A method for detecting prostate cancer lesions based on PET / CT images

Abstract

The present invention discloses a prostate cancer lesion detection method based on PET / CT images, and relates to the technical field of medical image processing. By acquiring the PSMA PET / CT images of the user, the soft tissue window CT images and the lung window CT images are extracted according to the CT images; the CT images are segmented to obtain organ segmentation images; the PSMA PET images, the soft tissue window CT images, the lung window CT images, and the organ segmentation images are respectively input into the PET lesion detection model and the CT lesion detection model to obtain the PET lesion detection images and the CT lesion detection images; and the PSMA PET / CT lesion detection information is generated according to the two lesion detection images. The present invention combines the PSMA PET / CT images with the CT anatomical information, and can more accurately detect and segment lesions, provide more accurate reference data for clinicians, and provide better data support for the formulation of individualized treatment plans.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a prostate cancer lesion detection method based on PET / CT images. Background Art

[0002] Prostate cancer (PCa) is the most common malignant tumor in men worldwide, and its mortality rate ranks second among all male malignant tumors. The incidence and mortality of prostate cancer in China are currently on the rise. Patients with moderate or high-risk prostate cancer need to undergo staging examinations to determine whether the tumor has spread beyond the prostate. Since the success of treatment depends largely on whether there is systemic spread, imaging examinations are of great significance in the diagnosis, staging and efficacy evaluation of prostate cancer.

[0003] Prostate specific membrane antigen (PSMA) is a type II transmembrane glycoprotein whose expression is significantly increased in prostate cancer cells. PSMA PET / CT has been successfully used for imaging of prostate cancer patients. Studies have found that PSMA has important value in the clinical application of prostate cancer and has obvious advantages in detecting prostate cancer metastases.

[0004] In today's clinical work, PCa lesions are usually identified by doctors through visual inspection, which has a large personal subjective error and causes excessive workload for doctors. In actual work, manually outlining all existing lesions is quite time-consuming and operator-dependent, so automated lesion detection and segmentation will be a good aid in the diagnosis, evaluation and treatment planning of prostate cancer.

[0005] In recent years, deep learning methods in the field of artificial intelligence have made great progress in the field of medical image processing, including target detection, image classification and segmentation. Among them, the research on deep learning methods based on PET / CT images is also a booming emerging field.

[0006] However, there are currently very few studies on prostate cancer lesion detection and segmentation based on PSMA PET / CT images, and the existing prostate cancer tumor recognition model only uses PET and CT image modalities for lesion detection, failing to make full use of more image information, resulting in low accuracy in prostate cancer lesion detection and segmentation.

[0007] Therefore, the existing technology still needs to be improved and developed. Summary of the invention

[0008] The technical problem to be solved by the present invention is that, in view of the above-mentioned defects of the prior art, a prostate cancer lesion detection method based on PET / CT images is provided, aiming to solve the problem that the deep learning method based on PET / CT images in the prior art only uses two image modalities, PET and CT, for lesion detection, fails to make full use of more image information, and leads to low accuracy of lesion detection and segmentation.

[0009] The technical solution adopted by the present invention to solve the problem is as follows:

[0010] In a first aspect, an embodiment of the present invention provides a method for detecting prostate cancer lesions based on PET / CT images, the method comprising:

[0011] Obtain the user's PSMA PET / CT image, and extract the soft tissue window CT image and lung window CT image based on the CT image;

[0012] Performing image segmentation on the CT image to obtain an organ segmentation image;

[0013] Inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image into the trained PET lesion detection model to obtain a PET lesion detection image;

[0014] Inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image;

[0015] The PSMA PET / CT lesion detection information corresponding to the user is generated according to the PET lesion detection image and the CT lesion detection image.

[0016] In one embodiment, after obtaining the PSMA PET / CT image of the user, the method further includes:

[0017] Performing intelligent registration on the PSMA PET / CT images to establish a correct correspondence between structure and function;

[0018] The coregistered PSMA PET images were normalized to the SUV images by injected dose and body weight.

