Method, device, equipment, storage medium and program product for processing CT image

Abstract

The present disclosure provides a CT image processing method, device, equipment, storage medium and program product, relating to deep learning technology, AI medical technology, comprising: determining a lung parenchyma region in a computed tomography (CT) image, and cutting out an initial lung parenchyma image from the CT image according to the lung parenchyma region; processing the initial lung parenchyma image to obtain a lesion segmentation mask; processing the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image; determining a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region. In this implementation, the lesion region is identified by segmentation and detection, and the target lesion region and the target lesion mask inside the target lesion region are obtained by fusing the two identification results, thereby reducing the probability of missed detection.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to a deep learning technology in artificial intelligence technology, an AI medical technology, and in particular to a CT image processing method and device, equipment, a storage medium and a program product. BACKGROUND

[0002] In recent years, deep learning (DL) and convolutional neural networks (CNNs) have been widely applied. When deep learning technology and convolutional neural networks are applied in the medical field, they can assist medical personnel in analyzing computed tomography (CT images).

[0003] In one application scenario, a model for identifying lung lesions can be pre-trained, and a CT image is input into the model to output a lesion region in the CT image. However, in this implementation scheme, the possibility of missing detection is high, resulting in missing of a real lesion region. SUMMARY

[0004] The present disclosure provides a CT image processing method, device, equipment, storage medium and program product for more accurately identifying a region that may have a lesion in a CT image.

[0005] According to a first aspect of the present disclosure, a computed tomography processing method is provided, comprising:

[0006] determining a lung parenchyma region in the computed tomography (CT) image, and cropping an initial lung parenchyma image from the CT image according to the lung parenchyma region; wherein the initial lung parenchyma image includes the lung parenchyma region;

[0007] processing the initial lung parenchyma image to obtain a lesion segmentation mask, the lesion segmentation mask being used to represent a lesion position in the initial lung parenchyma image; processing the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image;

[0008] determining a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent a lesion position in the initial lung parenchyma image.

[0009] According to a second aspect of the present disclosure, a computed tomography processing device is provided, comprising:

[0010] A lung parenchyma determination unit is configured to determine a lung parenchyma region in the computed tomography (CT) image, and crop an initial lung parenchyma image from the CT image according to the lung parenchyma region, wherein the initial lung parenchyma image comprises the lung parenchyma region.

[0011] A segmentation unit is configured to process the initial lung parenchyma image to obtain a lesion segmentation mask, wherein the lesion segmentation mask is used to represent a lesion position in the initial lung parenchyma image.

[0012] A detection unit is configured to process the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image.

[0013] A fusion unit is configured to determine a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent a lesion position in the initial lung parenchyma image.

[0014] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0015] at least one processor; and

[0016] a memory connected with the at least one processor; wherein

[0017] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect.

[0018] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to perform the method according to the first aspect.

[0019] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising: a computer program stored in a readable storage medium, and at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to enable the electronic device to perform the method according to the first aspect.

[0020] The processing method, device, equipment, storage medium and program product of the CT image provided by the present disclosure include: determining a lung parenchyma region in a computed tomography (CT) image, and cutting out an initial lung parenchyma image from the CT image according to the lung parenchyma region; wherein the initial lung parenchyma image includes the lung parenchyma region; processing the initial lung parenchyma image to obtain a lesion segmentation mask, the lesion segmentation mask being used to represent a lesion position in the initial lung parenchyma image; processing the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image; determining a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent the lesion position in the initial lung parenchyma image. In this embodiment, the lung parenchyma image is directly cut out from the CT image, so that the lung parenchyma image saves rich information of the lung parenchyma region in the CT image, and the lesion region of the lung parenchyma region is more accurately identified. Moreover, the scheme provided by the present disclosure identifies the lesion region by segmentation and detection, and fuses the two identification results to obtain the target lesion region and the target lesion mask inside the target lesion region. In this way, the lesion region and the lesion mask can be more accurately identified in the CT image, so as to improve the recall rate of the lesion region and the lesion mask and reduce the probability of missed detection.

[0021] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0023] Figure 1 A flowchart of a processing method of computed tomography is shown for an exemplary embodiment of the present disclosure;

[0024] Figure 2 A schematic diagram of a CT image and a lung parenchyma region is shown for an exemplary embodiment of the present disclosure;

[0025] Figure 3 A flowchart of a processing method of computed tomography is shown for an exemplary embodiment of the present disclosure;

[0026] Figure 4 A flowchart of processing a CT image is shown for an exemplary embodiment of the present disclosure;

[0027] Figure 5 A structural schematic diagram of a processing device of computed tomography is shown for an exemplary embodiment of the present disclosure;

[0028] Figure 6 A structural schematic diagram of a processing device for computer tomography according to another exemplary embodiment of the present disclosure is shown.

[0029] Figure 7 A block diagram of an electronic device for implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0030] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding them. These should be considered in their context only. Thus, those of ordinary skill in the art will recognize the various changes and modifications of the embodiments described herein, without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and structures are omitted in the following description for the sake of clarity and conciseness.

[0031] With the development of deep learning and convolutional neural network technology, the technology can be applied in the medical field, and specifically, medical personnel can analyze CT images with the assistance of deep learning and convolutional neural network technology.

[0032] In some relatively simple auxiliary analysis systems, a pre-trained neural network can be directly used to process CT images to obtain lesion analysis results.

[0033] In some complex auxiliary analysis systems, the lung region can be extracted from the CT image to obtain a lung CT image, the lung lobe CT image can be segmented in the lung CT image, and the abnormal region mask can be segmented in the lung lobe CT image in multiple ways. The abnormal region masks obtained in different ways are fused to obtain a fused abnormal region.

[0034] However, the above-mentioned way of segmenting the abnormal region in the CT image has a high possibility of missing detection, which further leads to the omission of the real lesion region in the output abnormal region image.

[0035] To solve the above technical problems, in the scheme provided by the present disclosure, the lung parenchyma region in the CT image is processed, the lesion segmentation mask is segmented, and the predicted lesion region is identified, so as to determine the target lesion region and the target lesion mask by combining the lesion segmentation mask and the predicted lesion region. In the scheme provided by the present disclosure, not only the segmentation method is used to segment the lesion mask, but also the method of detecting the lesion region is combined, and the detection is realized by using the detection and segmentation methods. The recall rate of the lesion region and the lesion mask is higher, and the probability of missing detection is lower.

[0036] Figure 1 A flowchart of a processing method for computer tomography according to an exemplary embodiment of the present disclosure is shown.

[0037] AsFigure 1 As shown, the method for processing computed tomography images provided in this disclosure includes:

[0038] Step 101: Determine the lung parenchyma region in the computed tomography CT image, and crop out the initial lung parenchyma image from the CT image based on the lung parenchyma region.

[0039] The method provided in this disclosure can be executed by an electronic device with computing capabilities, such as a computer.

[0040] In one alternative implementation, the computer may be connected to a medical device for capturing CT images, which can send the CT images to the computer so that the computer can analyze the CT images based on the methods provided in this disclosure and output the analysis results.

