Probability calculation method and device of raider-dtc, electronic equipment and storage medium

Abstract

The present disclosure provides a probability calculation method and device of RAIR-DTC, electronic equipment and storage medium; wherein the method comprises: registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images and anatomical structure images of a patient with differentiated thyroid cancer to obtain a set of registered images; tumor lesion feature extraction is performed on the set of registered images to obtain feature data of at least one tumor lesion on the patient; according to the feature data, the tumor lesions on the patient are classified according to metabolic heterogeneity to obtain lesion type information corresponding to each tumor lesion; and based on the lesion type information, the probability of the patient suffering from RAIR-DTC is predicted. The present disclosure can improve the probability prediction accuracy of DTC developing into RAIR-DTC.

Description

Technical Field

[0001] This disclosure relates to the field of computer processing technology, and in particular to a probability calculation method, apparatus, electronic device, and storage medium for RAIR-DTC. Background Technology

[0002] Differentiated thyroid cancer (DTC) is the most common malignant tumor of the endocrine system in clinical practice. Although the prognosis of DTC is good, about 20%-30% of patients will experience disease recurrence or distant metastasis. For these patients, the main treatment is surgery combined with high-dose radioactive iodine (RAI) or I-131 clearance or treatment.

[0003] As the disease progresses, some DTC patients experience dedifferentiation of their cancer cells, causing the cells to lose their ability to take up radioactive iodine, thus becoming ineffective against RAI treatment. This is known as RAIR-DTC (radioactive iodine-refractory differentiated thyroid cancer). RAIR-DTC has a significantly worse prognosis and is a major cause of death from thyroid cancer.

[0004] Current technologies often rely on a single test result to predict whether a patient's DTC will develop into RAIR-DTC, and the prediction results are unsatisfactory. At present, there is still a lack of a predictive technology with reference value. Summary of the Invention

[0005] In view of this, the purpose of this disclosure is to provide a method, apparatus, electronic device and storage medium for calculating the probability of RAIR-DTC, so as to improve the accuracy of probability prediction of DTC evolving into RAIR-DTC.

[0006] In a first aspect, embodiments of this disclosure provide a method for calculating the probability of RAIR-DTC, the method comprising: registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of a patient with differentiated thyroid cancer to obtain a registered image set; extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient; classifying the tumor lesions in the patient according to the feature data to classify them as metabolic heterogeneous lesions to obtain lesion type information corresponding to each tumor lesion; and predicting the probability of the patient having RAIR-DTC based on the lesion type information.

[0007] Secondly, embodiments of this disclosure provide a probability calculation device for RAIR-DTC, the device comprising: a registration module for registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of a patient with differentiated thyroid cancer to obtain a registered image set; an extraction module for extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient; a classification module for classifying the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion; and a prediction module for predicting the probability of the patient having RAIR-DTC based on the lesion type information.

[0008] Thirdly, embodiments of this disclosure provide an electronic device, including a processor and a memory, wherein the memory stores machine-executable instructions that can be executed by the processor, and the processor executes the machine-executable instructions to implement the above-described RAIR-DTC probability calculation method.

[0009] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to implement the aforementioned RAIR-DTC probability calculation method.

[0010] The embodiments disclosed herein bring the following beneficial effects:

[0011] The aforementioned RAIR-DTC probability calculation method, device, electronic equipment, and storage medium, by introducing multimodal image registration and lesion-level metabolic heterogeneity classification, align with the principles of dedifferentiation biology in thyroid cancer, significantly improving the accuracy of RAIR-DTC probability prediction and providing valuable data support for clinicians' treatment decisions.

[0012] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure are realized and obtained through the structures particularly pointed out in the description, claims and drawings.

[0013] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the specific embodiments of this disclosure or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0015] Figure 1 This is a flowchart of one embodiment of the probability calculation method for RAIR-DTC in this disclosure;

[0016] Figure 2 A schematic diagram of a probability calculation device for RAIR-DTC provided in an embodiment of this disclosure;

[0017] Figure 3 A registered image set of a local human body in a probability calculation method for RAIR-DTC provided in this embodiment of the present disclosure;

[0018] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0020] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of this disclosure is described below. Please refer to [link / reference]. Figure 1 One embodiment of the probability calculation method for RAIR-DTC in this disclosure includes:

[0022] Step S10: Register the whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set;

[0023] This disclosure pertains to a method for predicting the probability of differentiated thyroid cancer (DTC) progressing to RAIR-DTC in DTC patients. Differentiated thyroid cancer, a broad category, includes papillary thyroid carcinoma (PTC) and follicular thyroid carcinoma (FTC). DTC is characterized by cancer cells retaining some features of normal thyroid follicular cells, particularly the ability to take up radioactive iodine (RAI). Therefore, radioactive iodine is typically used post-operatively. Treatment or removal of residual thyroid tissue or metastatic lesions.

[0024] Radioactive iodine-refractory differentiated thyroid cancer (RAIR-DTC) is a subgroup of differentiated thyroid cancer (DTC) that refers to DTC that has lost or partially lost its response to radioactive iodine therapy. In some patients with locally advanced or metastatic DTC, tumor cell dedifferentiation leads to decreased expression or abnormal function of the sodium-iodine cotransporter (NIS), resulting in a loss of the ability to effectively take up radioactive iodine. This portion of DTC that is insensitive to radioactive iodine therapy is defined as RAIR-DTC.

[0025] Whole-body iodine metabolism scan images are taken by patients with DTC who have received oral or intravenous radioactive iodine (such as... Afterwards, images obtained through whole-body scintillation scanning or SPECT / CT (Single-Photon Emission Computed Tomography and computed tomography) can reflect the tumor cells' ability to take up iodine. Among them, SPECT / CT is a medical imaging technique that combines SPECT and CT.

[0026] Whole-body glucose metabolism scan images are taken when DTC patients receive intravenous injections of radioactive tracers (such as fluorodeoxyglucose). After FDG, images obtained by PET (Positron emission tomography) or PET / CT (Positron emission tomography and computed tomography) can reflect the glucose metabolism level of tumor cells. PET / CT is a medical imaging technique that combines PET and CT.

[0027] Anatomical imaging, specifically CT or MRI (Magnetic Resonance Imaging) images of DTC patients, can reflect the location and morphological information of tumor lesions. In one implementation, whole-body iodine metabolism scans can be taken 2-10 days after the first or any subsequent high-dose radioactive iodine therapy (RAI) in DTC patients. The timing of whole-body glucose metabolism scans and anatomical imaging can be the same as that of whole-body iodine metabolism scans, allowing for more accurate calculation of the RAIR-DTC probability.

