A lung nodule assisted detection device for paired CT images

Abstract

The present invention discloses a lung nodule auxiliary detection device for paired CT images. The device includes an acquisition module for obtaining a first CT image and a second CT image of a patient's lungs and respectively corresponding two-dimensional CT slice sequences; a segmentation module connected to the acquisition module to divide each two-dimensional CT slice into a plurality of sub-block matrices of the same size and number; a matrix transformation module connected to the segmentation module to extract the matrix transformation results of each sub-block matrix by using a preset image algorithm; a calculation module connected to the matrix transformation module to calculate the nodule progression factor by comparing the matrix transformation results corresponding to the first and second CT images of the same patient; and an output module connected to the calculation module to comprehensively output the detection result of nodule progression according to the nodule progression factor. The present invention uses a uniquely designed detection quantity to measure the changes of nodules in patients, which can help doctors observe and analyze the progression of lung nodules more accurately.

Description

Technical Field

[0001] The present invention relates to a lung nodule auxiliary detection device for paired CT images and belongs to the field of healthcare informatics. Background Art

[0002] The monitoring of lung nodules is crucial for the early detection and treatment of lung cancer. Traditionally, doctors evaluate the size and density changes of lung nodules by analyzing CT images, but this method is highly subjective and prone to errors. To improve accuracy, researchers have developed 3D nodule volume measurement methods based on image segmentation and 3D filtering techniques, which show better performance on lung images than 2D methods.

[0003] In the Chinese patent application with the application number 202410586882.8, a lung nodule historical examination and comparison method combining deep learning algorithms and image processing techniques is proposed. This method improves the detection and comparison efficiency by identifying the bifurcation points of the main bronchi in the lungs, avoiding the cumbersome steps of manually finding the corresponding positions of the lesions. Nevertheless, this method still relies on volume measurement and requires precise nodule morphology delineation, and its results may be affected by various factors.

[0004] Although these technologies have improved the detection efficiency, they have not fully solved the problem of the statistical significance of detected changes. Future research needs to further optimize the algorithms, improve the detection accuracy and robustness, and at the same time develop new evaluation methods to determine whether the detected changes are clinically significant. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a lung nodule auxiliary detection device for paired CT images.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] An embodiment of the present invention provides a lung nodule auxiliary detection device for paired CT images, including:

[0008] An acquisition module, configured to acquire a first CT image and a second CT image of a patient's lung scan, and respectively acquire two-dimensional CT slice sequences corresponding to the first CT image and the second CT image;

[0009] A segmentation module, connected to the acquisition module, to divide each two-dimensional CT slice in the two-dimensional CT slice sequence into a plurality of sub-block matrices with the same size and number; specifically including: determining the number n to be divided based on the size of the target lung nodule; dividing the two-dimensional CT slice into sub-block matrices; wherein each of the sub-block matrices is a matrix, where m is the size of the sub-block matrix; n is a positive integer and its value is related to the nodule size;

[0010] a matrix transformation module, connected to the segmentation module and built-in with a preset image algorithm, to extract the matrix transformation result of each sub-block matrix through the preset image algorithm for each sub-block matrix; wherein, perform a logarithmic transformation on the pixel HU values of each position element in the (n1, n2) sub-block matrix of the t-th two-dimensional CT slice of the first CT image to obtain the (n1, n2) sub-block matrix of the t-th two-dimensional CT slice ; for the (n1, n2) sub-block matrix of the t-th two-dimensional CT slice , define its binary indicator parameter , where, when the sub-block matrix contains the center point of the target nodule, , otherwise, ; for the t-th two-dimensional CT slice of the first CT image, assume that the sum of the binary indicator parameters of all sub-block matrices is 1, and the center point of the target nodule appears in the sub-block matrix at the position, define , 1 ≤ ≤ n; calculate , as the matrix transformation result of the (n1, n2) sub-block matrix in the t-th CT slice; wherein, 1 ≤ n1, n2 ≤ n, and t is a positive integer;