[0041] Specifically, a computer can acquire CT images and determine the lung parenchyma region within them. The computer may include a lung parenchyma segmentation module, which can be used to process the CT images, identify voxels belonging to and not belonging to the lung parenchyma region, and determine connected regions based on the voxels belonging to the lung parenchyma region, thereby locating the lung parenchyma region.

[0042] Furthermore, an initial lung parenchyma image can be cropped from CT images based on the lung parenchyma region. For example, the lung parenchyma region in the image can be cropped as the initial lung parenchyma image.

[0043] In practical applications, the cropped lung parenchyma image includes the lung parenchyma region from the CT image.

[0044] Figure 2 This is a schematic diagram of a CT image and a lung parenchyma region, illustrating an exemplary embodiment of the present disclosure.

[0045] like Figure 2 As shown, the lung parenchyma region is identified in CT image 21, and lung parenchyma image 22 is obtained by cropping based on the lung parenchyma region.

[0046] Step 102: Process the initial lung parenchyma image to obtain a lesion segmentation mask; process the initial lung parenchyma image to obtain the predicted lesion region in the initial lung parenchyma image.

[0047] The initial lung parenchyma image can be segmented and detected separately. For example, a mask segmentation module and a lesion detection module can be set up separately. The mask segmentation module processes the initial lung parenchyma image to obtain a lesion segmentation mask, and the lesion detection module processes the initial lung parenchyma image to obtain the predicted lesion region.

[0048] Specifically, the voxel points belonging to the lesion region and the voxel points not belonging to the lesion region can be identified in the initial lung parenchyma image, and then a lesion segmentation mask is obtained. The voxel points belonging to the lesion region can be marked as 1, and the voxel points not belonging to the lesion region can be marked as 0, and then the lesion segmentation mask is obtained. The lesion segmentation mask can represent the lesion position in the initial lung parenchyma image.

[0049] Further, the region in which the lesion may exist can also be identified in the initial lung parenchyma image. For example, if it is identified that a certain region may have a lesion, a detection box or a detection frame corresponding to the region can be output, and the detection box or the detection frame can wrap the lesion region.

[0050] In actual application, the predicted lesion region is used to represent the region in which the lesion may exist.

[0051] In step 103, a target lesion region and a target lesion mask corresponding to the target lesion region are determined according to the lesion segmentation mask and the predicted lesion region, and the target lesion mask is located in the target lesion region.

[0052] The target lesion mask is used to represent the lesion position in the initial lung parenchyma image.

[0053] Specifically, the CT image can be analyzed by combining the lesion segmentation mask and the predicted lesion region, and the target lesion region and the target lesion mask corresponding thereto are obtained.

[0054] In an optional implementation, a positive judgment module can also be provided, and each predicted lesion region can be input into the positive judgment module, so as to determine whether the predicted lesion region really has a lesion region, and then the predicted lesion region is confirmed again.

[0055] If it is determined that the judgment result of the predicted lesion region is positive, the predicted lesion region can be determined as the target lesion region, and then the target lesion mask of each target lesion region is obtained by fusing the target lesion region and the lesion segmentation mask.

[0056] In another implementation, the predicted lesion region and the lesion segmentation mask can also be fused directly. For example, if the predicted lesion region of the CT image also has the lesion segmentation mask, it is proved that the region in which the lesion may exist is determined by two ways, and therefore, the predicted lesion region can be determined as the target lesion region, and the lesion segmentation mask in the target lesion region can be taken as the target lesion mask corresponding thereto.

[0057] The method for processing computed tomography (CT) images disclosed herein includes: determining a lung parenchyma region in a CT image and cropping an initial lung parenchyma image from the CT image based on the lung parenchyma region; wherein the initial lung parenchyma image includes the lung parenchyma region; processing the initial lung parenchyma image to obtain a lesion segmentation mask, the lesion segmentation mask being used to characterize the lesion location in the initial lung parenchyma image; processing the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image; determining a target lesion region and a target lesion mask corresponding to the target lesion region based on the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located within the target lesion region and is used to characterize the lesion location in the initial lung parenchyma image. In this embodiment, the lung parenchyma image is directly cropped from the CT image, allowing the lung parenchyma image to retain rich information about the lung parenchyma region in the CT image, thereby more accurately identifying the lesion region within the lung parenchyma region. Moreover, the method provided in this disclosure identifies the lesion region through segmentation and detection, and merges the results of these two identifications to obtain the target lesion region and the target lesion mask within the target lesion region. In this way, the lesion region and lesion mask can be identified more accurately in CT images, thereby improving the recall rate of the lesion region and lesion mask and reducing the probability of missed detection.

[0058] Figure 3 This is a schematic flowchart illustrating a computed tomography processing method as an exemplary embodiment of the present disclosure.

[0059] like Figure 3 As shown, the method for processing computed tomography images provided in this disclosure includes:

[0060] Step 301: Process the CT images using a preset lung parenchyma extraction model to obtain lung parenchyma segmentation results.

[0061] In this context, a lung parenchyma extraction model can be pre-set in electronic devices to identify lung parenchyma regions in CT images. The lung parenchyma extraction model refers to a pre-trained 3D U-shaped neural network (such as U-Net, U2Net, etc.), which is trained to determine whether each voxel belongs to either "background" or "lung parenchyma" based on the input CT image, thus performing semantic segmentation of the lung parenchyma region.

[0062] Specifically, a lung parenchyma extraction model can be used to segment CT images to obtain lung parenchyma segmentation results. The lung parenchyma segmentation results can be a lung parenchyma segmentation mask, which can include voxel points in the CT image that belong to the lung parenchyma and voxel points that do not belong to the lung parenchyma.

[0063] Step 302: Determine the third lung parenchyma image in the CT images based on the lung parenchyma segmentation results.

[0064] Further, the lung parenchyma segmentation result is used to represent which voxel points in the preliminary determined CT image belong to the lung parenchyma region and which voxel points do not belong to the lung parenchyma region, and then a third lung parenchyma image can be determined in the CT image based on the result.

[0065] In actual application, the third lung parenchyma image can be a cuboid including the preliminary lung parenchyma region. The third lung parenchyma image can be cut out from the CT image according to the lung parenchyma segmentation result.

[0066] In this case, the size of the CT image can be adjusted to a first size, and then the adjusted CT image is input into the lung parenchyma extraction model to obtain the lung parenchyma segmentation result. At this time, the lung parenchyma segmentation result obtained is also of the first size, which is different from the size of the CT image.

[0067] The first size can be the size requirement of the input data of the lung parenchyma extraction model. Therefore, the method provided by the present disclosure can obtain the lung parenchyma segmentation result by using the lung parenchyma extraction model.

[0068] In another optional implementation, the CT value range of the CT image can also be adjusted to reduce the data in the CT image and improve the processing speed of the lung parenchyma extraction model on the input image.

[0069] For example, the original CT value range can be cropped to [HU min 1, HU max 1] and normalized to [0, 1]:

[0070]

[0071] HU is the CT value in the CT image.

[0072] Specifically, if the size of the lung parenchyma segmentation result is different from the size of the CT image, the size of the lung parenchyma segmentation result can be adjusted to the size of the CT image, and then a third lung parenchyma image can be determined in the CT image according to the lung parenchyma segmentation result with the adjusted size.