[0028] In this embodiment, registering the whole-body iodine metabolism scan image, whole-body glucose metabolism scan image, and anatomical structure image of the DTC patient can ensure that the images of the same location in the registered image set belong to the same location on the DTC patient, thereby performing multimodal tumor lesion analysis on the same location on the DTC patient and improving the accuracy of RAIR-DTC probability calculation.

[0029] In one implementation, a preset registration algorithm can be used for registration. The resulting registered image set is a set of images aligned with body parts in the same spatial coordinate system. During registration, rigid or non-rigid registration algorithms can be used. Specifically, rigid registration algorithms are based on bony landmarks (such as the spine) for body part registration, while non-rigid registration algorithms are based on voxels of mutual information or normalized cross-correlation. The specific method is not limited here.

[0030] In one implementation, the registration employs a two-stage strategy combining rigid registration based on mutual information with elastic registration using B-spline free deformation. Specifically: First, normalized mutual information is used as a similarity measure to perform three-dimensional rigid transformations on whole-body iodine metabolism scans and whole-body glucose metabolism scans with anatomical structure images, respectively. The transformation parameters include three translational degrees of freedom and three rotational degrees of freedom. The optimizer uses an adaptive gradient descent method, and the iteration termination condition is that the change in mutual information value is less than a certain threshold. .

[0031] After rigid registration, a B-spline-based free deformation model was used for local elastic registration. The control point grid spacing was set to 64×64×64 mm, and the bending energy penalty term weight was set to 0.05 to compensate for changes in patient position and minor organ deformation. Registration accuracy was assessed using the Euclidean distance of anatomical landmarks (such as the tracheal carina and the upper edge of the aortic arch). An average registration error of less than 3 mm was considered acceptable, ensuring a one-to-one spatial correspondence for subsequent multimodal features.

[0032] In another implementation, an unsupervised registration network based on deep learning (such as the VoxelMorph architecture) can be directly used. Anatomical images are used as fixed images, while iodine metabolism and glucose metabolism images are used as floating image inputs. The registered image set is output synchronously through a spatial transformation network. The training loss function consists of local normalized cross-correlation and deformation field smoothing regularization terms, which effectively improves the efficiency of processing large amounts of data.

[0033] like Figure 3 The image shown is a registered image set of a localized human body. The leftmost part is a portion of a whole-body iodine metabolism scan, with bright areas (such as the neck and abdomen) representing the accumulation of radioactive iodine. The middle part is a portion of a whole-body glucose metabolism scan, with highlighted areas representing regions of active glucose metabolism. The rightmost part is a portion of an anatomical structure image. Figure 3 It is evident that the whole-body iodine metabolism scan, whole-body glucose metabolism scan, and anatomical structure images in the registered image set are aligned with each other, and the image coordinates of the same location in the human body are the same. This provides multi-dimensional fused data for the feature analysis and classification of the same tumor lesion, thereby improving the accuracy of RAIR-DTC prediction.

[0034] In one implementation, whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images taken at different times of a DTC patient can be registered separately to obtain registered image sets corresponding to different times. These sets can be used to obtain feature data that can be used to indicate changes in tumor lesions when extracting tumor lesion features, thereby improving the accuracy of RAIR-DTC probability prediction.

[0035] Step S20: Extract tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient;

[0036] It is understood that combined image data of any part of a DTC patient can be obtained from the registered image set, including whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images, which is a type of multimodal image data. This embodiment extracts tumor lesion features from the multimodal image data to obtain feature data of at least one tumor lesion in a DTC patient. The feature data can include static features, change / dynamic features, and static / dynamic features reflected by at least two types of images. Static features can include iodine metabolism features, glucose metabolism features, morphological features, and texture features. The image types include the aforementioned whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images.

[0037] Specifically, the feature data may include features reflected by both whole-body iodine metabolism scans and whole-body glucose metabolism scans, features reflected by both whole-body iodine metabolism scans and anatomical images, features reflected by both whole-body glucose metabolism scans and anatomical images, and features reflected by both whole-body iodine metabolism scans, whole-body glucose metabolism scans and anatomical images; no specific limitation is made here.

[0038] In one implementation, an artificial intelligence model can be used to extract tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient. Specifically, the artificial intelligence model is trained in advance using sample data composed of the registered image set and labeled tumor lesion features. It can extract feature data of tumor lesions from the multimodal registered image set to improve the accuracy of RAIR-DTC probability prediction.

[0039] In one implementation, an artificial intelligence model can be used to extract morphological features, first-order statistical features, second-order statistical features, and higher-order filtering features from the registered image set to obtain feature data of at least one tumor lesion on the patient. This feature data is a high-dimensional vector that integrates all the features extracted above.

[0040] Specifically, morphological features can include the geometric shape of the tumor lesion, such as surface area, sphericity, and compactness. First-order features can include the distribution of pixel (or voxel) intensity within the tumor lesion, such as skewness and kurtosis. Second-order features can include the spatial relationship between pixel intensity within the tumor lesion, such as gray-level co-occurrence matrix and gray-level run matrix. Higher-order filtering features can be features extracted from images processed by wavelet transform or Laplacian of Gaussian (LoG) filtering, which can indicate information about different scales and textures in the image, as well as the microenvironment and angiogenesis status of the tumor lesion.

[0041] In one implementation, the aforementioned artificial intelligence model may adopt an end-to-end model structure that does not rely on a preset radiomics formula, but instead learns abstract features related to RAIR-DTC progression, dedifferentiation, or prognosis through neural networks as feature data of at least one tumor lesion in the patient.

[0042] Step S30: Based on the feature data, classify the tumor lesions on the patient according to metabolic heterogeneity to obtain the lesion type information corresponding to each tumor lesion;

[0043] In this embodiment, based on the obtained feature data, each tumor lesion in the patient can be classified as a metabolic heterogeneous lesion. The classification result can be a binary classification result or a multi-class classification result. When it is a binary classification result, the lesion type information corresponding to each tumor lesion can indicate whether the corresponding tumor lesion is a metabolic heterogeneous lesion or a metabolic non-heterogeneous lesion. When it is a multi-class classification result, the lesion type information corresponding to each tumor lesion can indicate whether the corresponding tumor lesion is an iodine-uptake lesion, a non-iodine-uptake lesion, a hypermetabolic lesion, a non-hypermetabolic lesion, a metabolic reversal lesion, or a non-metabolic reversal lesion. Metabolic heterogeneous lesions can also include other types, which are not limited here.