[0011] a calculation module, connected to the matrix transformation module, to obtain the nodule progression display factor corresponding to the sub-block matrix at this position based on the matrix transformation result of the first CT image and the matrix transformation result of the second CT image for the sub-block matrix at the same slice number and the same position; wherein, for the (n1, n2) sub-block matrix in the t-th CT slice, respectively obtain the matrix transformation result of the first CT image and the matrix transformation result of the second CT image ; calculate the nodule progression display factor of the (n1, n2) sub-block matrix in the t-th CT slice : is a preset hyperparameter and 0 < < 1; indicates that the nodule has a tendency to increase on the t-th CT slice; indicates that the nodule has a tendency to shrink on the t-th CT slice; indicates that there is no obvious change in the nodule on the t-th CT slice;

[0012] An output module, connected to the computing module, to obtain the nodule progression detection quantity of the two-dimensional CT slice at the same number of slices based on the nodule progression demonstration factors of all sub-block matrices of the two-dimensional CT slice; wherein, obtain the nodule progression demonstration factor of each sub-block matrix in the t-th two-dimensional CT slice; based on the nodule progression demonstration factors of all sub-block matrices in the t-th two-dimensional CT slice, calculate the average value As the nodule progression detection quantity of the t-th two-dimensional CT slice; among which , for any , ; and, based on the nodule progression detection quantity of each two-dimensional CT slice, comprehensively output the final nodule progression detection result; wherein, based on the nodule progression detection quantity of the t-th two-dimensional CT slice and hyperparameter , construct two-dimensional curve; calculate the area under the curve of the two-dimensional curve, denoted as the nodule progression detection result of the t-th two-dimensional CT slice; based on the nodule progression detection results of all two-dimensional CT slices, output the final nodule progression detection result.

[0013] Compared with the prior art, the present invention has the following technical effects:

[0014] (1) From the perspective of hypothesis testing, in cooperation with a specific image processing algorithm, a unique and delicate detection quantity is designed to measure the nodule changes of patients. Thus, by quantifying the changes of the lesions in the two CT images before and after, it helps doctors observe and analyze the progression of pulmonary nodules more accurately.

[0015] (2) By studying whether there are significant differences in the local nodule images in the two CT scans of the patient before and after, the detection quantity is used to answer whether the changes of the pulmonary nodules are statistically significant, making up for the key problem that the existing methods cannot answer whether the detected changes are statistically significant.

[0016] (3) The nodule progression detection quantity of each two-dimensional CT slice is represented by a preset hyperparameter, and a two-dimensional curve between the nodule progression detection quantity and the hyperparameter is constructed to eliminate the influence of the hyperparameter by calculating the area under the curve of the two-dimensional curve, improving the accuracy of the final nodule progression detection result. Description of the Drawings

[0017] Figure 1 is a flowchart of a method for assisting in the detection of pulmonary nodules for paired CT images provided by the first embodiment of the present invention;

[0018] Figure 2AIn the first embodiment of the present invention, it is a schematic diagram of two-dimensional CT slices of the same physical location before and after the treatment of patient A;

[0019] Figure 2B In the first embodiment of the present invention, it is a schematic diagram of a two-dimensional curve constructed for patient A;

[0020] Figure 3A In the first embodiment of the present invention, it is a schematic diagram of two-dimensional CT slices of the same physical location before and after the treatment of patient B;

[0021] Figure 3B In the first embodiment of the present invention, it is a schematic diagram of a two-dimensional curve constructed for patient B;

[0022] Figure 4 In the first embodiment of the present invention, it is a confusion matrix for each diameter group under the simulation experiment;

[0023] Figure 5 In the first embodiment of the present invention, it is the numerical value of the evaluation index under the simulation experiment;

[0024] Figure 6 In the first embodiment of the present invention, it is a confusion matrix for each diameter group and the full sample under the real clinical data;

[0025] Figure 7 In the first embodiment of the present invention, it is the numerical value of the evaluation index under the real clinical data;

[0026] Figure 8 In the second embodiment of the present invention, it is a structural diagram of a lung nodule auxiliary detection device for paired CT images;

[0027] Figure 9 In the third embodiment of the present invention, it is a structural diagram of a lung nodule auxiliary detection device for paired CT images. Detailed implementation mode

[0028] The following combines the accompanying drawings and specific embodiments to make a detailed and specific description of the technical content of the present invention.

