LASSO feature selection algorithm to screen out effective features, and the screened features have higher correlation with the etiology of CMB, so that the linear relationship (probability prediction model) constructed thereby is more practical.

[0155] Of course, in specific operations, feature extraction algorithms other than the LASSO feature selection algorithm can also be used to determine more targeted features from the initial features adopted, which can be configured as needed.

[0156] The above is the introduction of the SWI-based cerebral microbleeding image detection method provided by the present application. In order to better implement the SWI-based cerebral microbleeding image detection method provided by the present application, the present application also provides an SWI-based cerebral microbleeding image detection device from the perspective of functional modules.

[0157] Referring to Figure 7 , Figure 7 A structure diagram of the SWI-based cerebral microbleeding image detection device of the present application is shown. In the present application, the SWI-based cerebral microbleeding image detection device 700 can specifically include the following structures:

[0158] The acquisition unit 701 is configured to acquire T1 modality images and SWI images.

[0159] The CMB detection unit 702 is configured to input the T1 modality images and the SWI images into a pre-configured CMB detection module to perform CMB detection in combination with the T1 modality images and the SWI images, and obtain a CMB detection result.

[0160] The acquisition unit 701 is further configured to acquire MR plain scan images, wherein the MR plain scan images include multi-modality phases of T1, T2 and FLAIR.

[0161] The full lesion segmentation unit 703 is configured to input the MR plain scan images into a pre-configured full lesion segmentation module to perform full lesion segmentation based on the MR plain scan images, and obtain a full lesion segmentation result, wherein the lesions involved in the full lesion segmentation include WMH, PVS, LA and infarction.

[0162] The calculation unit 704 is configured to calculate corresponding imaging features in combination with the CMB detection result and the full lesion segmentation result, wherein the imaging features include brain region distribution, quantity, morphology and size.

[0163] In an exemplary implementation, the processing content of the CMB detection module includes:

[0164] inputting the T1 modality images into a brain region segmentation network to obtain a brain region segmentation result;

[0165] inputting the SWI images into a CMB segmentation network to obtain a CMB segmentation result;

[0166] registering the T1 modality images and the brain region segmentation result to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and the brain region segmentation result thereof.

[0167] In another exemplary implementation, the full lesion segmentation unit 703 is specifically configured to:

[0168] inputting the MR plain scan images into a pre-configured full lesion segmentation module to perform full lesion segmentation based on the MR plain scan images, and obtain an initial full lesion segmentation result;

[0169] registering the MR plain scan images and the initial full lesion segmentation result to the SWI image space through a rigid registration algorithm to obtain a full lesion segmentation result of the SWI modality.

[0170] In another exemplary implementation, the apparatus further includes a correction unit 705 configured to:

[0171] perform N4 bias field correction on the T1 modality images.

[0172] In yet another exemplary implementation, both the input brain region segmentation network and the CMB segmentation network adopt a U-Net framework in the form of an encoder-decoder cascade, learn global features of the image through a continuous downsampling module, decode the learned semantic features to the input image space through an upsampling module to obtain a segmentation result, and fuse the position information of the bottom layer with the semantic information of the deep layer through a skip connection module to obtain a more accurate segmentation result.

[0173] In yet another exemplary implementation, the apparatus further includes a CMB cause detection unit 706 configured to:

[0174] input the imaging features into a pre-configured CMB cause detection module to perform CMB cause detection based on the imaging features, and obtain a CMB cause detection result.

[0175] In yet another exemplary implementation, the processing content of the CMB cause detection module includes:

[0176] adopting an ordinal outcome variable nomogram algorithm to construct a linear relationship between the preset imaging features and the preset CMB causes, wherein the linear relationship simultaneously fits multiple dichotomous logistic models, and has:

[0177] Prob(y0+y1+y2+y3) = βX + α3

[0178] Prob(y0+y1+y2+y3) = βX + α3

[0179] Prob(y0+y1+y2+y3) = βX + α3

[0180] Prob(y0+y1+y2+y3) = βX + α3

[0181] ………

[0182] wherein Prob represents probability, y represents cause, X represents feature, β represents the linear relationship between the constructed feature and cause, and α represents different intercepts.