[0044] In one implementation, the lesion type information may include the lesion type and confidence level of the corresponding tumor lesion. The tumor lesions in the patient can be classified into metabolic heterogeneous lesions by clustering algorithms (such as K-means, hierarchical clustering) or machine learning classifiers (such as support vector machine SVM, decision tree, deep learning model) to obtain the lesion type information corresponding to each tumor lesion.

[0045] The aforementioned machine learning classifier can be a supervised learning-based classifier. It pre-collects pathologically or clinically confirmed RAIR-DTC cases as a training set, labels the true type of each lesion, and trains the extracted feature data using random forest or XGBoost algorithms. Hyperparameters (e.g., 500 trees, maximum depth 8) are adjusted through 10-fold cross-validation, and the final output is the predicted probability of the lesion type. Model performance is evaluated using the area under the ROC curve (AUC), with a value greater than 0.85 considered clinically usable.

[0046] Step S40: Based on the lesion type information, predict the probability that the patient has RAIR-DTC.

[0047] In this embodiment, based on the lesion type information corresponding to each tumor lesion in a DTC patient, an artificial intelligence model can be used to predict the risk probability that the DTC in the patient will develop into RAIR-DTC, i.e., the probability that the patient will develop RAIR-DTC. In one embodiment, logistic regression, Cox proportional hazards regression model, or deep neural network can be used, and the specific method is not limited here.

[0048] If the probability prediction model for a patient developing RAIR-DTC is constructed using a logistic regression model or a Cox proportional hazards regression model, the input variables can include: the proportion of metabolic reversal lesions among all tumor lesions in the patient, the highest confidence probability among metabolic reversal lesions, the total lesion volume, and the patient's thyroglobulin level. The model outputs a probability value ranging from 0 to 1, with a cutoff value set to 0.5, meaning a probability greater than 0.5 is considered high risk.

[0049] In one implementation, the probabilistic prediction model is evaluated using a calibration curve on an independent validation set, with a Brier score less than 0.15, indicating good consistency between the predicted probability and the actual frequency of occurrence. Furthermore, personalized prediction reports can be generated, displaying the classification results of each lesion and its contribution to the overall risk, assisting clinicians in making treatment decisions.

[0050] In one implementation, the probability of a patient developing RAIR-DTC can be predicted by combining lesion type information with the patient's clinical data (such as physiological data and TNM staging of malignant tumors) to improve prediction accuracy.

[0051] In one implementation, the number of tumor lesions of a specified type on the patient can be compared with multiple preset ranges based on lesion type information. The preset probability corresponding to the range is then determined as the probability that the DTC patient has RAIR-DTC. For example, assuming the number of tumor lesions of a specified type on the DTC patient is 3, belonging to the preset range [3,5], and assuming the preset probability corresponding to this range is 80%, then the probability that the DTC patient has RAIR-DTC is 80%. Specific details are not limited here.

[0052] The RAIR-DTC probability calculation method provided by the above embodiments, by introducing multimodal image registration and lesion-level metabolic heterogeneity classification, conforms to the principle of dedifferentiation biological process of thyroid cancer, which greatly improves the accuracy of RAIR-DTC probability prediction and provides valuable data support for clinicians' treatment decisions.

[0053] Next, we will explain in detail how the probability calculation method of RAIR-DTC is performed.

[0054] In one embodiment, the feature data includes index feature data and heterogeneous feature data; the step of extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient includes: identifying regions of interest (ROIs) of tumor lesions in the registered image set to obtain regions of interest (ROIs) of at least one tumor lesion in the registered image set; calculating iodine uptake capacity and glucose metabolism level indicators for the ROIs based on the registered image set to obtain index feature data; and identifying metabolic heterogeneity features for the ROIs based on the registered image set to obtain heterogeneous feature data.

[0055] A region of interest (ROI) is the three-dimensional boundary of a suspected tumor lesion identified on registered multimodal images. It can be segmented using thresholding algorithms (e.g., based on glucose metabolism images, SUV > 2.5 and higher than 40% of the maximum SUV), fully automated segmentation algorithms (e.g., deep learning-based U-Net models), or multimodal joint segmentation algorithms; the specific method is not limited here. In one implementation, when identifying the ROI, the morphology can be determined based on anatomical images (CT / MRI) and functional images (FDG-PET and...). Positive signal regions in the scan serve as the boundaries of biological activity.

[0056] When using a thresholding algorithm, firstly, a nuclear medicine physician can delineate a rough bounding box containing suspected lesions on anatomical images (CT / MRI). Then, on a whole-body glucose metabolism scan, voxels within the bounding box are segmented using a preset standardized uptake threshold. For example, the threshold can be set to 40% of the maximum standardized uptake value (SUVmax), retaining voxels with a standardized uptake value greater than 0.4 × SUVmax as the active portion of the tumor lesion. For whole-body iodine metabolism scans, a lower limit threshold for the target-to-background ratio (e.g., target-to-background ratio > 1.5) can be set to exclude interference from physiologically iodine-uptaken tissues. After segmentation, three-dimensional connected component analysis is used to remove isolated noise regions smaller than 0.5 mL, and the final set of voxels constitutes the region of interest.

[0057] When sampling the U-Net model based on deep learning, a 3D U-Net convolutional neural network model can be pre-trained for automatic identification of regions of interest. The training data consists of a multi-center historical case registration image set and physician-annotated lesion masks. The network input is multi-channel 3D volumetric data stitched from the registration image set, and the output is a probability map of each voxel belonging to a tumor lesion. Binarization is performed by setting a probability threshold (e.g., 0.5), and the boundaries are smoothed using a conditional random field post-processing method to obtain the final region of interest segmentation result. This method can significantly reduce the workload of manual delineation and improve segmentation consistency across operators.

[0058] To further improve segmentation accuracy when using multimodal joint segmentation algorithms, a multimodal fuzzy C-means clustering algorithm can be employed. This algorithm constructs a three-dimensional feature vector from the three modal values ​​(iodine metabolism count, glucose metabolism SUV value, and anatomical structure CT value) of each voxel in the registered image set. By iteratively optimizing the objective function, voxels are classified into lesion, normal tissue, and background classes, obtaining soft segmentation results for the region of interest. The final boundary is then determined using the maximum membership principle. This approach utilizes complementary information between different modalities and is particularly suitable for identifying invasive lesions with blurred boundaries.