[0029] The first embodiment

[0030] The first embodiment of the present invention proposes a method for auxiliary detection of lung nodules in paired CT images. This method is based on the principle of hypothesis testing and focuses on analyzing whether there are significant differences in the lung nodule images of patients after two CT scans. This method aims to help doctors more accurately observe and evaluate the progression of lung nodules. In medical practice, doctors usually judge whether a nodule is progressing based on the changes between the nodule and the surrounding normal tissue. If there are no significant changes in the nodule compared with the surrounding tissue, the nodule is considered stable; if there are significant changes, the nodule is considered to be progressing.

[0031] Based on this clinical experience, the embodiments of the present invention design a unique and delicate statistic - the Detection Result of Nodule Progression (DNP), which is used to quantify the nodule changes of an individual patient. The DNP is not only constructed based on the doctor's clinical experience, but also can provide objective conclusions with statistical significance, so as to assist doctors in making more accurate judgments on the progression of patients' pulmonary nodules. This method reduces subjectivity and errors in traditional methods through quantitative analysis, and improves the accuracy and reliability of pulmonary nodule detection.

[0032] Next, the specific steps of this pulmonary nodule auxiliary detection method will be described:

[0033] S1: Obtain a sequence of two-dimensional CT slices.

[0034] First, collect the first CT image of the patient's lung scan before treatment, and use a three-dimensional tensor to define this first CT image; where represents the t-th two-dimensional CT slice, T is the total number of CT slices, p represents the dimension of the two-dimensional CT; R represents the real number field; t is a positive integer.

[0035] In actual situations, a set of real pulmonary CT images diagnosed as positive often contains a large number of CT slices without nodules, and such CT slices will be excluded in subsequent research and analysis to reduce the interference of nodule-irrelevant information. Therefore, in actual implementation, only select a continuous sequence of two-dimensional CT slices containing the target pulmonary nodule, denoted as , where represents the logarithmic transformation of the HU value of the pixel point at the position in the t-th CT slice of

[0036] In an embodiment of the present invention, the process of obtaining a sequence of two-dimensional CT slices is as follows: First, directly obtain the patient's pulmonary CT images through tomographic scanning technology. Subsequently, professional doctors carefully review and annotate these images to identify and determine specific regions covering the range of pulmonary nodules. According to the doctors' annotations, select continuous two-dimensional CT slices containing the target pulmonary nodule to form a sequence. This sequence of two-dimensional CT slices will be used in the subsequent pulmonary nodule auxiliary detection method for accurate analysis and evaluation of the presence and changes of pulmonary nodules.

[0037] S2: Divide each two-dimensional CT slice into multiple sub-block matrices.

[0038] Specifically, after obtaining the two-dimensional CT slice sequence based on step S1, for each two-dimensional CT slice in the two-dimensional CT slice sequence, first determine the number of sub-blocks n to be divided based on the size of the target lung nodule. Here, n is a positive integer, and its value is mainly related to the nodule size. For example, if the dimension of a two-dimensional CT slice is , and the sub-blocks to be segmented are , that is, the size m of each sub-block matrix is 32, then a two-dimensional CT slice can be segmented into sub-blocks of . That is, cut 16 times horizontally and 16 times vertically, so n = 16.

[0039] Then, based on the number of sub-blocks n to be divided, divide each two-dimensional CT slice into sub-block matrices, that is, , where each sub-block matrix is a matrix, . Moreover, the sizes and numbers of the sub-block matrices divided from each two-dimensional CT slice are equal, which is convenient for subsequent comparative analysis.