[0019] In one embodiment, performing image segmentation on the CT image to obtain an organ segmentation image includes:

[0020] The CT image is input into a trained whole-body organ segmentation network to obtain the organ segmentation image.

[0021] In one embodiment, the PET lesion detection model comprises:

[0022] A first-level PET segmentation network, used for outputting a first-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image;

[0023] A second-level PET segmentation network is used to output a second-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level PET lesion segmentation image;

[0024] The PET lesion classification network is used to determine the lesion probability of each lesion in the second-level PET lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level PET lesion segmentation image, and output a final PET lesion detection image according to the lesion probabilities of all lesions.

[0025] In one embodiment, the CT lesion detection model includes:

[0026] A first-level CT segmentation network, used for outputting a first-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image;

[0027] A second-level CT segmentation network is used to output a second-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level CT lesion segmentation image;

[0028] The CT lesion classification network is used to determine the lesion probability of each lesion in the second-level CT lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level CT lesion segmentation image, and output the final CT lesion detection image according to the lesion probabilities of all lesions.

[0029] In one implementation, generating PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image includes:

[0030] Determine a plurality of lesions and attribute categories corresponding to the lesions according to the PET lesion detection image and the CT lesion detection image, wherein the attribute category of each lesion is one of double-detected lesions, PET-only detected lesions, and CT-only detected lesions;

[0031] Acquiring structural position information corresponding to each lesion according to the organ segmentation image;

[0032] Obtaining index information corresponding to each lesion according to the PSMA PET image, wherein the index information includes at least one index data of maximum standard uptake value, average standard uptake value, total glycolysis of lesion, and metabolic volume of tumor;

[0033] PSMA PET / CT lesion detection information is generated according to the attribute category, the structural location information and the indicator information of each lesion.

[0034] In one implementation, the organ segmentation image, the PET lesion detection image, and the CT lesion detection image are all mask images.

[0035] In a second aspect, an embodiment of the present invention further provides a prostate cancer lesion detection system based on PET / CT imaging, the system comprising:

[0036] An image processing module is used to obtain the user's PSMA PET / CT image and extract the soft tissue window CT image and the lung window CT image according to the CT image;

[0037] An organ extraction module, used for performing image segmentation on the CT image to obtain an organ segmentation image;

[0038] A PET lesion detection module, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained PET lesion detection model to obtain a PET lesion detection image;

[0039] A CT lesion detection module, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image;

[0040] An information fusion module is used to generate PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image.

[0041] In a third aspect, an embodiment of the present invention also provides a prostate cancer lesion detection terminal based on PET / CT imaging, the terminal comprising a memory and one or more processors; the memory stores one or more programs; the program comprises instructions for executing any of the prostate cancer lesion detection methods based on PET / CT imaging as described above; the processor is used to execute the program.

[0042] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a plurality of instructions stored thereon, wherein the instructions are suitable for being loaded and executed by a processor to implement any of the steps of the above-mentioned method for detecting prostate cancer lesions based on PET / CT images.

[0043] Beneficial effects of the present invention: The embodiment of the present invention obtains the PSMA PET / CT image of the user, extracts the soft tissue window CT image and the lung window CT image according to the CT image; performs image segmentation according to the CT image to obtain an organ segmentation image; inputs the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into the trained PET lesion detection model to obtain a PET lesion detection image; inputs the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into the trained CT lesion detection model to obtain a CT lesion detection image; generates lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image. The present invention not only utilizes the two image modalities of PET and CT in the process of lesion detection, but also further utilizes the anatomical information of the CT image as auxiliary information, so that more accurate lesion detection and segmentation can be achieved, providing more accurate reference data for clinicians, and providing better data support for the formulation of individualized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 It is a flow chart of a method for detecting prostate cancer lesions based on PET / CT images provided in an embodiment of the present invention.

[0046] Figure 2 It is a schematic diagram of the principle of a prostate cancer lesion detection method based on PET / CT imaging provided in an embodiment of the present invention.