[0059] Indicator characteristic data refers to macroscopic and global quantitative indicators that reflect the overall function and metabolic level of the tumor, including iodine uptake capacity indicators and glucose metabolism level indicators. Among them, iodine uptake capacity indicators may include target-to-background ratio, maximum uptake value of lesions, total uptake value of lesions, and iodine uptake ratio (i.e., the ratio of tumor uptake to the count of reference background tissue (such as blood pool). Glucose metabolism level indicators may include maximum standardized uptake value, peak standardized uptake value, average standardized uptake value, and total lesion glycolysis.

[0060] Taking the target-to-background ratio as an example, the calculation process of the iodine uptake capacity index includes: first, determining the average radioactivity count within the region of interest. The selection of the background region can be adaptively determined based on the lesion location—for neck lesions, contralateral normal thyroid tissue or posterior cervical triangle muscle is preferentially selected; for lung lesions, adjacent normal lung parenchyma is selected; for bone lesions, contralateral normal bone is selected. Next, the average radioactivity count of the background region is calculated. Then the target-to-background ratio is .

[0061] If there is uncertainty in the selection of the background region, multiple candidate background regions can be selected simultaneously, and the median of the target-to-background ratio can be used as a robust estimate. Furthermore, the maximum uptake value of a lesion is the highest radioactive count of a single voxel within the region of interest; the total uptake value of a lesion is the sum of the radioactive counts of all voxels within the region of interest, reflecting the overall iodine uptake load of the lesion.

[0062] When calculating glucose metabolism levels, the voxel values ​​of whole-body glucose metabolism scans can be converted into standardized uptake values ​​based on the patient's weight, injection dose, and radioactive decay correction factor. :

[0063]

[0064] in, Let x be the radioactivity concentration at voxel x, and W be the patient's weight. This is the injection dose. for The decay constant is given by t, where t is the time interval between injection and scanning.

[0065] In glucose metabolism indicators, the maximum standardized uptake (MAU) is the maximum value of the standardized uptake values ​​on a voxel within the region of interest (ROI); the peak standardized uptake (PSU) is the voxel containing the MAU located within the ROI. The average standardized uptake within the spherical volume; the average standardized uptake is the mean of the standardized uptake values ​​of all voxels within the region of interest; total lesion glycolysis is the product of the average standardized uptake and the volume of the region of interest.

[0066] Heterogeneous feature data refers to quantitative indicators that reflect the differences in the tumor's internal microenvironment, cell density, or functional distribution, including at least texture and shape features extracted from iodine metabolism images and glucose metabolism images.

[0067] Specifically, heterogeneous feature data can include: first-order statistical features (such as skewness—indicating the asymmetry of the SUV or I signal distribution, kurtosis—indicating the sharpness of the SUV or I signal distribution), gray-level co-occurrence matrix features (such as contrast—indicating the gray-level difference between high-metabolic and low-metabolic regions within the lesion, entropy—indicating the complexity or disorder of the local gray-level distribution), and the heterogeneity of ratio features (comparing iodine uptake capacity indicators with glucose metabolism level indicators, and then calculating the local variance or range of the ratio image to quantify the spatial variability of the "Flip-Flop" phenomenon).

[0068] When identifying metabolic heterogeneity features, first-order, second-order, and higher-order statistical features of the registered image set can be identified. First-order statistical features include: mean, variance, skewness, kurtosis, energy, and entropy. Skewness reflects the asymmetry of the gray-level distribution; a positive value indicates a right-skewed distribution (the presence of a few high uptake hotspots). Kurtosis reflects the sharpness of the gray-level distribution; a high kurtosis value suggests voxel aggregation with extreme uptake values.

[0069] Second-order statistical features can include: contrast, homogeneity, correlation, and angular second moments calculated based on the gray-level co-occurrence matrix. Among them, high contrast indicates significant differences in uptake levels in local areas, suggesting the existence of cell subpopulations with varying degrees of differentiation within the lesion; low homogeneity indicates that metabolic activity is distributed unevenly in a patchy manner.

[0070] The process of recognizing high-order texture features may include: performing wavelet decomposition on the image (e.g., using the Daubechies-4 wavelet basis, first-order decomposition), and repeatedly calculating the aforementioned first-order and second-order features for the low-frequency approximation subband and the high-frequency detail subband, respectively, to obtain high-order texture features. In this embodiment, the registered image set is decomposed into different frequency components through wavelet decomposition, wherein the high-frequency subband features are related to the microstructure inside the lesion (such as micronecrotic foci and fibrous septa).

[0071] If a registered image set from multiple time points of a patient is obtained, dynamic heterogeneous feature data can be extracted. Specifically, the rate of change of iodine uptake capacity and glucose metabolism indicators (such as the rate of decrease in the target-to-background ratio and the doubling time of total lesion glycolysis) of the same lesion at different time points can be calculated, along with the temporal difference vector of texture features, to describe the evolutionary trajectory of lesion dedifferentiation. These dynamic features have additional predictive value for the early identification of lesions that are about to develop into RAIR-DTC.

[0072] In one embodiment, the iodine uptake capacity index includes the target-to-background ratio, and the glucose metabolism level index includes the standardized uptake value of fluorodeoxyglucose. The step of calculating the iodine uptake capacity index and glucose metabolism level index for the region of interest based on the registered image set to obtain index feature data includes: calculating the ratio between the average radioactivity count in each region of interest and the average radioactivity count in the background region based on the whole-body iodine metabolism scan image to obtain the target-to-background ratio; and calculating the maximum standardized uptake value and the peak standardized uptake value of fluorodeoxyglucose in each region of interest based on the whole-body glucose metabolism scan image to obtain the standardized uptake value.

[0073] The target-to-background ratio (TBR) is the ratio of the mean radioactive count within a tumor lesion's region of interest (ROI) to the mean radioactive count in a reference background region (background). Using TBR as a relative indicator can effectively eliminate absolute count differences caused by factors such as patient injection dosage, imaging time, and equipment calibration, thereby improving the comparability and robustness of iodine uptake quantification.

[0074] The process of obtaining the mean radioactivity count of the region of interest may include: First, in the whole-body iodine metabolism scan image, based on the identified region of interest, extract the radioactivity count of all voxels in the region and calculate the arithmetic mean; then, based on the anatomical structure image, for lesions in different anatomical locations, use different adaptive background selection rules to select background regions, calculate the arithmetic mean of the radioactivity count of all voxels in the corresponding background regions, and finally calculate the ratio between the mean radioactivity count of each region of interest and the mean radioactivity count of the background regions to obtain the target-background ratio.

[0075] To improve robustness, in one implementation, the median target-to-background ratio can be used. This involves randomly sampling multiple sub-regions within the background area, calculating the target-to-background ratio for each sub-region, and taking the median as the iodine uptake capacity indicator to reduce the impact of occasional high uptake points within the background area. In another implementation, the logarithmic target-to-background ratio can also be calculated as the iodine uptake capacity indicator. The logarithmic transformation makes the indicator distribution closer to a normal distribution, facilitating subsequent statistical analysis.