[0040] It can be understood that for any and , the element at the position of can be given by , where , . For the convenience of description, define the index set , so can be re-recorded as . Subsequently, can be used as the th sub-block matrix of the t-th CT slice.

[0041] S3: Extract the matrix transformation result of each sub-block matrix through a preset image algorithm.

[0042] For each two-dimensional CT slice, after it is divided into multiple sub-block matrices, the matrix transformation result needs to be extracted through a preset image algorithm for each sub-block matrix of the two-dimensional CT slice.

[0043] In this embodiment, the preset image algorithm is as follows:

[0044] ⅰ. Perform a logarithmic transformation on the HU values of the pixel points at each position element in each sub-block matrix.

[0045] Specifically, for the sub-block matrix at the position (n1, n2) of the t-th two-dimensional CT slice of the first CT image, perform a logarithmic transformation on the HU values of the pixel points at each position in the sub-block matrix, so as to obtain the sub-block matrix at the position (n1, n2) of the t-th two-dimensional CT slice ; where 1 ≤ n1, n2 ≤ n.

[0046] ii. For the sub-block matrix in step i , define its binary indicator parameter .

[0047] Specifically, when the sub-block matrix contains the center point of the target nodule, otherwise, .

[0048] iii. Based on hypothesis testing, calculate the matrix transformation results of each sub-block matrix.

[0049] In this embodiment, according to the binary indicator parameters of each sub-block matrix in step ii, for each two-dimensional CT slice of the first CT image, assume that the sum of the binary indicator parameters of all sub-block matrices is , that is, there is only and only one target nodule in each two-dimensional CT slice. And, assume that for each two-dimensional CT slice, the center point of the target nodule appears in the sub-block matrix at the position, that is, the sub-block position where the nodule appears is independent of the CT slice number t.

[0050] Based on this, for any , define , 1 ≤ ≤ n; thus calculate , as the matrix transformation result of the sub-block matrix at the position (n1, n2) of the t-th CT slice. Where, " " and " " are matrix element-by-element calculation factors.

[0051] S4: Calculate the nodule progression display factor corresponding to each sub-block matrix.

[0052] In this embodiment, after calculating the matrix transformation results corresponding to each sub-block matrix of each two-dimensional CT slice of the first CT image through the above steps S1 to S3. Perform the same operations and calculations on the follow-up second CT scan image of the same patient after a period of time, and obtain the matrix transformation result of the sub-block matrix at the position (n1, n2) of the t-th two-dimensional CT slice of the second CT scan image , , Based on the above results, calculate the nodule progression factor of the sub-block matrix at the (n1, n2) position of the t-th two-dimensional CT slice. :

[0053]

[0054] where is a pre-set hyperparameter; indicates that the nodule has an increasing trend on the t-th CT slice; indicates that the nodule has a decreasing trend on the t-th CT slice; indicates that there is no obvious change in the nodule on the t-th CT slice.

[0055] S5: Obtain the nodule progression detection quantity for each two-dimensional CT slice.

[0056] Specifically, since each two-dimensional CT slice is composed of n 2 sub-block matrices, and each sub-block matrix has its corresponding nodule progression factor , therefore, for the same two-dimensional CT slice, it is necessary to integrate the nodule progression factors of the n2 sub-block matrices to form a more effective and comprehensive nodule progression detection quantity.

[0057] In this embodiment, the method of taking the average value is adopted to integrate the nodule progression factors. Specifically, the integrated nodule progression detection quantity .

[0058] S6: Output the final nodule progression detection result.

[0059] When the nodule progression detection quantity for each two-dimensional CT slice is calculated based on step S5, it is easy to verify that for any , . However, the nodule progression detection quantity still has drawbacks, that is, is a function of the hyperparameter , and it is difficult to select the hyperparameter a priori. To eliminate the influence of the hyperparameter , in this embodiment, a two-dimensional curve of is constructed, .