[0073] Further, the lung parenchyma segmentation result can be a lung parenchyma mask, and a third lung parenchyma image can be directly determined in the CT image according to the lung parenchyma mask with the same size as the CT image. Specifically, the third lung parenchyma image can be cropped in the CT image according to the lung parenchyma mask.

[0074] The range of the third lung parenchyma image can be expanded based on the lung parenchyma mask.

[0075] In this implementation, the CT image can be processed by using the lung parenchyma extraction model, and the third lung parenchyma image can also be determined when the size of the input data of the lung parenchyma extraction model is different from the original size of the CT image.

[0076] In step 303, the third lung parenchyma image is processed by using the preset lung lobe segmentation model to obtain a lung lobe segmentation result.

[0077] In the electronic device, a lung lobe segmentation model can also be set, which is used to identify lung lobe regions in the image.

[0078] The lung lobe segmentation model refers to a pre-trained 3D U-shaped neural network, which is trained to identify each voxel as one of six categories (background, left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe).

[0079] Specifically, the lung lobe segmentation model can be used to process the third lung parenchyma image to identify the voxel points belonging to each lung lobe and the background voxel points in the third lung parenchyma image, and further identify the lung lobe regions in the third lung parenchyma image.

[0080] Further, the lung lobes can include the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe. The voxel points belonging to the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe can be determined in the third lung parenchyma image, and further the regions of each lung lobe can be identified in the third lung parenchyma image.

[0081] In actual application, the lung lobe segmentation result output by the lung lobe segmentation model can include segmentation results of multiple lung lobes, such as segmentation results of the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe.

[0082] The lung lobe segmentation result can be a lung lobe segmentation mask, which includes voxel points belonging to lung lobes.

[0083] In step 304, the lung parenchyma region is determined in the CT image according to the lung lobe segmentation result; the lung parenchyma region includes the regions of the lung lobes.

[0084] The lung parenchyma region includes the regions of the lung lobes, so the lung parenchyma region can be determined in the CT image according to the lung lobe segmentation results.

[0085] Specifically, the regions of the lung lobes can be taken as the lung parenchyma region. For example, the regions of the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe can be taken as the lung parenchyma region.

[0086] In this implementation, the lung parenchyma region is preliminarily segmented from the CT image, and then each lung lobe region is more accurately segmented in the region, and further the lung parenchyma region is re-determined according to each lung lobe region, so that the lung parenchyma region determined in the CT image is more accurate.

[0087] In this way, the size of the third lung parenchyma image can be adjusted to the second size first, and then the adjusted third lung parenchyma image is input into the lung lobe segmentation model, and then the lung lobe segmentation result is obtained. At this time, the lung lobe segmentation result obtained is also of the second size, which is different from the size of the third lung parenchyma image.

[0088] The second size can be the size requirement of the input data of the lung lobe segmentation model, and therefore, the method provided by the present disclosure can obtain the lung lobe segmentation result by using the lung lobe segmentation model.

[0089] In another optional implementation, the CT value range of the third lung parenchyma image can also be adjusted to reduce the data in the third lung parenchyma image and improve the processing speed of the lung lobe segmentation model on the input image.

[0090] Specifically, if the size of the lung lobe segmentation result is different from the size of the third lung parenchyma image, the size of the lung lobe segmentation result can be adjusted to the size of the third lung parenchyma image.

[0091] Further, based on the position information of the third lung parenchyma image in the CT image, blank voxels can be supplemented in the periphery of the lung lobe segmentation result with the adjusted size to obtain a lung lobe segmentation result with the same size as the CT image.

[0092] In the lung lobe segmentation result with the same size as the CT image, the position information of the lung lobe is located in the lung parenchyma region in the CT image, and therefore, the region where the lung lobe is located in the CT image can be cut out according to the lung lobe segmentation result.

[0093] In this implementation, the third lung parenchyma image can be processed by using the lung lobe segmentation model, and in the case where the size of the input data of the lung lobe segmentation model is different from the size of the third lung parenchyma image, the lung lobe segmentation result can also be used to cut out the lung lobe part in the original CT image, so as to retain more information of the lung lobe region in the CT image.

[0094] In actual application, the lung lobe segmentation result includes the information of multiple lung lobes, and therefore, multiple lung lobe regions can be determined in the CT image. For example, the left upper lung lobe region, the left lower lung lobe region, the right upper lung lobe region, the right middle lung lobe region, and the right lower lung lobe region can be obtained.

[0095] In this way, after the lung lobe regions are determined, the voxel points at the boundary between two adjacent lung lobes can be determined as the interlobar fissure result.

[0096] Specifically, the computer can also output the interlobar fissure result, which can assist medical personnel in understanding the lung structure in the CT image, so that the medical personnel can make more accurate diagnoses.

[0097] Further, the computer can determine lung lobe regions corresponding to each lung lobe in the CT image according to the lung lobe information in the lung lobe segmentation result.

[0098] Then, each voxel in the lung lobe region is adjusted to a unit volume, and the lung lobe volume is determined according to the number of voxels in the lung lobe region and the original volume of the voxels.

[0099] The unit volume may, for example, be 1mm*1mm*1mm in size. Thereafter, the number of voxels in the lung lobe region can be determined, and the product of the number of voxels and the original volume of the voxels can be taken as the lung lobe volume.

[0100] The lung lobe volume corresponding to each lung lobe region can be determined, and the lung lobe volume can be output, so as to assist medical personnel in diagnosis by means of the lung lobe volume.

[0101] Step 305: cropping an initial lung parenchyma image from the CT image according to the lung parenchyma region; wherein the initial lung parenchyma image includes the lung parenchyma region.

[0102] Step 305 is similar to the implementation of cropping the initial lung parenchyma image in step 101, and will not be described again.

[0103] Step 306: preprocessing the initial lung parenchyma image to obtain a first lung parenchyma image.

[0104] The preprocessing may, for example, include size adjustment processing or CT value range adjustment processing.

[0105] Specifically, the size of the initial lung parenchyma image can be adjusted to a third size, and the original CT value range of the initial lung parenchyma image can be cropped to [HU min 2, HU max 2] and normalized to [0, 1].

[0106] Step 307: processing the first lung parenchyma image by using a preset lesion segmentation model to obtain a first lesion mask.

[0107] After obtaining the first lung parenchyma image, the first lung parenchyma image can be processed by using the preset lesion segmentation model to obtain a first lesion mask.

[0108] The lesion segmentation model can be a pre-trained 3D U-shaped neural network, which is trained to be able to determine, according to an input CT image, whether each voxel point belongs to any of the two categories of “background” and “abnormal lesion”.

[0109] After processing the first lung parenchyma image by using the lesion segmentation model, a first lesion mask of the lung parenchyma image can be obtained, which can include which voxels are background points and which voxels are abnormal lesion points.

[0110] In step 308, a lesion segmentation mask of the initial lung parenchyma image is determined according to the first lesion mask.

[0111] The first lesion mask is a lesion mask of the first lung parenchyma image, and the lesion segmentation mask of the initial lung parenchyma image can be determined according to the first lesion mask.

[0112] Through this implementation, the lesion segmentation mask of the lung parenchyma image can be obtained by using the lesion segmentation model, and then the mask of the lesion region in the lung parenchyma image can be segmented.