[0076] Maximum standardized uptake (MAX) is the highest value of fluorinated deoxyglucose (FDG) uptake within a region of interest (ROI). It is an indicator of glucose metabolic activity and represents the most metabolically active cell population in the lesion. The highest voxel SUV value in the ROI can be directly extracted as the MAX.

[0077] The standardized peak uptake (PV) is obtained by taking a spherical volume of fixed size (e.g., 1 cm in diameter) within the region of interest (ROI), calculating the average SUV value within that volume, and identifying the largest average SUV value as the standardized peak uptake. SUV is a normalized index calculated by dividing the FDG uptake value (kBq / mL) by the injection dose and the patient's weight or body surface area, ensuring comparability between different patients.

[0078] In one embodiment, the heterogeneous feature data is used to indicate: the metabolic heterogeneity characteristics of the tumor lesion, and the characteristics of the microenvironment and angiogenesis status; the step of identifying metabolic heterogeneity features of the region of interest based on the registered image set to obtain heterogeneous feature data includes: for each region of interest, identifying the spatial relationship between pixel intensities of aligned pixels in the registered image set to generate a gray-level co-occurrence matrix and a gray-level run-length matrix; wherein the gray-level co-occurrence matrix and the gray-level run-length matrix are used to indicate the metabolic heterogeneity characteristics of the tumor lesion within each region of interest; for each region of interest, performing wavelet transform and Laplacian filtering on the registered image set to obtain tumor lesion size texture information; wherein the tumor lesion size texture information is used to indicate the characteristics of the tumor lesion's microenvironment and angiogenesis status.

[0079] The Gray Level Co-Occurrence Matrix (GLCM) is used to record the frequency of intensity combinations of two pixels within a Region of Interest (ROI) that are separated by a certain distance (displacement) and angle. Features derived from the GLCM (such as contrast, homogeneity, and entropy) can quantify metabolic non-uniformity. Specifically, contrast reflects the degree of difference in local metabolic activity, homogeneity reflects the uniformity of metabolic distribution, and entropy reflects the complexity and disorder of metabolic patterns.

[0080] For example, a contrast ratio above a certain threshold indicates an SUV or [other type of SUV] in a local area. Significant differences in signals suggest the existence of regions with marked metabolic differences within the tumor, such as the coexistence of dedifferentiated "hot zones" and highly differentiated "cold zones," which is a key indicator of metabolic heterogeneity.

[0081] The gray-level co-occurrence matrix (GLCM) indicates the joint probability of voxels with specific gray-level value pairs in space at a given direction and distance. For a 3D region of interest, the GLCM is defined in 13 standard directions (an independent set of directions in 3D space corresponding to vertices, edge centers, face centers, etc.) at distances... Calculate the gray-level co-occurrence matrix using voxels as the step size. Its elements are defined as:

[0082]

[0083] in, For direction The upper length is displacement vector, This is the indicator function. Averaging the gray-level co-occurrence matrices in the 13 directions yields the rotation-invariant gray-level co-occurrence matrix.

[0084] The Gray Level Run Length Matrix (GLRLM) is used to record the number (run length) of pixels with the same gray value that appear consecutively along a certain direction within a Region of Interest (ROI). Features derived from GLRLM (such as long run emphasis, short run emphasis, and run percentage) can quantify the roughness and directionality of tumor internal structures. A larger long run emphasis value indicates the presence of a large, continuous, and uniform metabolic region; a larger short run emphasis value indicates a more fragmented distribution of high-metabolic regions; and the run percentage reflects the coarseness of the texture. For example, a high long run emphasis value suggests the presence of a large area of ​​homogeneous tissue (such as necrosis or a homogeneous high-metabolic region).

[0085] The grayscale run-length matrix indicates the length distribution of a continuous sequence of voxels (i.e., "runs") with the same grayscale value in a given direction. For a 3D region of interest, the grayscale run-length matrix is ​​calculated in each of the 13 directions. Its elements represent direction The number of runs with grayscale value i and run length l. Similarly, averaging over directions yields the rotation-invariant grayscale run matrix.

[0086] Wavelet transform is a process that decomposes an image into different frequency sub-bands (high frequencies correspond to details, and low frequencies correspond to macroscopic structures) to obtain multi-scale, multi-resolution texture information. This helps to discover tiny structures of different sizes (such as micronecrosis or microvessels).

[0087] When performing wavelet transform, a three-dimensional discrete wavelet transform can be used to decompose the region of interest into multiple scales. The wavelet basis functions are selected as Coiflet-1 wavelet or Daubechies-4 wavelet. A first-level decomposition is performed to obtain one low-frequency approximation subband (LLL) and seven high-frequency detail subbands (LLH, LHL, LHH, HLL, HLH, HHL, HHH), which serve as the subband images after wavelet transform.

[0088] The Laplacian filter is a high-pass filter used for edge detection and sharpening. It enhances details and gradients in images, highlighting minute structures, boundary irregularities, and cystic regions within the tumor microenvironment. For example, applying Laplacian filtering to FDG-PET images extracts the shape irregularities (e.g., fractal dimension) of the enhanced image, indicating the invasiveness of tumor edges. During Laplacian filtering, a three-dimensional Laplacian-Gaussian filter can be used to convolve the region of interest to obtain the Laplacian-filtered response image.

[0089] Furthermore, the entropy value based on the gray-level co-occurrence matrix and the short run emphasis feature based on the gray-level run matrix can be calculated for each sub-band image after wavelet transform and the response image after Laplacian filtering, respectively. The set of these recalculated feature parameters in the transform domain / filtering domain is defined as the tumor lesion size texture information.

[0090] In one embodiment, the step of classifying the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion includes: classifying the region of interest of each tumor lesion according to the indicator feature data to obtain a first classification result corresponding to each region of interest; classifying the region of interest of each tumor lesion according to the heterogeneous feature data to obtain a second classification result corresponding to each region of interest; and determining the lesion type information corresponding to each tumor lesion according to the first classification result and the second classification result.

[0091] The first classification result is a preliminary, functional classification of the lesions based on these macroscopic indicators. The classification model can be a threshold-based determination or a preliminary classifier trained based on indicator features (such as logistic regression or SVM). As an example, and not a limitation, the first classification result can be divided into three or four categories, such as double-positive, double-negative, iodine-hypermeable, and glucose-hypermeable types. In one implementation, a logistic regression model can be used to obtain the first classification result corresponding to each region of interest.