[0060] Figure 2A As shown in this embodiment, it is a schematic diagram of two-dimensional CT slices at the same physical position before and after the treatment of patient A. Among them, the position marked by the square is the position where the nodule is located, and the diameter of the nodule of patient A has increased by about 50%. Figure 2B As shown in this embodiment, it is the two-dimensional curve schematic diagram constructed for patient A. Similarly,Figure 3A Shown is a schematic diagram of two-dimensional CT slices of the same physical location before and after the treatment of Patient B in this embodiment. Among them, the diameter of the nodule in Patient B has decreased by approximately 50%. Figure 3B Shown is the schematic diagram of a two-dimensional curve constructed for Patient B.

[0061] After constructing the two-dimensional curve, by calculating the area under the curve of the curve, it is denoted as the nodule progression detection result DNPt of the t-th two-dimensional CT slice. Then, based on the nodule progression detection results of all two-dimensional CT slices, the final nodule progression detection result is comprehensively output; among them, .

[0062] It can be understood that in the actual application of the embodiments of the present invention, only the CT images of the same patient scanned twice before and after need to be obtained, and then based on the nodule centroid positions marked by the doctor, the value of the final nodule progression detection result can be calculated, and then a hypothesis test can be performed to draw a conclusion on whether the nodule has progressed.

[0063] To verify the effectiveness of the lung nodule auxiliary detection method provided by the embodiments of the present invention, the inventors verified this technical solution through large-scale simulation experiments and real clinical data respectively.

[0064] I. Verification by Simulation Experiment

[0065] First is the design of the simulation experiment. Select the CT image of a certain patient when the nodule was first discovered, denoted as , and then add Gaussian random perturbations with a mean of and a variance of to all pixel points in the nodule area.

[0066] Randomly select two such samples, denoted as and , take as the baseline CT sample, and as the CT sample after follow-up. Because and are both generated by adding the same random perturbation to the same CT image, they can be considered as a pair of nodule samples without change.

[0067] The above method generates the nodule samples before and after from the perspective of nodule density change. Similarly, we can also generate invariant samples by changing the diameter of the nodule. Therefore, in order to generate invariant samples, we designed different combinations based on , and the nodule diameter as three parameters to generate nodule invariant samples. In this embodiment, takes values of , takes values of , the change rate of the nodule diameter considered for value increase is 10%, 20% or 50%, and for value decrease is 10%, 20% or 50%.

[0068] Based on the above simulation experiments, 100,000 pairs of non-changing nodule samples were generated. Then, the Bootstrap method was used to calculate the empirical distribution of the nodule progression detection results, and a 95% confidence interval was given to obtain the sampling distribution of the test statistic under the null hypothesis (nodules do not change). Based on the same method above, 20,000 pairs of test samples were generated, and the proportion of progressing nodules and non-progressing nodules in this sample is approximately 1:1. The generation parameters of the progressing nodule samples in the test samples are a set of parameters outside the specified ranges of the above three parameters.

[0069] For the convenience of subsequent analysis, the above nodule sample pairs were further divided into four groups of subsamples according to the original nodule diameter size, namely less than or equal to 5mm, 5 - 10mm, 10 - 15mm, and 15 - 30mm. Next, denote { , } as the 95% quantile values of DNP obtained under the full sample and the four subsamples, which are 0.087, 0.002, 0.009, 0.033, 0.134 respectively. Compare the DNP of the full sample and each group of test nodule subsamples with the corresponding , where . If DNP > , reject the null hypothesis and consider that the nodule has progressed; otherwise, predict that the nodule has not progressed. Since the conclusion of this study is to judge whether the nodule has progressed, which belongs to a binary classification problem, the subsequent experimental results will all be evaluated using the evaluation indicators of the classification model.