[0183] The present application also provides a processing device from the hardware structure, refer to Figure 8 , Figure 8 a structural schematic diagram of the processing device of the present application is shown, specifically, the processing device of the present application can include a processor 801, a memory 802 and an input and output device 803, the processor 801 is used to execute the computer program stored in the memory 802 to realize the functions of the above Figure 1 the steps of the SWI-based cerebral microbleeding image detection method in the corresponding embodiment; or the processor 801 is used to execute the computer program stored in the memory 802 to realize the functions of the above Figure 7 the functions of each unit in the corresponding embodiment, the memory 802 is used to store the computer program executed by the processor 801Figure 1 Computer program required for the SWI-based brain microbleed image detection method in the corresponding embodiments.

[0184] For example, the computer program can be divided into one or more modules / units, one or more modules / units are stored in the memory 802 and executed by the processor 801 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device.

[0185] The processing device can include, but is not limited to, the processor 801, the memory 802, the input and output device 803. Those skilled in the art can understand that the schematic is only an example of the processing device, and does not constitute a limitation on the processing device, and can include more or less components than the schematic, or combine certain components, or different components, for example, the processing device can also include a network access device, a bus, etc., and the processor 801, the memory 802, the input and output device 803 are connected through the bus.

[0186] The processor 801 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the processing device, which connects all parts of the device through various interfaces and lines.

[0187] The memory 802 can be used to store computer programs and / or modules, and the processor 801 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 802, and calling data stored in the memory 802. The memory 802 can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, application programs required by at least one function, etc.; and the data storage area can store data created according to the use of the processing device, etc. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0188] When the processor 801 is used to execute the computer programs stored in the memory 802, the following functions can be specifically implemented:

[0189] The T1 modality image and the SWI image are input into a pre-configured CMB detection module to combine the T1 modality image and the SWI image to perform CMB detection, and obtain a CMB detection result;

[0190] An MR plain scan image is acquired, wherein the MR plain scan image includes multiple modality phases of T1, T2 and FLAIR;

[0191] The MR plain scan image is input into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image, and obtain a whole lesion segmentation result, wherein the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction;

[0192] The corresponding imaging features are calculated in combination with the CMB detection result and the whole lesion segmentation result, wherein the imaging features include brain region distribution, quantity, morphology and size.

[0193] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described SWI-based cerebral microbleeding image detection device, processing device and corresponding units thereof can be referred to as Figure 1 The description of the SWI-based cerebral microbleeding image detection method in the corresponding embodiments will not be repeated here.

[0194] Those skilled in the art can understand that all or part of the steps in the various methods of the above-described embodiments can be completed by instructions, or by controlling related hardware by the instructions, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0195] To this end, the present application provides a computer readable storage medium, wherein a plurality of instructions are stored, the instructions can be loaded by a processor to execute the present application as follows Figure 1 Steps of the SWI-based cerebral microbleed image detection method in the corresponding embodiments can be specifically implemented with reference to the description of the SWI-based cerebral microbleed image detection method in the corresponding embodiments as follows Figure 1 The description of the SWI-based cerebral microbleed image detection method in the corresponding embodiments is not repeated here.

[0196] The computer readable storage medium can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or the like.

[0197] The instructions stored in the computer readable storage medium can execute the present application as follows Figure 1 Steps of the SWI-based cerebral microbleed image detection method in the corresponding embodiments, so the present application as follows Figure 1 The beneficial effects of the SWI-based cerebral microbleed image detection method in the corresponding embodiments can be achieved, as described above, and are not repeated here.