[0092] The second classification result is based on these microscopic features to categorize the lesions by their invasiveness risk. The classification model can employ a high-dimensional data classifier (such as random forest, SVM, or shallow neural networks). As an example, and not a limitation, the second classification result can be divided into highly heterogeneous invasive lesions and low heterogeneous stable lesions.

[0093] Taking the random forest model as an example of a classifier, the training process may include: collecting RAIR-DTC and non-RAIR-DTC cases confirmed by pathological biopsy or clinical follow-up, extracting heterogeneous feature data of their lesions as training samples, and labeling them with RAIR-DTC status. Ten-fold cross-validation is used to adjust the model hyperparameters; typical settings include 500 decision trees, a maximum depth of 8, and a minimum number of split samples of 5. Model performance is evaluated using the area under the curve (AUC), with a value greater than 0.80 considered clinically usable.

[0094] When determining the lesion type information corresponding to each tumor lesion, weighted voting or fusion models can be used to integrate the first classification result and the second classification result. For example, if the first classification result is glucose hypermetabolism (functional dedifferentiation) or the second classification result is high heterogeneous invasion (structural invasion), the final type is determined to be a high RAIR-DTC risk lesion.

[0095] In one embodiment, the step of predicting the probability of the patient having RAIR-DTC based on the lesion type information includes: if the lesion type information indicates that the probability of any tumor lesion being a metabolic reversal is greater than a preset probability threshold, then determining the probability of the patient having RAIR-DTC based on the metabolic reversal probability corresponding to the tumor lesion whose probability of being a metabolic reversal is greater than the preset probability threshold.

[0096] Metabolic flip-flop refers to a tumor lesion exhibiting both decreased (or absent) iodine uptake and significantly increased glucose metabolism. If at least one tumor lesion in a patient has a probability of exhibiting flip-flop greater than a preset probability threshold, the maximum, minimum, or median probability can be directly used as the probability of the patient having RAIR-DTC. Alternatively, the probability of the patient having RAIR-DTC can be determined using other methods based on the metabolic flip-flop probabilities corresponding to tumor lesions with a probability of flip-flop greater than a preset probability threshold; details will not be elaborated here.

[0097] In one embodiment, the step of registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of a patient with differentiated thyroid cancer to obtain a registered image set includes: structurally aligning the whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of the patient with differentiated thyroid cancer to obtain a registered image set.

[0098] Structural alignment is the process of spatially transforming two or more images (including translation, rotation, scaling, etc.) using a registration algorithm to make corresponding anatomical structures in the images overlap. The anatomical structure image can be used as a reference image. An iodine metabolism scan image (floating image) is then transformed into the space of the reference image using a registration algorithm, and the glucose metabolism scan image (floating image) is also transformed into the space of the reference image using a registration algorithm.

[0099] For the corresponding method embodiments described above, see [link to relevant documentation]. Figure 2The diagram shows a schematic of a probability calculation device for RAIR-DTC. The device includes: a registration module 20, used to register whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of a patient with differentiated thyroid cancer to obtain a registered image set; an extraction module 22, used to extract tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient; a classification module 24, used to classify the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion; and a prediction module 26, used to predict the probability of the patient having RAIR-DTC based on the lesion type information.

[0100] The aforementioned RAIR-DTC probability calculation device, by introducing multimodal image registration and lesion-level metabolic heterogeneity classification, aligns with the principles of dedifferentiation biology in thyroid cancer, significantly improving the accuracy of RAIR-DTC probability prediction and providing valuable data support for clinicians' treatment decisions.

[0101] Optionally, the feature data includes index feature data and heterogeneous feature data; the step of extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient includes: identifying regions of interest (ROIs) of tumor lesions in the registered image set to obtain regions of interest of at least one tumor lesion in the registered image set; calculating iodine uptake capacity and glucose metabolism level indices for the regions of interest based on the registered image set to obtain index feature data; and identifying metabolic heterogeneity features for the regions of interest based on the registered image set to obtain heterogeneous feature data.

[0102] Optionally, the iodine uptake capacity index includes the target-to-background ratio, and the glucose metabolism level index includes the standardized uptake value of fluorodeoxyglucose. The step of calculating the iodine uptake capacity index and glucose metabolism level index for the region of interest based on the registered image set to obtain index feature data includes: calculating the ratio between the average radioactivity count in each region of interest and the average radioactivity count in the background region based on the whole-body iodine metabolism scan image to obtain the target-to-background ratio; and calculating the maximum standardized uptake value and the peak standardized uptake value of fluorodeoxyglucose in each region of interest based on the whole-body glucose metabolism scan image to obtain the standardized uptake value.

[0103] Optionally, the heterogeneous feature data is used to indicate: the metabolic heterogeneity characteristics of the tumor lesion, as well as the microenvironment and angiogenesis status characteristics; the step of identifying metabolic heterogeneity features of the region of interest based on the registered image set to obtain heterogeneous feature data includes: for each region of interest, identifying the spatial relationship between pixel intensities of aligned pixels in the registered image set to generate a gray-level co-occurrence matrix and a gray-level run-length matrix; wherein, the gray-level co-occurrence matrix and the gray-level run-length matrix are used to indicate the metabolic heterogeneity characteristics of the tumor lesion within each region of interest; for each region of interest, performing wavelet transform and Laplacian filtering on the registered image set to obtain tumor lesion size texture information; wherein, the tumor lesion size texture information is used to indicate the microenvironment and angiogenesis status characteristics of the tumor lesion.

[0104] Optionally, the step of classifying the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion includes: classifying the region of interest of each tumor lesion according to the indicator feature data to obtain a first classification result corresponding to each region of interest; classifying the region of interest of each tumor lesion according to the heterogeneous feature data to obtain a second classification result corresponding to each region of interest; and determining the lesion type information corresponding to each tumor lesion according to the first classification result and the second classification result.

[0105] Optionally, the step of predicting the probability of the patient having RAIR-DTC based on the lesion type information includes: if the lesion type information indicates that the probability of any tumor lesion being a metabolic reversal is greater than a preset probability threshold, then determining the probability of the patient having RAIR-DTC based on the metabolic reversal probability corresponding to the tumor lesion whose probability of being a metabolic reversal is greater than the preset probability threshold.

[0106] Optionally, the step of registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set includes: structurally aligning the whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set.

[0107] This embodiment also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor. The processor executes the machine-executable instructions to implement the RAIR-DTC probability calculation method described above. This electronic device can be a server or a terminal device.