[0070] Figure 4 The confusion matrix under each diameter grouping is shown Figure 5 is the evaluation indicator value. Considering the confusion matrix and the evaluation indicators comprehensively, the sensitivity of the smallest diameter grouping is the lowest. Only 75.1% of the test samples that are actually progressing are successfully detected. This value increases as the diameter of the test samples increases, and the sensitivity of the largest diameter grouping reaches 99.9%.

[0071] Based on the settings of this simulation experiment, the final results show that for nodules with a diameter greater than 15 mm in the test nodule samples, if they progress, the hypothesis testing method proposed by the present invention can give a correct prediction with a probability of 99.9% without measuring the nodule diameter and volume. The above results are consistent with the actual situation, that is, it is difficult to accurately judge the progress of small nodules based on two consecutive CT scans, while the progress of large nodules is more obvious and easier to judge.

[0072] In addition, from the perspective of the full sample of test nodules, the sensitivity, specificity, positive predictive value (PPV for short), and negative predictive value (NPV for short) are 85.8%, 93.4%, 92.5%, and 87.4% respectively. When the PPV is 92.5%, the sensitivity of the proposed grouped hypothesis testing method can reach 85.8%, which means that the proposed method makes a correct prediction for 85.8% of the progressing samples at the cost of making 7.5% incorrect progressing predictions.

[0073] II. Verification with real clinical data

[0074] To verify the effectiveness of the lung nodule auxiliary detection method provided in the embodiments of the present invention on a real clinical data set, 436 nodule pair samples were collected from the thoracic surgery department of a certain domestic tertiary hospital. These 436 nodule pair samples came from 364 sets of CT pairs of 342 patients. Further, the progress information of the 436 nodule pair samples was extracted from the CT reports of these 342 patients and jointly reviewed by 3 experienced thoracic surgery and radiology experts. Finally, objective and reliable nodule pair progress labels, nodule pair center coordinates X, Y, and the Z-axis range covered by the nodules (based on the first CT scan) were determined.

[0075] Among them, there were 185 progressing nodule pairs and 251 non-progressing nodule pairs. Among these 436 nodule pair samples, 100 had an original nodule diameter less than 5 mm, 153 were in the range of (5, 10] mm, 87 were in the range of (10, 15] mm, and 96 were in the range of (15, 30] mm. The experimental process of the real data set was exactly the same as that of the simulation experiment. After calculating the DNP of the real nodule pair samples in each group, it was compared with the 95th percentile of the invariant nodule samples in the corresponding group. If the DNP of the real nodule pair sample was greater than this 95th percentile, it was predicted that the sample had progressed; otherwise, it was predicted that the sample had not progressed.

[0076] Figure 6 Shows the confusion matrices of each group and the full sample. Figure 7is the value of the evaluation index. The PPVs of each group are 65.85%, 73.33%, 88.00% and 93.88% respectively, and the accuracies of each group are 83.00%, 86.27%, 89.66% and 90.63% respectively. The PPV and accuracy generally show an increasing trend with the increase of diameter, and the above results are consistent with the actual situation.

[0077] In addition, from the perspective of the whole-sample test nodules, the sensitivity, specificity, PPV and NPV are 92.97%, 82.87%, 80.00% and 94.12% respectively. When the PPV is 80.00%, the sensitivity of the proposed grouped hypothesis testing method can reach 92.97%, which means that the proposed method makes a correct prediction for 92.97% of the progressing samples at the cost of making 20% false progressing predictions. When the NPV is 94.12%, the specificity of the proposed grouped hypothesis testing method can reach 82.87%, which means that the method proposed by the present invention makes a correct prediction for 82.87% of the non-progressing samples at the cost of making 5.88% false non-progressing predictions.

[0078] Second Embodiment

[0079] As Figure 8 shown, on the basis of the above first embodiment, the second embodiment of the present invention further provides a lung nodule auxiliary detection device for paired CT images, including an acquisition module 1, a segmentation module 2, a matrix transformation module 3, a calculation module 4 and an output module 5.