[0198] The SWI-based cerebral microbleed image detection method, device, processing equipment, and computer readable storage medium provided by the present application are described in detail above. The principles and implementation modes of the present application are described by applying specific examples. The above description of the embodiments is only to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation modes and application ranges can be changed according to the idea of the present application. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A SWI-based cerebral microbleed image detection method, characterized in that, The method comprises: acquiring T1 modality images and SWI images; inputting the T1 modality images and the SWI images into a pre-configured CMB detection module to combine the T1 modality images and the SWI images for CMB detection and obtain a CMB detection result; acquiring MR plain scan images, wherein the MR plain scan images comprise multiple modality phases of T1, T2 and FLAIR; inputting the MR plain scan images into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan images and obtain a whole lesion segmentation result, wherein the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction; combining the CMB detection result and the whole lesion segmentation result to calculate corresponding imaging features, wherein the imaging features include brain region distribution, quantity, morphology and size; the processing content of the CMB detection module comprises: inputting the T1 modality images into a brain region segmentation network to obtain a brain region segmentation result; inputting the SWI images into a CMB segmentation network to obtain a CMB segmentation result; registering the T1 modality images and the brain region segmentation result to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and its brain region segmentation result; the inputting of the MR plain scan images into the pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan images and obtain a whole lesion segmentation result comprises: the inputting of the MR plain scan images into the pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan images and obtain an initial whole lesion segmentation result; registering the MR plain scan images and the initial whole lesion segmentation result to the SWI image space through the rigid registration algorithm to obtain the whole lesion segmentation result of the SWI modality; both the brain region segmentation network and the CMB segmentation network adopt a U-Net framework in the form of an encoder-decoder cascade, learn global features of images through a continuous downsampling module, decode the learned semantic features to the input image space to obtain a segmentation result through an upsampling module, and fuse the position information at the bottom layer with the semantic information at the deep layer through a skip connection module to obtain a more accurate segmentation result.

2. The method of claim 1, wherein, Before the inputting of the T1 modality images into the brain region segmentation network to obtain a brain region segmentation result, the method further comprises: performing N4 bias correction on the T1 modality images.

3. The method of claim 1, wherein, After the combining of the CMB detection result and the whole lesion segmentation result to calculate corresponding imaging features, the method further comprises: inputting the imaging features into a pre-configured CMB etiology detection module to perform CMB etiology detection based on the imaging features and obtain a CMB etiology detection result.

4. The method of claim 1, wherein, the processing content of the CMB etiology detection module comprises: adopting an ordinal outcome variable nomogram algorithm to construct a linear relationship between pre-set imaging features and pre-set CMB etiologies, wherein the linear relationship simultaneously fits multiple dichotomous logistic models, and there are: Wherein, Prob represents probability, y represents cause, X represents feature, β represents linear relationship between constructed feature and cause, and α represents different intercepts.

5. A cerebral microbleed image detection apparatus based on SWI, characterized by, The device comprises: An acquisition unit configured to acquire a T1 modality image and a SWI image; A CMB detection unit configured to input the T1 modality image and the SWI image into a pre-configured CMB detection module to perform CMB detection in combination with the T1 modality image and the SWI image and obtain a CMB detection result; The acquisition unit is further configured to acquire an MR plain scan image, wherein the MR plain scan image comprises multiple modality phases of T1, T2 and FLAIR; A whole lesion segmentation unit configured to input the MR plain scan image into a pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain a whole lesion segmentation result, wherein the lesions involved in the whole lesion segmentation include WMH, PVS, LA and infarction; A calculation unit configured to calculate corresponding imaging features in combination with the CMB detection result and the whole lesion segmentation result, wherein the imaging features include brain region distribution, quantity, morphology and size; The processing content of the CMB detection module comprises: inputting the T1 modality image into a brain region segmentation network to obtain a brain region segmentation result; inputting the SWI image into a CMB segmentation network to obtain a CMB segmentation result; registering the T1 modality image and the brain region segmentation result to the SWI image space of the CMB segmentation result through a rigid registration algorithm to obtain the CMB segmentation result of the SWI modality and its brain region segmentation result; The whole lesion segmentation unit is specifically configured to: input the MR plain scan image into the pre-configured whole lesion segmentation module to perform whole lesion segmentation based on the MR plain scan image and obtain an initial whole lesion segmentation result; register the MR plain scan image and the initial whole lesion segmentation result to the SWI image space through the rigid registration algorithm to obtain the whole lesion segmentation result of the SWI modality; Both the brain region segmentation network and the CMB segmentation network adopt a U-Net framework in the form of an encoder-decoder cascade, learn global features of an image through a continuous downsampling module, decode the learned semantic features to an input image space through an upsampling module to obtain a segmentation result, and fuse the position information of a bottom layer with the semantic information of a deep layer through a skip connection module to obtain a more accurate segmentation result.

6. A processing device, characterized by The device comprises a processor and a memory, the memory stores a computer program, and the processor executes the computer program in the memory to perform the method of any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a plurality of instructions, which are suitable for being loaded by a processor to execute the method of any one of claims 1 to 4.

Images