[0047] Figure 3 It is a schematic diagram of whole body anatomical structure segmentation provided by an embodiment of the present invention.

[0048] Figure 4 It is a schematic diagram of the PET lesion detection process provided by an embodiment of the present invention.

[0049] Figure 5 It is a schematic diagram of a network for classifying true and false positive lesions provided in an embodiment of the present invention.

[0050] Figure 6 Schematic diagram of a PSMA PET / CT image and a lesion mask provided by an embodiment of the present invention.

[0051] Figure 7 It is a module schematic diagram of a prostate cancer lesion detection system based on PET / CT imaging provided by an embodiment of the present invention.

[0052] Figure 8 It is a principle block diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention discloses a method for detecting prostate cancer lesions based on PET / CT images. In order to make the purpose, technical scheme and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0055] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0056] In view of the above-mentioned defects of the prior art, the present invention provides a prostate cancer lesion detection method based on PET / CT images. The method obtains the PSMA PET / CT image of the user, extracts the soft tissue window CT image and the lung window CT image according to the CT image; performs image segmentation on the CT image to obtain an organ segmentation image; inputs the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained PET lesion detection model to obtain a PET lesion detection image; inputs the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image; and generates PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image. The present invention not only utilizes the two image modalities of PET and CT in the process of lesion detection, but also further utilizes the anatomical information of CT images as auxiliary information, thereby achieving more accurate lesion detection and segmentation, providing clinicians with more accurate reference data, and providing better data support for the formulation of individualized treatment plans.

[0057] like Figure 1 As shown, the method includes:

[0058] Step S100: Acquire the PSMA PET / CT image of the user, and extract the soft tissue window CT image and the lung window CT image according to the CT image.

[0059] Specifically, PET / CT is a combination of PET and CT, which are designed as one and controlled by a workstation. During scanning, PET imaging and CT imaging are performed simultaneously according to demand, that is, the PSMA PET / CT image in this embodiment represents the PSMA PET image and the CT image. PSMA PET images can reflect the information and changes of tissue metabolism, that is, the functional information of the organs in the user's body; CT images can reflect the structural information of the organs in the user's body. In order to facilitate subsequent image analysis, it is necessary to extract local images of the numerical range of soft tissue and lungs from the CT images through the preset window width and window position, that is, to obtain soft tissue window CT images and lung window CT images.

[0060] In one implementation, after obtaining the PSMA PET / CT image of the user, the method further includes:

[0061] Performing intelligent registration on the PSMA PET / CT images to establish a correct correspondence between structure and function;

[0062] The coregistered PSMA PET images were normalized to the SUV images by injected dose and body weight.

[0063] Specifically, due to the difference in image space size between the originally acquired PSMA PET images and CT images, it may be difficult to accurately correspond the organ function reflected by the PSMA PET images and the organ structure reflected by the CT images. Figure 2 As shown, this embodiment first needs to perform intelligent registration on the two medical images to make the image space sizes of the two images consistent, thereby ensuring the correct correspondence between structure and function. Then, the PSMA PET image is standardized to the SUV image by injection dose and body weight.

[0064] For example, for the obtained PET / CT images, the CT images are first intelligently registered to the PSMA PET images using the trained registration network to keep the PSMA PET images and CT images in the same spatial size. Then, through intensity threshold processing, the lung window-CT images and soft tissue window-CT images are extracted from the CT images using the window width and window position of [1200,-600] and [300,40]. The PSMA PET images are then standardized to SUV images using the standardization formula: SUV = PET pixel value / (injected dose (MBq) / body weight (kg)).

[0065] Step S200: performing image segmentation on the CT image to obtain an organ segmentation image.

[0066] Specifically, since the metabolism of each organ in the PSMA PET image is different, such as the kidney, heart, and bladder are often accompanied by physiological high uptake, the differences between different organs need to be considered in the process of detecting abnormal high uptake lesions in PET / CT. In this embodiment, the CT image is segmented according to human organs to obtain structural semantic information represented by human organs.