[0080] Specifically, the acquisition module 1 is an image acquisition unit for obtaining a first CT image of a patient's lung scan and obtaining a two-dimensional CT slice sequence corresponding to the first CT image.

[0081] The segmentation module 2 is connected to the acquisition module 1 to divide each two-dimensional CT slice in the two-dimensional CT slice sequence into a plurality of sub-block matrices.

[0082] The matrix transformation module 3 is connected to the segmentation module 2 and is built-in with a preset image algorithm to extract the matrix transformation result of each sub-block matrix through the preset image algorithm for each sub-block matrix.

[0083] The calculation module 4 is connected to the matrix transformation module 3 to obtain the nodule progression demonstration factor corresponding to the sub-block matrix at this position based on the matrix transformation results of the first CT image and the second CT image for the sub-block matrices with the same slice number and the same position; and, based on the nodule progression demonstration factors of all sub-block matrices of the two-dimensional CT slice under the same slice number, obtain the nodule progression detection amount of the two-dimensional CT slice at this slice number.

[0084] The output module 5 is connected to the calculation module 4 to comprehensively output the final nodule progression detection result based on the nodule progression detection quantity of each two-dimensional CT slice.

[0085] It can be understood that the above acquisition module 1, segmentation module 2, matrix transformation module 3, calculation module 4, and output module 5 are module structures corresponding to the above steps S1 to S6 for implementing the operations of S1 to S6. Here, the specific composition of the modules is not specifically elaborated, and it only needs to meet the above functions.

[0086] Third Embodiment

[0087] As Figure 9 shown, on the basis of the above first embodiment, the third embodiment of the present invention further provides a lung nodule auxiliary detection device for paired CT images. The lung nodule auxiliary detection device includes one or more processors 21 and a memory 22. Among them, the memory 22 is coupled to the processor 21 and is used to store one or more programs. When the one or more programs are executed by the one or more processors 21, the one or more processors 21 implement the method for auxiliary detection of lung nodules in paired CT images as in the above embodiment.

[0088] Among them, the processor 21 is used to control the overall operation of the lung nodule auxiliary detection device to complete all or part of the steps of the above method for auxiliary detection of lung nodules in paired CT images. The processor 21 can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The memory 22 is used to store various types of data to support the operation of the lung nodule auxiliary detection device. These data can include, for example, instructions for any application program or method operating on the lung nodule auxiliary detection device, as well as data related to the application program. The memory 22 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory, etc.

[0089] In an exemplary embodiment, the lung nodule auxiliary detection device may specifically be implemented by a computer chip or an entity, or by a product with certain functions, and is used to execute the above-mentioned method for auxiliary detection of lung nodules in paired CT images, and achieve the same technical effects as the above method. A typical embodiment is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0090] In another exemplary embodiment, the present invention further provides a computer-readable storage medium including program instructions, and when the program instructions are executed by a processor, the steps of the method for auxiliary detection of lung nodules in paired CT images in any one of the above embodiments are implemented. For example, the computer-readable storage medium may be the above-mentioned memory including program instructions, and the above program instructions may be executed by the processor of the lung nodule auxiliary detection device to complete the above-mentioned method for auxiliary detection of lung nodules in paired CT images, and achieve the same technical effects as the above method.

[0091] In summary, the method and device for auxiliary detection of lung nodules in paired CT images provided by the embodiments of the present invention have the following beneficial effects:

[0092] (1) From the perspective of hypothesis testing, in cooperation with a specific image processing algorithm, a unique and delicate detection measure is designed to measure the nodule changes of the patient. Thus, by quantifying the changes of the lesions in the two CT images before and after, it helps doctors observe and analyze the progress of lung nodules more precisely.

[0093] (2) By studying whether there are significant differences in the local nodule images in the two CT scans of the patient before and after, the detection measure is used to answer whether the changes in the lung nodules are statistically significant, making up for the key problem that the existing methods cannot answer whether the detected changes are statistically significant.