[0113] Specifically, if the size of the first lung parenchyma image is different from the size of the initial lung parenchyma image, the size of the first lesion mask can be adjusted to the size of the initial lung parenchyma image when determining the lesion segmentation mask of the initial lung parenchyma image.

[0114] Further, the lesion points in the first lesion mask after size adjustment can be determined according to the lesion probabilities of the voxel points in the first lesion mask after size adjustment. For example, if the lesion probability of a voxel point in the first lesion mask after size adjustment is greater than a probability threshold, the voxel point can be determined as a lesion point.

[0115] In actual application, the lesion segmentation mask in the initial lung parenchyma image can be determined according to the lesion points in the first lesion mask after size adjustment. The connected domain can be determined according to the lesion points, and then the connected domain can be taken as the lesion region in the initial lung parenchyma image.

[0116] Through this implementation, the first lung parenchyma image can be processed by using the lesion segmentation model, and the lesion segmentation mask in the initial lung parenchyma image can also be determined by using the lesion segmentation model when the size of the first lung parenchyma image is different from the size of the initial lung parenchyma image. Therefore, the voxel points that may belong to the lesion can be determined in the initial lung parenchyma image, and more abundant lesion information in the initial lung parenchyma image can be retained.

[0117] In an optional implementation, the lesion connected domain can be determined according to the lesion points in the first lesion mask after size adjustment. For example, the adjacent lesion points can be connected to obtain the lesion connected domain.

[0118] In order to avoid determining the normal region as the lesion region, it can also be judged whether the volume of each connected domain is greater than a volume threshold. If the volume is less than or equal to the volume threshold, the connected domain can be discarded.

[0119] In actual application, the lesion segmentation mask in the initial lung parenchyma image can also be determined according to the lesion connected domain, and the lesion segmentation mask includes multiple lesion connected domains.

[0120] The finally determined connected domain can be determined as a lesion connected domain, and a lesion segmentation mask in the initial lung parenchyma image is obtained. The lesion segmentation mask includes the lesion connected domain in which the lesion may exist. Through this implementation manner, the lesion connected domain can be determined in the first lesion mask after the size is adjusted, so that the lesion segmentation mask in the initial lung parenchyma image is more accurately determined.

[0121] In step 309, the initial lung parenchyma image is preprocessed to obtain a second lung parenchyma image.

[0122] The computer can also preprocess the initial lung parenchyma image to obtain a second lung parenchyma image. The processing manner is similar to that in step 306, but the parameters used can be different. For example, the size of the adjusted second lung parenchyma image is different, and the CT value range is different.

[0123] In step 310, a preset lesion detection model is used to process the second lung parenchyma image to obtain a current lesion region.

[0124] Specifically, the lesion detection model can be preset. The lesion detection model is a pre-trained target detection model based on 2D CNN or 3D CNN. If the target detection model is constructed based on 2D CNN, the second lung parenchyma image is sliced along any direction, and each slice is predicted using the 2D target detection model. If the target detection model is constructed based on 3D CNN, the second lung parenchyma image is predicted after the sliding window is selected.

[0125] Further, the lesion detection model is used to process the input CT image to obtain a lesion region in the CT image. The lesion detection model can output a detection frame or a detection box to mark the lesion region.

[0126] In actual application, the current lesion region output by the lesion detection model refers to the lesion region determined in the second lung parenchyma image.

[0127] In step 311, a predicted lesion region in the initial lung parenchyma image is determined according to the current lesion region.

[0128] The computer can also determine the predicted lesion region in the initial lung parenchyma image according to the current lesion region in the second lung parenchyma image.

[0129] Through this implementation manner, the lesion detection model can be used to determine the predicted lesion region in the initial lung parenchyma image, so that the information of the lesion region in the initial lung parenchyma image is fully retained.

[0130] Specifically, the second lung parenchyma image has a different image size from the initial lung parenchyma image. For example, after preprocessing the initial lung parenchyma image, the second lung parenchyma image has a different size from the initial lung parenchyma image.

[0131] Further, if the second lung parenchyma image has a different image size from the initial lung parenchyma image, when determining the predicted lesion region, the predicted lesion region can be mapped into the initial lung parenchyma image according to the image size of the second lung parenchyma image and the image size of the initial lung parenchyma image, to obtain the predicted lesion region in the initial lung parenchyma image.

[0132] In actual application, the current lesion region can include position information and size information, and the position information and the size information of the current lesion region can be mapped into the size space of the initial lung parenchyma image according to the conversion relationship between the image size of the second lung parenchyma image and the image size of the initial lung parenchyma image, to obtain the predicted lesion region in the initial lung parenchyma image.

[0133] Through this implementation manner, the lesion region that can exist can be accurately predicted in the initial lung parenchyma image.

[0134] In an optional implementation manner, the lesion detection model can be a two-dimensional neural network model, and the current lesion region output by the lesion detection model is a detection box. In this implementation manner, the lesion region that can exist in the initial lung parenchyma image can be obtained based on the detection boxes.

[0135] In another optional implementation manner, the lesion detection model can be a three-dimensional neural network model, and the current lesion region output by the lesion detection model is a detection box. In this implementation manner, the detection result obtained is smoother, so that the final detection result is more accurate.

[0136] The result output by the lesion detection model can further include a confidence, for example, the confidence of each detection box or each detection box can be output, and finally the detection box or detection box with a confidence higher than a threshold value can be selected as the current lesion region.

[0137] Specifically, the detection box can be a cube.

[0138] In step 312, the suspected lesion region is determined according to the lesion segmentation mask, and the possible lesion region is determined according to the suspected lesion region and the predicted lesion region.

[0139] After step 308 and step 311, step 312 can be further performed to determine the suspected lesion region according to the lesion segmentation mask.

[0140] The suspected lesion region can be determined according to the lesion segmentation mask. For example, the suspected lesion region can be determined according to a region belonging to the lesion in the lesion segmentation mask. For example, if a region is marked as a lesion region, a cube in which the lesion region is located can be regarded as the suspected lesion region.

[0141] Specifically, the possible lesion region can be determined in combination with the predicted lesion region output by the lesion detection model and the determined suspected lesion region.

[0142] Further, the determination methods of the predicted lesion region and the suspected lesion region are different, and therefore the results are not completely the same. The predicted lesion region and the suspected lesion region can be fused to obtain the possible lesion region.

[0143] In actual application, the lesion segmentation mask includes a plurality of lesion connected domains, and the lesion connected domains are not connected to each other. Therefore, the suspected lesion region can be generated for each lesion connected domain in the lesion segmentation mask, and the suspected lesion region includes a center point parameter and a size parameter.

[0144] The center point parameter can include three parameters (x1, y1, and z1), and the size parameter can include a parameter d1.

[0145] In this implementation, the suspected lesion region can be obtained from the lesion connected domain, so that the suspected lesion region can wrap the lesion connected domain, and the suspected lesion region in the CT image can be predicted.

[0146] Specifically, the predicted lesion region can also include the center point parameter and the size parameter, and specifically can include a second center point parameter and a second size parameter. The second center point parameter can include three parameters (x2, y2, and z2), and the second size parameter can include a parameter d2.

[0147] Further, the two lesion regions can be fused according to the center point parameters and the size parameters of the two lesion regions.