[0067] In one implementation, performing image segmentation on the CT image to obtain an organ segmentation image includes:

[0068] The CT image is input into a trained whole-body organ segmentation network to obtain the organ segmentation image.

[0069] Specifically, in order to accurately and quickly complete the organ segmentation of CT images, this embodiment pre-builds a whole-body organ segmentation network based on a deep learning network. The input data of the network is the user's CT image, and the output data is the organ segmentation image obtained after performing organ segmentation on the CT image. By performing organ segmentation on the CT image, this embodiment can extract identifiable organs within the user's whole body as structural semantic information, thereby obtaining anatomical information in the CT image.

[0070] For example, Figure 3As shown, the trained whole-body organ segmentation network model is used to perform anatomical structure segmentation on the CT image to obtain a whole-body anatomical region mask image / organ mask image (Mask). The mask image is the organ segmentation image, which can guide the subsequent detection of abnormal high uptake lesions in PET.

[0071] Step S300: input the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into the trained PET lesion detection model to obtain a PET lesion detection image.

[0072] Specifically, this embodiment will establish a PET lesion detection model based on PSMA PET images, CT images (i.e., soft tissue window CT images and lung window CT images) and organ segmentation images to obtain lesion recognition results on PSMA PET images. In order to improve the model prediction ability, this embodiment will also use a pre-constructed training data set to converge the gap between the prediction results of the PET lesion detection model and the true label. In actual application scenarios, the current user's PSMA PET images, soft tissue window CT images, lung window CT images and organ segmentation images are input into the trained PET lesion detection model, and the PET lesion detection model will output the corresponding PET lesion detection image, which can reflect the pathological high uptake lesions identified from the PSMA PET images.

[0073] In one implementation, the PET lesion detection model is a heterogeneous cascade, including:

[0074] A first-level PET segmentation network, used for outputting a first-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image;

[0075] A second-level PET segmentation network is used to output a second-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level PET lesion segmentation image;

[0076] The PET lesion classification network is used to determine the lesion probability of each lesion in the second-level PET lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level PET lesion segmentation image, and output a final PET lesion detection image according to the lesion probabilities of all lesions.

[0077] Specifically, high uptake of PSMA PET images may be physiological high uptake, such as bilateral kidneys and bladder, or pathological high uptake, such as malignant tumors. The differential diagnosis of these two types of high uptake requires the combination of anatomical knowledge and the experience of imaging physicians. In order to reduce the workload of physicians, Figure 4 As shown, this embodiment pre-builds a PET lesion detection model using a heterogeneous cascade method for identifying pathological high uptake from PET high uptake. Heterogeneous cascade can improve lesion detection performance and effectively suppress false positive detection of lesions without reducing the lesion detection rate. The heterogeneous cascade of the PET lesion detection model mainly includes three parts: a first-level PET segmentation network, a second-level PET segmentation network, and a PET lesion classification network. Among them, the input data of the first-level PET segmentation network are PSMA PET images, soft tissue window CT images, lung window CT images and organ segmentation images, and the output data are the first-level PET lesion segmentation images, which can preliminarily extract the pathological high uptake areas in the PSMA PET images, but contain many false positive detections; the input data of the second-level PET segmentation network are PSMA PET images, soft tissue window CT images, lung window CT images, organ segmentation images and the first-level PET lesion segmentation images, and the output data are the second-level PET lesion segmentation images, which can optimize the deficiencies of the first-level PET segmentation network, improve the segmentation accuracy, and remove some false positive detections; the input data of the PET lesion classification network are the connected domain segmentation images and local PET / CT images corresponding to each lesion in the second-level PET lesion segmentation images, and the output data are the true positive probability values ​​of the corresponding lesions, and the final PET lesion detection image is output according to the true positive probability values ​​of all lesions. By analyzing the local PET / CT images of each lesion, the false positive lesions are further removed, and the detection effect of the overall pathological high uptake lesions is improved. The schematic diagram of the PET lesion classification network is shown in the figure. Figure 5 shown.