[0108] See Figure 4As shown, the electronic device includes a processor 100 and a memory 101. The memory 101 stores machine-executable instructions that can be executed by the processor 100. The processor 100 executes the machine-executable instructions to implement the probability calculation method of RAIR-DTC described above.

[0109] Furthermore, Figure 4 The electronic device shown also includes a bus 102 and a communication interface 103, with the processor 100, the communication interface 103 and the memory 101 connected via the bus 102.

[0110] The memory 101 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 102 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0111] The processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 100 or by instructions in software form. The processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 101. The processor 100 reads information from memory 101 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments, for example:

[0112] Whole-body iodine metabolism scans, whole-body glucose metabolism scans, and anatomical structure images of patients with differentiated thyroid cancer are registered to obtain a registered image set. Tumor lesion features are extracted from the registered image set to obtain feature data of at least one tumor lesion in the patient. Based on the feature data, the tumor lesions in the patient are classified according to metabolic heterogeneity to obtain lesion type information corresponding to each tumor lesion. Based on the lesion type information, the probability of the patient having RAIR-DTC is predicted.

[0113] This approach, by introducing multimodal image registration and lesion-level metabolic heterogeneity classification, aligns with the principles of dedifferentiation biology in thyroid cancer, significantly improving the accuracy of RAIR-DTC probability prediction and providing valuable data support for clinicians' treatment decisions.

[0114] Optionally, the feature data includes index feature data and heterogeneous feature data; the step of extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient includes: identifying regions of interest (ROIs) of tumor lesions in the registered image set to obtain regions of interest of at least one tumor lesion in the registered image set; calculating iodine uptake capacity and glucose metabolism level indices for the regions of interest based on the registered image set to obtain index feature data; and identifying metabolic heterogeneity features for the regions of interest based on the registered image set to obtain heterogeneous feature data.

[0115] Optionally, the iodine uptake capacity index includes the target-to-background ratio, and the glucose metabolism level index includes the standardized uptake value of fluorodeoxyglucose. The step of calculating the iodine uptake capacity index and glucose metabolism level index for the region of interest based on the registered image set to obtain index feature data includes: calculating the ratio between the average radioactivity count in each region of interest and the average radioactivity count in the background region based on the whole-body iodine metabolism scan image to obtain the target-to-background ratio; and calculating the maximum standardized uptake value and the peak standardized uptake value of fluorodeoxyglucose in each region of interest based on the whole-body glucose metabolism scan image to obtain the standardized uptake value.

[0116] Optionally, the heterogeneous feature data is used to indicate: the metabolic heterogeneity characteristics of the tumor lesion, as well as the microenvironment and angiogenesis status characteristics; the step of identifying metabolic heterogeneity features of the region of interest based on the registered image set to obtain heterogeneous feature data includes: for each region of interest, identifying the spatial relationship between pixel intensities of aligned pixels in the registered image set to generate a gray-level co-occurrence matrix and a gray-level run-length matrix; wherein, the gray-level co-occurrence matrix and the gray-level run-length matrix are used to indicate the metabolic heterogeneity characteristics of the tumor lesion within each region of interest; for each region of interest, performing wavelet transform and Laplacian filtering on the registered image set to obtain tumor lesion size texture information; wherein, the tumor lesion size texture information is used to indicate the microenvironment and angiogenesis status characteristics of the tumor lesion.

[0117] Optionally, the step of classifying the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion includes: classifying the region of interest of each tumor lesion according to the indicator feature data to obtain a first classification result corresponding to each region of interest; classifying the region of interest of each tumor lesion according to the heterogeneous feature data to obtain a second classification result corresponding to each region of interest; and determining the lesion type information corresponding to each tumor lesion according to the first classification result and the second classification result.

[0118] Optionally, the step of predicting the probability of the patient having RAIR-DTC based on the lesion type information includes: if the lesion type information indicates that the probability of any tumor lesion being a metabolic reversal is greater than a preset probability threshold, then determining the probability of the patient having RAIR-DTC based on the metabolic reversal probability corresponding to the tumor lesion whose probability of being a metabolic reversal is greater than the preset probability threshold.

[0119] Optionally, the step of registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set includes: structurally aligning the whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set.

[0120] This embodiment also provides a computer-readable storage medium storing computer-executable instructions. When these computer-executable instructions are invoked and executed by a processor, they cause the processor to implement the aforementioned RAIR-DTC probability calculation method, for example:

[0121] Whole-body iodine metabolism scans, whole-body glucose metabolism scans, and anatomical structure images of patients with differentiated thyroid cancer are registered to obtain a registered image set. Tumor lesion features are extracted from the registered image set to obtain feature data of at least one tumor lesion in the patient. Based on the feature data, the tumor lesions in the patient are classified according to metabolic heterogeneity to obtain lesion type information corresponding to each tumor lesion. Based on the lesion type information, the probability of the patient having RAIR-DTC is predicted.

[0122] This approach, by introducing multimodal image registration and lesion-level metabolic heterogeneity classification, aligns with the principles of dedifferentiation biology in thyroid cancer, significantly improving the accuracy of RAIR-DTC probability prediction and providing valuable data support for clinicians' treatment decisions.

[0123] Optionally, the feature data includes index feature data and heterogeneous feature data; the step of extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient includes: identifying regions of interest (ROIs) of tumor lesions in the registered image set to obtain regions of interest of at least one tumor lesion in the registered image set; calculating iodine uptake capacity and glucose metabolism level indices for the regions of interest based on the registered image set to obtain index feature data; and identifying metabolic heterogeneity features for the regions of interest based on the registered image set to obtain heterogeneous feature data.

[0124] Optionally, the iodine uptake capacity index includes the target-to-background ratio, and the glucose metabolism level index includes the standardized uptake value of fluorodeoxyglucose. The step of calculating the iodine uptake capacity index and glucose metabolism level index for the region of interest based on the registered image set to obtain index feature data includes: calculating the ratio between the average radioactivity count in each region of interest and the average radioactivity count in the background region based on the whole-body iodine metabolism scan image to obtain the target-to-background ratio; and calculating the maximum standardized uptake value and the peak standardized uptake value of fluorodeoxyglucose in each region of interest based on the whole-body glucose metabolism scan image to obtain the standardized uptake value.