[0148] In actual application, the region combination can be determined according to the first center point parameters of the suspected lesion regions and the second center point parameters of the predicted lesion regions. The region combination includes one predicted lesion region and one suspected lesion region.

[0149] The matched suspected lesion region and the predicted lesion region can be determined according to the first center point parameters and the second center point parameters. For example, the distance between the first center point parameter and each second center point parameter can be determined, and the combination with the shortest distance is regarded as the region combination.

[0150] After the region combinations are determined, the possible lesion region corresponding to each region combination can be generated.

[0151] Specifically, the center point parameter and the size parameter of the possible lesion region can be determined according to the first center point parameter and the first size parameter of the predicted lesion region and the second center point parameter and the second size parameter of the presumed lesion region.

[0152] For example, the average of the first center point parameter and the second center point parameter can be determined as the center point parameter of the possible lesion region. The average of the first size parameter and the second size parameter can also be determined as the size parameter of the possible lesion region.

[0153] In another implementation, the average of the first center point parameter and the second center point parameter can also be determined as the center point parameter of the possible lesion region, and the maximum of the first size parameter and the second size parameter can be determined as the size parameter of the possible lesion region.

[0154] Through this implementation, the presumed lesion region and the predicted lesion region can be fused to obtain the possible lesion region, and the region where the lesion may exist in the CT image can be more accurately determined.

[0155] Step 313: determining a lesion image in the CT image according to the possible lesion region, and processing each lesion image by using a preset true positive lesion discrimination model to determine a positive lesion image.

[0156] In actual application, the lesion image can be intercepted in the CT image according to the possible lesion region. The lesion image obtained in this way contains the original information in the CT image, so that the subsequent recognition result is more accurate.

[0157] The true positive lesion discrimination model refers to a discrimination model composed of a 3D CNN. The model can identify whether the center point of the input CT image is located inside the lesion region.

[0158] Specifically, in order to avoid that the possible lesion region is a false positive region misrecognized, the scheme provided by the present disclosure can use the true positive lesion discrimination model to further identify each possible lesion region, so as to reduce the result of outputting a false positive region.

[0159] If the identification result of the possible lesion region output by the true positive lesion discrimination model is positive, the possible lesion region can be determined as a positive lesion image, that is, it can be considered that the image indeed includes a lung lesion part.

[0160] Step 314: determining the possible lesion region corresponding to the positive lesion image as a target lesion region.

[0161] Among them, the possible lesion area corresponding to the positive lesion image can be identified as the target lesion area, that is, the target lesion area is considered to include the lesion area.

[0162] Step 315: Determine the target lesion mask corresponding to the target lesion area based on the target lesion area and the lesion segmentation mask.

[0163] Specifically, after identifying the target lesion area containing the lesion, a target lesion mask can also be determined for each target lesion area.

[0164] Furthermore, the lesion segmentation mask obtained from the lesion segmentation model can be combined to determine the target lesion mask in each target lesion region.

[0165] In practical applications, if the target lesion area includes a lesion segmentation mask, then the lesion segmentation mask included in the target lesion area is determined as the target lesion mask corresponding to the target lesion area.

[0166] If the target lesion region includes a segmented lesion mask, then that segmented lesion mask can be identified as the target lesion mask within the target lesion region. In this way, the lesion region and the lesion mask can be effectively fused, outputting an accurate target lesion region and its internal target lesion mask.

[0167] If the target lesion area does not include a lesion segmentation mask, a target lesion mask can be generated based on the target lesion area. The generated target lesion mask is located inside the target lesion area.

[0168] Since the target lesion area is a region that actually includes the lesion, determined by multiple models, if there is no mask segmented by the segmentation model in this area, a lesion segmentation mask can be generated in it to avoid missing the lesion area.

[0169] Furthermore, since the target lesion region has center point parameters and size parameters, a target lesion mask can be generated based on these parameters. Specifically, a sphere located inside the target lesion region can be generated based on the center point parameters and size parameters of the target lesion region to obtain the target lesion mask.

[0170] Since the lesion area is generally spherical, the target lesion mask generated in this way is closer to the actual lesion mask.

[0171] Figure 4 This is a flowchart illustrating CT image processing as an exemplary embodiment of the present disclosure.

[0172] like Figure 4 As shown, the CT image 41 can be processed to identify the lung parenchyma region and then crop out the lung parenchyma image 42.

[0173] The lung parenchyma image 42 can be respectively subjected to segmentation and detection processing to obtain a lesion segmentation mask 43 and a predicted lesion region 44.

[0174] The lesion segmentation mask 43 and the predicted lesion region 44 are fused again to obtain a target lesion region 45 internally including a target lesion mask.

[0175] Figure 5 A structural schematic diagram of a processing device for computer tomography according to an example embodiment of the present disclosure is shown.

[0176] As shown in Figure 5 The processing device for computer tomography 500 provided by the present disclosure includes:

[0177] A lung parenchyma determination unit 510 is configured to determine a lung parenchyma region in the CT image of the computer tomography and crop an initial lung parenchyma image from the CT image according to the lung parenchyma region; wherein the initial lung parenchyma image includes the lung parenchyma region.

[0178] A segmentation unit 520 is configured to process the initial lung parenchyma image to obtain a lesion segmentation mask, which is used to represent a lesion position in the initial lung parenchyma image.

[0179] A detection unit 530 is configured to process the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image.

[0180] A fusion unit 540 is configured to determine a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent a lesion position in the initial lung parenchyma image.

[0181] The processing device for computer tomography provided by the present disclosure directly crops a lung parenchyma image from a CT image, so that the lung parenchyma image retains rich information of a lung parenchyma region in the CT image, and further more accurately identifies a lesion region of the lung parenchyma region. Moreover, the method provided by the present disclosure identifies a lesion region by two ways of segmentation and detection, and fuses the two identification results to obtain a target lesion region and a target lesion mask inside the target lesion region. In this way, the lesion region and the lesion mask can be more accurately identified in the CT image, so as to improve the recall rate of the lesion region and the lesion mask and reduce the probability of missed detection.

[0182] Figure 6 A structural schematic diagram of a processing device for computer tomography according to another example embodiment of the present disclosure is shown.

[0183] As Figure 6 shown, the computer tomography processing device 600 provided by the present disclosure includes a lung parenchyma determination unit 610, which is similar to the lung parenchyma determination unit 510 shown in the first aspect of the present disclosure, a segmentation unit 620, which is similar to the segmentation unit 520 shown in the first aspect of the present disclosure, a detection unit 630, which is similar to the detection unit 530 shown in the first aspect of the present disclosure, and a fusion unit 640, which is similar to the fusion unit 540 shown in the first aspect of the present disclosure. Figure 5 Figure 5 Figure 5 Figure 5

[0184] The segmentation unit 620 includes:

[0185] A first preprocessing module 621, configured to perform preprocessing on the initial lung parenchyma image to obtain a first lung parenchyma image.

[0186] A segmentation module 622, configured to process the first lung parenchyma image by using a preset lesion segmentation model to obtain a first lesion mask.

[0187] A mask determination module 623, configured to determine a lesion segmentation mask of the initial lung parenchyma image according to the first lesion mask.

[0188] The first lung parenchyma image and the initial lung parenchyma image are different in image size.