[0078] In one implementation, the first-level PET lesion segmentation image, the second-level PET lesion segmentation image, and the PET lesion detection image are all lesion prediction mask images.

[0079] Step S400: input the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image.

[0080] Specifically, there are not only pathological high uptake lesions that can be identified from PSMA PET images in PET / CT images, but also structural change lesions that can be identified from CT images, such as lung nodules. Therefore, this embodiment will also establish a CT lesion detection model based on PSMA PET images, CT images (i.e., soft tissue window CT images and lung window CT images) and organ segmentation images to obtain lesion recognition results on CT images. In order to improve the model prediction ability, this embodiment will also use a pre-constructed training data set to converge the gap between the prediction results of the CT lesion detection model and the true label. In actual application scenarios, the PSMA PET images, soft tissue window CT images, lung window CT images and organ segmentation images of the current user are input into the trained CT lesion detection model, and the CT lesion detection model will output the corresponding CT lesion detection image, which can reflect the structural change lesions identified from the CT image.

[0081] In one implementation, the CT lesion detection model includes:

[0082] A first-level CT segmentation network, configured to output a first CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image;

[0083] A second-level CT segmentation network is used to output a second-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level CT lesion segmentation image;

[0084] The CT lesion classification network is used to determine the lesion probability of each lesion in the second-level CT lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level CT lesion segmentation image, and output the final CT lesion detection image according to the lesion probabilities of all lesions.

[0085] Specifically, similar to the PET lesion detection model, the CT lesion detection model is also constructed using a heterogeneous cascade method to identify structural change lesions in CT images. The heterogeneous cascade of the CT lesion detection model mainly includes three parts: the first-level CT segmentation network, the second-level CT segmentation network, and the CT lesion classification network. Among them, the input data of the first-level CT segmentation network are PSMA PET images, soft tissue window CT images, lung window CT images and organ segmentation images, and the output data are the first-level CT lesion segmentation images, which can preliminarily extract the areas of structural changes in the CT images, but contain many false positive detections; the input data of the second-level CT segmentation network are PSMA PET images, soft tissue window CT images, lung window CT images, organ segmentation images and first-level CT lesion segmentation images, and the output data are the second-level CT lesion segmentation images, which can optimize the deficiencies of the first-level CT segmentation network, improve the segmentation accuracy, and remove some false positive detections; the input data of the CT lesion classification network are the connected domain segmentation images and local PET / CT images corresponding to each lesion in the second-level CT lesion segmentation images, and the output data are the true positive probability values ​​of the corresponding lesions, and the final CT lesion detection image is output according to the true positive probability values ​​of all lesions. By analyzing the local PET / CT images of each lesion, the false positive lesions are further removed, and the detection effect of the overall structural change lesions is improved.

[0086] In one implementation, the first-level CT lesion segmentation image, the second-level CT lesion segmentation image, and the CT lesion detection image are all lesion prediction mask images.

[0087] In one implementation, a method for generating a training data set of the PET lesion detection model / the CT lesion detection model includes:

[0088] Collect PSMA PET / CT imaging data and clinical information of user groups with specific disease types through medical institutions, and obtain PSMA PET / CT imaging data of several users;

[0089] In the collected PSMA PET / CT image data, PET pathological high uptake lesions and CT structural change lesions are respectively annotated to obtain the corresponding pathological high uptake lesion masks and structural change lesion masks;

[0090] All labeled image data are divided into training data set, verification data set and test data set according to preset ratios;

[0091] For the PET lesion detection model, the convolutional neural network model was trained using the training set data and the corresponding annotated pathological high uptake lesion masks;

[0092] For the CT lesion detection model, the convolutional neural network model is trained using the training set data and the corresponding annotated structural change lesion masks;

[0093] Use the validation set data to fine-tune the trained PET lesion detection model / trained CT lesion detection model;

[0094] The test set data is input into the tuned PET lesion detection model / tuned CT lesion detection model for result prediction, and the pathological high uptake lesion detection results corresponding to the PET lesion detection model / the structural change lesion detection results corresponding to the CT lesion detection model are obtained. At the same time, the algorithm performance indicators are statistically analyzed to evaluate the accuracy and effect of the algorithm.