[0125] Optionally, the heterogeneous feature data is used to indicate: the metabolic heterogeneity characteristics of the tumor lesion, as well as the microenvironment and angiogenesis status characteristics; the step of identifying metabolic heterogeneity features of the region of interest based on the registered image set to obtain heterogeneous feature data includes: for each region of interest, identifying the spatial relationship between pixel intensities of aligned pixels in the registered image set to generate a gray-level co-occurrence matrix and a gray-level run-length matrix; wherein, the gray-level co-occurrence matrix and the gray-level run-length matrix are used to indicate the metabolic heterogeneity characteristics of the tumor lesion within each region of interest; for each region of interest, performing wavelet transform and Laplacian filtering on the registered image set to obtain tumor lesion size texture information; wherein, the tumor lesion size texture information is used to indicate the microenvironment and angiogenesis status characteristics of the tumor lesion.

[0126] Optionally, the step of classifying the tumor lesions in the patient according to the feature data to obtain lesion type information corresponding to each tumor lesion includes: classifying the region of interest of each tumor lesion according to the indicator feature data to obtain a first classification result corresponding to each region of interest; classifying the region of interest of each tumor lesion according to the heterogeneous feature data to obtain a second classification result corresponding to each region of interest; and determining the lesion type information corresponding to each tumor lesion according to the first classification result and the second classification result.

[0127] Optionally, the step of predicting the probability of the patient having RAIR-DTC based on the lesion type information includes: if the lesion type information indicates that the probability of any tumor lesion being a metabolic reversal is greater than a preset probability threshold, then determining the probability of the patient having RAIR-DTC based on the metabolic reversal probability corresponding to the tumor lesion whose probability of being a metabolic reversal is greater than the preset probability threshold.

[0128] Optionally, the step of registering whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set includes: structurally aligning the whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set.

[0129] The computer program product of the RAIR-DTC probability calculation method, apparatus, electronic device and storage medium provided in this disclosure includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0131] Furthermore, in the description of the embodiments of this disclosure, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this disclosure based on the specific circumstances.

[0132] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0133] In the description of this disclosure, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this disclosure and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this disclosure. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0134] Finally, it should be noted that the above embodiments are merely specific implementations of this disclosure, used to illustrate the technical solutions of this disclosure, and not to limit it. The protection scope of this disclosure is not limited thereto. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this disclosure. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be covered within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the protection scope of the claims.

Claims

1. A method for calculating the probability of RAIR-DTC, characterized in that, The method includes: The whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer were registered to obtain a registered image set. Tumor lesion features are extracted from the registered image set to obtain feature data of at least one tumor lesion in the patient. Based on the aforementioned feature data, the tumor lesions in the patient are classified according to metabolic heterogeneity to obtain lesion type information corresponding to each tumor lesion. Based on the lesion type information, the probability of the patient having RAIR-DTC is predicted.

2. The method according to claim 1, characterized in that, The feature data includes index feature data and heterogeneous feature data; The step of extracting tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient includes: The region of interest (ROI) of tumor lesions is identified in the registered image set to obtain at least one ROI of a tumor lesion in the registered image set. Based on the registered image set, iodine uptake capacity index and glucose metabolism level index are calculated for the region of interest to obtain index feature data. Based on the registered image set, metabolic heterogeneity features are identified in the region of interest to obtain heterogeneous feature data.

3. The method according to claim 2, characterized in that, The iodine uptake capacity index includes the target-to-background ratio, and the glucose metabolism level index includes the standardized uptake value of fluorodeoxyglucose. The steps of calculating iodine uptake capacity and glucose metabolism level indicators for the region of interest based on the registered image set to obtain indicator feature data include: Based on the whole-body iodine metabolism scan images, the ratio between the average radioactivity count in each region of interest and the average radioactivity count in the background region is calculated to obtain the target-background ratio. Based on the whole-body glucose metabolism scan images, the maximum standardized uptake value and the peak standardized uptake value of fluorodeoxyglucose in each region of interest are calculated to obtain the standardized uptake value.

4. The method according to claim 2, characterized in that, The heterogeneous feature data are used to indicate: the metabolic heterogeneity of tumor lesions, as well as the characteristics of the microenvironment and angiogenesis. The step of identifying metabolic heterogeneity features in the region of interest based on the registered image set to obtain heterogeneous feature data includes: For each region of interest, spatial relationships between pixel intensities are identified for pixels aligned in the registered image, generating a gray-level co-occurrence matrix and a gray-level run-length matrix; wherein, the gray-level co-occurrence matrix and the gray-level run-length matrix are used to indicate the metabolic heterogeneity characteristics of tumor lesions within each region of interest; For each region of interest, wavelet transform and Laplacian filtering are performed on the registered image set to obtain tumor lesion size and texture information; wherein, the tumor lesion size and texture information is used to indicate the microenvironment and angiogenesis status characteristics of the tumor lesion.

5. The method according to claim 2, characterized in that, The steps of classifying the tumor lesions in the patient according to the aforementioned feature data and obtaining the lesion type information corresponding to each tumor lesion include: Based on the aforementioned indicator feature data, metabolic heterogeneity lesion classification is performed on each region of interest of tumor lesion to obtain the first classification result corresponding to each region of interest; Based on the heterogeneous feature data, metabolic heterogeneous lesion classification is performed on the region of interest of each tumor lesion to obtain the second classification result corresponding to each region of interest; Based on the first classification result and the second classification result, the lesion type information corresponding to each tumor lesion is determined.

6. The method according to claim 1, characterized in that, The step of predicting the probability that the patient has RAIR-DTC based on the lesion type information includes: If the lesion type information indicates that the probability of any tumor lesion being a metabolic reversal is greater than a preset probability threshold, then the probability of the patient having RAIR-DTC is determined based on the metabolic reversal probability corresponding to the tumor lesion whose probability of being a metabolic reversal is greater than the preset probability threshold.

7. The method according to claim 1, characterized in that, The steps for registering whole-body iodine metabolism scans, whole-body glucose metabolism scans, and anatomical images of patients with differentiated thyroid cancer to obtain a registered image set include: Whole-body iodine metabolism scans, whole-body glucose metabolism scans, and anatomical images of patients with differentiated thyroid cancer were structurally aligned to obtain a registered image set.

8. A probability calculation device for RAIR-DTC, characterized in that, The device includes: The registration module is used to register whole-body iodine metabolism scan images, whole-body glucose metabolism scan images, and anatomical structure images of patients with differentiated thyroid cancer to obtain a registered image set. The extraction module is used to extract tumor lesion features from the registered image set to obtain feature data of at least one tumor lesion in the patient. The classification module is used to classify the tumor lesions in the patient according to the feature data, and obtain the lesion type information corresponding to each tumor lesion. The prediction module is used to predict the probability that the patient has RAIR-DTC based on the lesion type information.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the RAIR-DTC probability calculation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the probability calculation method of RAIR-DTC as described in any one of claims 1-7.

Images