[0189] The mask determination module 623 is specifically configured to:

[0190] adjust the size of the first lesion mask to the size of the initial lung parenchyma image;

[0191] determine lesion points in the first lesion mask of the adjusted size according to the lesion probabilities of the voxels in the first lesion mask of the adjusted size;

[0192] determine a lesion segmentation mask in the initial lung parenchyma image according to the lesion points.

[0193] The mask determination module 623 is specifically configured to:

[0194] determine a lesion connected domain in the first lesion mask of the adjusted size according to the lesion points;

[0195] determine a lesion segmentation mask in the initial lung parenchyma image according to the lesion connected domain, wherein the lesion segmentation mask includes a plurality of lesion connected domains.

[0196] The detection unit 630 includes:

[0197] ​​​​The second preprocessing module 631 is configured to perform preprocessing on the initial lung parenchyma image to obtain a second lung parenchyma image.

[0198] The detection module 632 is configured to perform processing on the second lung parenchyma image by using a preset lesion detection model to obtain a current lesion region.

[0199] The region determination module 633 is configured to determine a predicted lesion region in the initial lung parenchyma image according to the current lesion region.

[0200] The second lung parenchyma image and the initial lung parenchyma image are different in image size.

[0201] The region determination module 633 is specifically configured to:

[0202] Map the current lesion region to the initial lung parenchyma image according to the image size of the second lung parenchyma image and the initial lung parenchyma image to obtain the predicted lesion region in the initial lung parenchyma image.

[0203] If the lesion detection model is a two-dimensional neural network model, the current lesion region is a detection box.

[0204] If the lesion detection model is a three-dimensional neural network model, the current lesion region is a detection box.

[0205] The fusion unit 640 includes:

[0206] The region fusion module 641 is configured to determine a suspected lesion region according to the lesion segmentation mask, and determine the possible lesion region according to the suspected lesion region and the predicted lesion region.

[0207] The positive region determination module 642 is configured to determine a lesion image in the CT image according to the possible lesion region, and perform processing on each lesion image by using a preset lesion true positive discrimination model to determine a positive lesion image.

[0208] The target region determination module 643 is configured to determine a possible lesion region corresponding to the positive lesion image as the target lesion region.

[0209] The target mask determination module 644 is configured to determine a target lesion mask corresponding to the target lesion region according to the target lesion region and the lesion segmentation mask.

[0210] The lesion segmentation mask includes a plurality of lesion connected domains.

[0211] The region fusion module 641 is specifically configured to:

[0212] generate a suspected lesion region for each of the lesion connected domains in the lesion segmentation mask, the suspected lesion region comprising a center point parameter and a size parameter.

[0213] wherein the suspected lesion region comprises a first center point parameter and a first size parameter; and the predicted lesion region comprises a second center point parameter and a second size parameter.

[0214] The region fusion module 641 is specifically configured to:

[0215] determine a region combination according to the first center point parameter of each suspected lesion region and the second center point parameter of each predicted lesion region; the region combination comprising one predicted lesion region and one suspected lesion region.

[0216] determine the center point parameter and the size parameter of the possible lesion region according to the first center point parameter and the first size parameter of the predicted lesion region comprised in the region combination, and the second center point parameter and the second size parameter of the suspected lesion region comprised in the region combination.

[0217] The target mask determination module 644 is specifically configured to:

[0218] If the target lesion region comprises the lesion segmentation mask, determine the lesion segmentation mask comprised in the target lesion region as a target lesion mask corresponding to the target lesion region.

[0219] The target mask determination module 644 is specifically configured to:

[0220] If the target lesion region does not comprise the lesion segmentation mask, generate a target lesion mask according to the target lesion region, and the generated target lesion mask is located inside the target lesion region.

[0221] The target lesion region has a center point parameter and a size parameter.

[0222] The target mask determination module 644 is specifically configured to:

[0223] generate a sphere located inside the target lesion region according to the center point parameter and the size parameter of the target lesion region, to obtain the target lesion mask.

[0224] The lung parenchyma determination unit 610 comprises:

[0225] The lung parenchyma segmentation module 611 is configured to process the CT image by using a pre-set lung parenchyma extraction model to obtain a lung parenchyma segmentation result.

[0226] The preliminary determination module 612 is configured to determine a third lung parenchyma image in the CT image according to the lung parenchyma segmentation result.

[0227] The lung lobe segmentation module 613 is configured to process the third lung parenchyma image by using a preset lung lobe segmentation model to obtain a lung lobe segmentation result.

[0228] The re-determination module 614 is configured to determine the lung parenchyma region in the CT image according to the lung lobe segmentation result, wherein the lung parenchyma region includes a region where the lung lobe is located.

[0229] The lung parenchyma segmentation result is different in size from the CT image.

[0230] The lung lobe segmentation module 613 is specifically configured to:

[0231] adjust the size of the lung parenchyma segmentation result to the size of the CT image, and determine a third lung parenchyma image in the CT image according to the lung parenchyma segmentation result after the size adjustment.

[0232] The lung lobe segmentation result is different in size from the third lung parenchyma image.

[0233] The re-determination module 614 is specifically configured to:

[0234] adjust the size of the lung lobe segmentation result to the size of the third lung parenchyma image.

[0235] According to the position information of the third lung parenchyma image in the CT image, supplement blank voxels in the periphery of the lung lobe segmentation result after the size adjustment to obtain a lung lobe segmentation result with the same size as the CT image.

[0236] According to the lung lobe segmentation result with the same size as the CT image, determine the lung parenchyma region in the CT image.

[0237] The lung lobe segmentation result includes information of multiple lung lobes.

[0238] The device further includes a lung interstitial fissure determination unit 650, configured to:

[0239] Determine a voxel point at a boundary between two adjacent lung lobes as a lung interstitial fissure result.

[0240] The lung lobe segmentation result includes information of multiple lung lobes.

[0241] The device further includes a lung lobe volume determination unit 660, configured to:

[0242] According to each lung lobe information in the lung lobe segmentation result, determine a lung lobe region corresponding to each lung lobe in the CT image.

[0243] Adjust each voxel in the lung lobe region to a unit volume, and determine the lung lobe volume according to the number of voxels, the original volume of voxels in the lung lobe region.

[0244] The present disclosure provides a CT image processing method and device, equipment, storage medium and program product, which are applied to deep learning technology in artificial intelligence technology and AI medical technology, to more accurately identify a region possibly having a lesion in a CT image.

[0245] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0246] According to embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium and a computer program product.

[0247] According to embodiments of the present disclosure, the present disclosure further provides a computer program product, which comprises a computer program stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and the at least one processor executes the computer program to make the electronic device execute the scheme provided in any of the above embodiments.

[0248] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0249] As shown in Figure 7 The device 700 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. Various programs and data required for the operation of the device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0250] A plurality of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0251] The computing unit 701 can be various general and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above, such as the processing method of computed tomography. For example, in some embodiments, the processing method of computed tomography can be implemented as a computer software program, which is tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the processing method of computed tomography described above can be performed. Alternatively, in other embodiments, the computing unit 701 can be configured to perform the processing method of computed tomography by any other appropriate means, such as by means of firmware.

[0252] Various implementations of the systems and techniques described above herein can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0253] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package, or entirely on a remote machine or server.