[0095] For example, PSMA PET / CT imaging data and clinical data of prostate cancer patients were retrospectively collected from the hospital, and PSMA PET / CT images of 500 patients were collected. The inclusion criteria are as follows:

[0096] 1) Aged 18 and above, regardless of gender;

[0097] 2) Whole-body PET / CT imaging with image slice thickness ≤5 mm;

[0098] 3) The patients were compliant and signed informed consent forms. Except for some patients with special needs, all other patients were conscious.

[0099] Professional annotators annotate the PET pathological high uptake lesions and CT structural change lesions in the collected PET / CT data, i.e., delineate the gold standard. The annotation results are reviewed by two doctors, and when there is a disagreement, the higher-level doctor will make the final decision on the modification. Figure 6 As shown, the left side is the CT image, the middle is the PET image, and the right side is the lesion mask. The 500 data sets were divided into training set, validation set, and test set in a ratio of 8:1:1. For the PET lesion detection model, the training set data (80%) and the corresponding annotated pathological high uptake lesion mask were used to train the convolutional neural network model; for the CT lesion detection model, the training set data (80%) and the corresponding annotated structural change lesion mask were used to train the convolutional neural network model. Then the validation set data (10%) was used to tune the trained model, and the test set data (10%) was input into the final algorithm model for result prediction, and the pathological high uptake lesion detection results / structural change lesion detection results were obtained. At the same time, the algorithm performance indicators were statistically analyzed to evaluate the accuracy and effect of the algorithm.

[0100] It should be noted that for lesion segmentation, if different deep learning segmentation networks and different framework strategies are selected, the segmentation and detection effects of the final model will be different, which will in turn affect the final lesion recognition ability and analysis effect; for lesion classification, if different deep learning classification networks are selected, the classification effect of the final model will be different, which will in turn lead to differences in the final segmentation and detection effects.

[0101] In one implementation, the training method of the PET lesion detection model is as follows: the first-level PET segmentation network and the second-level PET segmentation network are trained separately. In view of the lack of output data of the first-level PET segmentation network, the second-level PET segmentation network needs to be supplemented with a pathological high uptake mask image generated based on the PSMA PET image and a preset threshold as input data during training.

[0102] In one implementation, the training method of the CT lesion detection model is: first, train the first-level CT segmentation network, and when the first-level CT segmentation network is trained to a certain extent (not fully trained), train the second-level CT segmentation network through the output results of the first-level CT segmentation network.

[0103] Step S500: Generate PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image.

[0104] Specifically, this embodiment will finally fuse the lesions identified from the PSMA PET images with the lesions identified from the CT images to obtain the final PSMA PET / CT lesion detection results.

[0105] In one implementation, generating PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image includes:

[0106] Determine a plurality of lesions and attribute categories corresponding to the lesions according to the PET lesion detection image and the CT lesion detection image, wherein the attribute category of each lesion is one of double-detected lesions, PET-only detected lesions, and CT-only detected lesions;

[0107] Acquiring structural position information corresponding to each lesion according to the organ segmentation image;

[0108] Obtaining index information corresponding to each lesion according to the PSMA PET image, wherein the index information includes at least one index data of maximum standard uptake value, average standard uptake value, total glycolysis of lesion, and metabolic volume of tumor;

[0109] PSMA PET / CT lesion detection information is generated according to the attribute category, the structural location information and the indicator information of each lesion.

[0110] Specifically, this embodiment will combine and analyze the PET lesion detection image and the CT lesion detection image to determine the attribute category of each lesion, for example, combining the pathological high uptake lesion mask from the PSMA PET image and the structural change lesion mask from the CT image to obtain the final PET / CT lesion mask, and distinguishing between double-detected lesions (i.e., lesions detected by PET / CT at the same time), only PET-detected lesions, and only CT-detected lesions according to the judgment method of whether the connected domains intersect (i.e., whether the covered areas overlap). Combined with the organ segmentation image extracted based on the CT image, the structural position information of each lesion can be analyzed. Combined with the PSMA PET image, the maximum standard uptake value (SUVmax), the average standard uptake value (SUVmean), the total glycolysis of the lesion (TLG), the disease tumor metabolic volume (MTV) and other indicator information can be obtained. By comprehensively obtaining the above information, doctors can be provided with more comprehensive and rich information about the lesion, assisting doctors in obtaining more accurate image analysis results, and providing better data support for the formulation of individualized treatment plans.

[0111] In one implementation, the registration network, the whole body organ segmentation network, the PET lesion detection model, and the CT lesion detection model can be integrated into a large model as a PET / CT lesion automatic segmentation model.

[0112] The advantages of the present invention are that it can establish an automatic segmentation model of PSMA PET / CT tumor lesions through deep learning methods, and use CT anatomical information as an aid to fully utilize the information of PET and CT images to achieve more accurate prostate cancer lesion detection and segmentation, provide more accurate data analysis results for clinicians, and provide better data support for the formulation of individualized treatment plans.

[0113] Based on the above embodiments, the present invention also provides a prostate cancer lesion detection system based on PET / CT imaging, such as Figure 7 As shown, the system comprises:

[0114] Image processing module 01, used to obtain the user's PSMA PET / CT image, and extract the soft tissue window CT image and the lung window CT image according to the CT image;

[0115] The organ extraction module 02 is used to perform image segmentation on the CT image to obtain an organ segmentation image;

[0116] PET lesion detection module 03, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into the trained PET lesion detection model to obtain a PET lesion detection image;

[0117] CT lesion detection module 04, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into the trained CT lesion detection model to obtain a CT lesion detection image;

[0118] The information fusion module 05 is used to generate PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image.

[0119] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be shown as follows: Figure 8 As shown. The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. Among them, the processor of the terminal is used to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a prostate cancer lesion detection method based on PET / CT images is implemented. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0120] Those skilled in the art will understand that Figure 8 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0121] In one implementation, the terminal stores one or more programs in its memory and is configured to be executed by one or more processors. The one or more programs include instructions for performing a prostate cancer lesion detection method based on PET / CT images.

[0122] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0123] In summary, the present invention discloses a prostate cancer lesion detection method based on PET / CT images, which relates to the technical field of medical image processing. The method obtains the PSMA PET / CT images of the user, extracts the soft tissue window CT images and the lung window CT images according to the CT images; performs image segmentation on the CT images to obtain organ segmentation images; inputs the PSMA PET images, soft tissue window CT images, lung window CT images and organ segmentation images into the PET lesion detection model and the CT lesion detection model respectively to obtain PET lesion detection images and CT lesion detection images; generates PSMA PET / CT lesion detection information according to the two lesion detection images. The present invention combines PSMA PET / CT images with CT anatomical information, which can more accurately detect and segment lesions, provide more accurate reference data for clinicians, and provide better data support for the formulation of individualized treatment plans.

[0124] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for detecting prostate cancer lesions based on PET / CT imaging, characterized in that: The method comprises: Obtain the user's PSMA PET / CT image, and extract the soft tissue window CT image and lung window CT image based on the CT image; Performing image segmentation on the CT image to obtain an organ segmentation image; Inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image into the trained PET lesion detection model to obtain a PET lesion detection image; Inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image, and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image; Generating PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image; The PET lesion detection model includes: A first-level PET segmentation network, used for outputting a first-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image; A second-level PET segmentation network is used to output a second-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level PET lesion segmentation image; a PET lesion classification network, for determining, for each lesion in the second-level PET lesion segmentation image, a lesion probability of the lesion according to a local PET / CT image of the lesion and a connected domain segmentation image corresponding to the lesion in the second-level PET lesion segmentation image, and outputting a final PET lesion detection image according to the lesion probabilities of all lesions; The CT lesion detection model includes: A first-level CT segmentation network, used for outputting a first-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image; A second-level CT segmentation network is used to output a second-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level CT lesion segmentation image; The CT lesion classification network is used to determine the lesion probability of each lesion in the second-level CT lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level CT lesion segmentation image, and output the final CT lesion detection image according to the lesion probabilities of all lesions.