[0254] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0255] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0256] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0257] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions typically taking place over a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are a host product in the cloud computing service system to solve the defects of large management difficulty and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS"). The servers can also be servers of a distributed system, or servers combined with a blockchain.

[0258] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be executed in parallel, in series, or in a different order, without departing from the desired results of the technical solutions of the present disclosure, and the present disclosure is not limited herein.

[0259] The above detailed description does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A processing method of computed tomography, comprising: determining a lung parenchyma region in a computed tomography (CT) image, and cropping an initial lung parenchyma image from the CT image according to the lung parenchyma region, wherein the initial lung parenchyma image comprises the lung parenchyma region; processing the initial lung parenchyma image to obtain a lesion segmentation mask, the lesion segmentation mask being used to represent a lesion position in the initial lung parenchyma image, and the lesion segmentation mask comprising a plurality of lesion connected domains; processing the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image; determining a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent the lesion position in the initial lung parenchyma image; wherein the target lesion mask is generated by fusing the lesion segmentation mask and the predicted lesion region, if the lesion segmentation mask is included in the target lesion region, the lesion segmentation mask included in the target lesion region is determined as the target lesion mask; if the lesion segmentation mask is not included in the target lesion region, a sphere located inside the target lesion region is generated according to a center point parameter and a size parameter of the target lesion region as the target lesion mask, and the generated target lesion mask is located inside the target lesion region.

2. The method of claim 1, wherein, the processing of the initial lung parenchyma image to obtain the lesion segmentation mask comprises: preprocessing the initial lung parenchyma image to obtain a first lung parenchyma image; processing the first lung parenchyma image using a preset lesion segmentation model to obtain a first lesion mask; determining the lesion segmentation mask of the initial lung parenchyma image according to the first lesion mask.

3. The method of claim 2, wherein, the first lung parenchyma image and the initial lung parenchyma image have different image sizes; the determination of the lesion segmentation mask of the initial lung parenchyma image according to the first lesion mask comprises: adjusting the size of the first lesion mask to the size of the initial lung parenchyma image; determining lesion points in the first lesion mask with adjusted size according to lesion probabilities of each voxel point in the first lesion mask with adjusted size; determining the lesion segmentation mask in the initial lung parenchyma image according to the lesion points.

4. The method of claim 3, wherein, the determination of the lesion segmentation mask in the initial lung parenchyma image according to the lesion points comprises: determining lesion connected domains in the first lesion mask with adjusted size according to the lesion points; determining the lesion segmentation mask in the initial lung parenchyma image according to the lesion connected domains, wherein the lesion segmentation mask comprises a plurality of lesion connected domains.

5. The method according to any one of claims 1 to 4, wherein, the processing of the initial lung parenchyma image to obtain the predicted lesion region in the initial lung parenchyma image comprises: preprocessing the initial lung parenchyma image to obtain a second lung parenchyma image; processing the second lung parenchyma image using a preset lesion detection model to obtain a current lesion region; According to the current lesion region, a predicted lesion region in the initial lung parenchyma image is determined.

6. The method of claim 5, wherein, The second lung parenchyma image is different from the initial lung parenchyma image in image size; The determining of the predicted lesion region in the initial lung parenchyma image according to the current lesion region comprises: According to the image size of the second lung parenchyma image and the initial lung parenchyma image, the current lesion region is mapped into the initial lung parenchyma image to obtain the predicted lesion region in the initial lung parenchyma image.

7. The method of claim 6, wherein, if the lesion detection model is a two-dimensional neural network model, the current lesion region is a detection box; if the lesion detection model is a three-dimensional neural network model, the current lesion region is a detection box.

8. The method according to claim 6 or 7, wherein, The determining of the target lesion region and the target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region comprises: According to the lesion segmentation mask, a suspected lesion region is determined, and the possible lesion region is determined according to the suspected lesion region and the predicted lesion region; According to the possible lesion region, a lesion image in the CT image is determined, and a preset lesion true positive discrimination model is used to process each lesion image to determine a positive lesion image; The possible lesion region corresponding to the positive lesion image is determined as the target lesion region; According to the target lesion region and the lesion segmentation mask, a target lesion mask corresponding to the target lesion region is determined.

9. The method of claim 8, wherein, The lesion segmentation mask comprises a plurality of lesion connected domains; The determining of the suspected lesion region according to the lesion segmentation mask comprises: A suspected lesion region is generated for each lesion connected domain in the lesion segmentation mask, and the suspected lesion region comprises a center point parameter and a size parameter.

10. The method of claim 9, wherein, The suspected lesion region comprises a first center point parameter and a first size parameter, and the predicted lesion region comprises a second center point parameter and a second size parameter; The determining of the possible lesion region according to the suspected lesion region and the predicted lesion region comprises: According to the first center point parameter of each suspected lesion region and the second center point parameter of each predicted lesion region, a region combination is determined, and the region combination comprises one predicted lesion region and one suspected lesion region; According to the first center point parameter and the first size parameter of the predicted lesion region included in the region combination and the second center point parameter and the second size parameter of the suspected lesion region included in the region combination, a center point parameter and a size parameter of a possible lesion region are determined.

11. The method of any one of claims 1-4, wherein, The determining of the lung parenchyma region in the CT image comprises: A preset lung parenchyma extraction model is used to process the CT image to obtain a lung parenchyma segmentation result; According to the lung parenchyma segmentation result, a third lung parenchyma image in the CT image is determined; A preset lung lobe segmentation model is used to process the third lung parenchyma image to obtain a lung lobe segmentation result; According to the lung lobe segmentation result, the lung parenchyma region in the CT image is determined; wherein the lung parenchyma region comprises a region where a lung lobe is located.

12. The method of claim 11, wherein, The lung parenchyma segmentation result is different from the size of the CT image; The third lung parenchyma image is determined in the CT image according to the lung parenchyma segmentation result, and the third lung parenchyma image includes the lung parenchyma segmentation result. The size of the lung parenchyma segmentation result is adjusted to the size of the CT image, and the third lung parenchyma image is determined in the CT image according to the lung parenchyma segmentation result after the size is adjusted.

13. The method of claim 12, wherein, The size of the lung lobe segmentation result is different from the size of the third lung parenchyma image. The lung parenchyma region is determined in the CT image according to the lung lobe segmentation result, and the lung parenchyma region includes the lung lobe segmentation result. The size of the lung lobe segmentation result is adjusted to the size of the third lung parenchyma image. The lung lobe segmentation result is adjusted to the size of the third lung parenchyma image. The lung lobe segmentation result is adjusted to the size of the third lung parenchyma image.

14. The method of claim 12 or 13, wherein, The lung lobe segmentation result is adjusted to the size of the third lung parenchyma image. The lung lobe segmentation result includes information of a plurality of lung lobes. The method further includes:

15. The method of claim 12 or 13, wherein, The voxel points at the junction of the adjacent two lung lobes are determined as the interlobar fissure result. The lung lobe segmentation result includes information of a plurality of lung lobes. The method further includes: According to the lung lobe information in the lung lobe segmentation result, a lung lobe region corresponding to each lung lobe is determined in the CT image. Each voxel in the lung lobe region is adjusted to a unit volume, and the lung lobe volume is determined according to the number of voxels and the original volume of the voxels in the lung lobe region.

16. A processing device for computed tomography, comprising: a lung parenchyma determination unit configured to determine a lung parenchyma region in a computed tomography (CT) image, and to crop an initial lung parenchyma image from the CT image according to the lung parenchyma region, wherein the initial lung parenchyma image includes the lung parenchyma region; a segmentation unit configured to process the initial lung parenchyma image to obtain a lesion segmentation mask, the lesion segmentation mask being used to represent a lesion position in the initial lung parenchyma image, and the lesion segmentation mask including a plurality of lesion connected domains; a detection unit configured to process the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image; a fusion unit configured to determine a target lesion region and a target lesion mask corresponding to the target lesion region according to the lesion segmentation mask and the predicted lesion region, wherein the target lesion mask is located in the target lesion region, and the target lesion mask is used to represent a lesion position in the initial lung parenchyma image; 17. The apparatus of claim 16, wherein, wherein the target lesion mask is generated by fusing the lesion segmentation mask and the predicted lesion region, if the target lesion region includes the lesion segmentation mask, the lesion segmentation mask included in the target lesion region is determined as the target lesion mask, if the target lesion region does not include the lesion segmentation mask, a sphere located inside the target lesion region is generated as the target lesion mask according to a center point parameter and a size parameter of the target lesion region, and the generated target lesion mask is located inside the target lesion region. The segmentation unit includes: The first preprocessing module is configured to preprocess the initial lung parenchyma image to obtain a first lung parenchyma image. The segmentation module is configured to process the first lung parenchyma image by using a preset lesion segmentation model to obtain a first lesion mask. The mask determination module is configured to determine a lesion segmentation mask of the initial lung parenchyma image according to the first lesion mask.

18. The apparatus of claim 17, wherein, The first lung parenchyma image has a different image size from the initial lung parenchyma image. The mask determination module is specifically configured to: adjust the size of the first lesion mask to the size of the initial lung parenchyma image; determine lesion points in the first lesion mask after size adjustment according to lesion probabilities of each voxel point in the first lesion mask after size adjustment; and determine a lesion segmentation mask in the initial lung parenchyma image according to the lesion points.

19. The apparatus of claim 18, wherein, The mask determination module is specifically configured to: determine a lesion connected domain in the first lesion mask after size adjustment according to the lesion points; and determine a lesion segmentation mask in the initial lung parenchyma image according to the lesion connected domain, wherein the lesion segmentation mask includes a plurality of lesion connected domains.

20. The apparatus of any of claims 16-19, wherein, The detection unit includes: The second preprocessing module is configured to preprocess the initial lung parenchyma image to obtain a second lung parenchyma image. The detection module is configured to process the second lung parenchyma image by using a preset lesion detection model to obtain a current lesion region. The region determination module is configured to determine a predicted lesion region in the initial lung parenchyma image according to the current lesion region.

21. The apparatus of claim 20, wherein, The second lung parenchyma image has a different image size from the initial lung parenchyma image. The region determination module is specifically configured to: map the current lesion region to the initial lung parenchyma image according to the image sizes of the second lung parenchyma image and the initial lung parenchyma image to obtain a predicted lesion region in the initial lung parenchyma image.

22. The apparatus of claim 21, wherein, if the lesion detection model is a two-dimensional neural network model, the current lesion region is a detection box; if the lesion detection model is a three-dimensional neural network model, the current lesion region is a detection box.

23. The apparatus of claim 21 or 22, wherein, The fusion unit includes: The region fusion module is configured to determine a suspected lesion region according to the lesion segmentation mask, and determine the possible lesion region according to the suspected lesion region and the predicted lesion region. The positive region determination module is configured to determine a lesion image in the CT image according to the possible lesion region, and process each lesion image by using a preset lesion true positive discrimination model to determine a positive lesion image. The target region determination module is configured to determine a possible lesion region corresponding to the positive lesion image as the target lesion region. The target mask determination module is configured to determine a target lesion mask corresponding to the target lesion region according to the target lesion region and the lesion segmentation mask.

24. The apparatus of claim 23, wherein, The lesion segmentation mask includes a plurality of lesion connected domains. The region fusion module is specifically configured to: generate a suspected lesion region for each of the lesion connected domains in the lesion segmentation mask, and the suspected lesion region includes a center point parameter and a size parameter.

25. The apparatus of claim 24, wherein, The suspected lesion region comprises a first center point parameter and a first size parameter; the predicted lesion region comprises a second center point parameter and a second size parameter; The region fusion module is specifically configured to: determine a region combination according to the first center point parameter of each suspected lesion region and the second center point parameter of each predicted lesion region; the region combination comprises one predicted lesion region and one suspected lesion region; determine the center point parameter and the size parameter of the possible lesion region according to the first center point parameter and the first size parameter of the predicted lesion region included in the region combination and the second center point parameter and the second size parameter of the suspected lesion region included in the region combination.

26. The apparatus of any one of claims 16-19, wherein, The lung parenchyma determination unit comprises: a lung parenchyma segmentation module configured to process the CT image by using a preset lung parenchyma extraction model to obtain a lung parenchyma segmentation result; a preliminary determination module configured to determine a third lung parenchyma image in the CT image according to the lung parenchyma segmentation result; a lung lobe segmentation module configured to process the third lung parenchyma image by using a preset lung lobe segmentation model to obtain a lung lobe segmentation result; a re-determination module configured to determine the lung parenchyma region in the CT image according to the lung lobe segmentation result; wherein the lung parenchyma region comprises a region where a lung lobe is located.

27. The apparatus of claim 26, wherein, The lung parenchyma segmentation result has a size different from that of the CT image; The lung lobe segmentation module is specifically configured to: adjust the size of the lung parenchyma segmentation result to the size of the CT image, and determine a third lung parenchyma image in the CT image according to the lung parenchyma segmentation result with the adjusted size.

28. The apparatus of claim 27, wherein, The lung lobe segmentation result has a size different from that of the third lung parenchyma image; The re-determination module is specifically configured to: adjust the size of the lung lobe segmentation result to the size of the third lung parenchyma image; supplement blank voxels in the periphery of the lung lobe segmentation result with the adjusted size according to the position information of the third lung parenchyma image in the CT image, to obtain a lung lobe segmentation result with the same size as that of the CT image; determine the lung parenchyma region in the CT image according to the lung lobe segmentation result with the same size as that of the CT image.

29. The apparatus of claim 27 or 28, wherein, The lung lobe segmentation result comprises information of a plurality of lung lobes; The device further comprises a interlobar fissure determination unit configured to: determine a voxel point at the junction of two adjacent lung lobes as an interlobar fissure result.

30. The apparatus of claim 27 or 28, wherein, The lung lobe segmentation result comprises information of a plurality of lung lobes; The device further comprises a lung lobe volume determination unit configured to: determine a lung lobe region corresponding to each lung lobe in the CT image according to each lung lobe information in the lung lobe segmentation result; adjust each voxel in the lung lobe region to a unit volume, and determine a lung lobe volume according to the number of voxels and the original volume of the voxels in the lung lobe region. 31.An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication;wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-15.

32. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method according to any one of claims 1-15.

33. A computer program product comprising computer instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1-15.

Images