2. The method for detecting prostate cancer lesions based on PET / CT imaging according to claim 1, characterized in that: After obtaining the PSMA PET / CT image of the user, the method further includes: Performing intelligent registration on the PSMA PET / CT images to establish a correct correspondence between structure and function; The coregistered PSMA PET images were normalized to the SUV images by injected dose and body weight.

3. The method for detecting prostate cancer lesions based on PET / CT imaging according to claim 1, characterized in that: The step of performing image segmentation on the CT image to obtain an organ segmentation image includes: The CT image is input into a trained whole-body organ segmentation network to obtain the organ segmentation image.

4. The method for detecting prostate cancer lesions based on PET / CT imaging according to claim 1, characterized in that: The generating PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image includes: Determine a plurality of lesions and attribute categories corresponding to each lesion according to the PET lesion detection image and the CT lesion detection image, wherein the attribute category of each lesion is one of double-detected lesions, PET-only detected lesions, and CT-only detected lesions; Acquiring structural position information corresponding to each lesion according to the organ segmentation image; Obtaining index information corresponding to each lesion according to the PSMA PET image, wherein the index information includes at least one index data of maximum standard uptake value, average standard uptake value, total glycolysis of lesion, and metabolic volume of tumor; PSMA PET / CT lesion detection information is generated according to the attribute category, the structural location information and the indicator information of each lesion.

5. The method for detecting prostate cancer lesions based on PET / CT imaging according to claim 1, characterized in that: The organ segmentation image, the PET lesion detection image, and the CT lesion detection image are all mask images.

6. A prostate cancer lesion detection system based on PET / CT imaging, characterized in that: The system comprises: An image processing module is used to obtain the user's PSMA PET / CT image and extract the soft tissue window CT image and the lung window CT image according to the CT image; An organ extraction module, used for performing image segmentation on the CT image to obtain an organ segmentation image; A PET lesion detection module, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained PET lesion detection model to obtain a PET lesion detection image; A CT lesion detection module, used for inputting the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image into a trained CT lesion detection model to obtain a CT lesion detection image; An information fusion module, used to generate PSMA PET / CT lesion detection information corresponding to the user according to the PET lesion detection image and the CT lesion detection image; The PET lesion detection model includes: A first-level PET segmentation network, used for outputting a first-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image; A second-level PET segmentation network is used to output a second-level PET lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level PET lesion segmentation image; a PET lesion classification network, for determining, for each lesion in the second-level PET lesion segmentation image, a lesion probability of the lesion according to a local PET / CT image of the lesion and a connected domain segmentation image corresponding to the lesion in the second-level PET lesion segmentation image, and outputting a final PET lesion detection image according to the lesion probabilities of all lesions; The CT lesion detection model includes: A first-level CT segmentation network, used for outputting a first-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image and the organ segmentation image; A second-level CT segmentation network is used to output a second-level CT lesion segmentation image according to the PSMA PET image, the soft tissue window CT image, the lung window CT image, the organ segmentation image and the first-level CT lesion segmentation image; The CT lesion classification network is used to determine the lesion probability of each lesion in the second-level CT lesion segmentation image according to the local PET / CT image of the lesion and the connected domain segmentation image corresponding to the lesion in the second-level CT lesion segmentation image, and output the final CT lesion detection image according to the lesion probabilities of all lesions.

7. A prostate cancer lesion detection terminal based on PET / CT imaging, characterized in that: The terminal includes a memory and one or more processors; the memory stores one or more programs; the program contains instructions for executing the prostate cancer lesion detection method based on PET / CT images as described in any one of claims 1-5; and the processor is used to execute the program.

8. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by a processor to implement the steps of the prostate cancer lesion detection method based on PET / CT images as described in any one of claims 1 to 5 above.

Images