[0094] (3) The nodule progression detection measure of each two-dimensional CT slice is represented by preset hyperparameters, and a two-dimensional curve between the nodule progression detection measure and the hyperparameters is constructed, so as to eliminate the influence of the hyperparameters by calculating the area under the curve of the two-dimensional curve, improving the accuracy of the final nodule progression detection result.

[0095] It should be noted that the above-mentioned multiple embodiments are only examples. The technical solutions of each embodiment can be combined, and all are within the protection scope of the present invention.

[0096] In addition, the terms "first" and "second" are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0097] The above has provided a detailed description of the method and apparatus for assisting in the detection of pulmonary nodules in paired CT images according to the present invention. For those of ordinary skill in the art, any obvious changes made without departing from the substantial content of the present invention will constitute an infringement of the patent right of the present invention and will bear corresponding legal responsibilities.

Claims

1. A pulmonary nodule auxiliary detection device for paired CT images, characterized in that include: An acquisition module, used to acquire a first CT image and a second CT image of a patient's lung scan, and respectively acquire a two-dimensional CT slice sequence corresponding to the first CT image and the second CT image; a segmentation module connected to the acquisition module to divide each two-dimensional CT slice in the two-dimensional CT slice sequence into a plurality of sub-block matrices of the same size and number; Specifically, it includes: determining the number n to be divided based on the size of the target pulmonary nodule; dividing the two-dimensional CT slice into sub-block matrices; wherein each of the sub-block matrices is Matrix, m is the size of the sub-block matrix; n is a positive integer, and its value is related to the size of the nodule; A matrix transformation module is connected to the segmentation module and has a preset image algorithm built in, so as to extract the matrix transformation result of each sub-block matrix through the preset image algorithm for each sub-block matrix; wherein the HU value of the pixel point of each position element in the (n1, n2)th sub-block matrix of the t-th two-dimensional CT slice of the first CT image is logarithmically transformed to obtain the (n1, n2)th sub-block matrix of the t-th two-dimensional CT slice. ; For the (n1, n2)th sub-block matrix of the t-th two-dimensional CT slice , define its binary characteristic parameter , where when the sub-block matrix When the target nodule center point is included, ,otherwise, For the t-th two-dimensional CT slice of the first CT image, it is assumed that the sum of the binary characteristic parameters of all sub-block matrices is 1, and the center point of the target nodule appears at In the sub-block matrix of the position, define , 1≤ ≤n; calculation , as the matrix transformation result of the (n1, n2)th sub-block matrix in the tth CT slice; where 1≤n1, n2≤n, and t is a positive integer; A calculation module is connected to the matrix transformation module to obtain the nodule display factor corresponding to the sub-block matrix at the same position for the sub-block matrix with the same number of slices and the same position based on the matrix transformation result of the first CT image and the matrix transformation result of the second CT image; wherein, for the (n1, n2)th sub-block matrix in the tth CT slice, the matrix transformation results of the first CT image are respectively obtained. And the matrix transformation result of the second CT image ; Calculate the nodule display factor of the (n1, n2)th sub-block matrix in the tth CT slice : ,in, is a pre-set hyperparameter and 0< <1; It means that the nodule has a tendency to increase in size on the tth CT slice; It means that the nodule has a tendency to shrink on the tth CT slice; It means that the nodule has no clear changes on the tth CT slice; The output module is connected to the calculation module to obtain the nodule progression detection amount of the two-dimensional CT slice under the same number of slices based on the nodule progression revelation factors of all sub-block matrices of the two-dimensional CT slice; wherein the nodule progression revelation factor of each sub-block matrix in the t-th two-dimensional CT slice is obtained; based on the nodule progression revelation factors of all sub-block matrices in the t-th two-dimensional CT slice, an average value is calculated. As the nodule progression detection quantity of the t-th two-dimensional CT slice; , for any , ; and, based on the nodule progression detection amount of each of the two-dimensional CT slices, the final nodule progression detection result is comprehensively output; wherein, the nodule progression detection amount based on the t-th two-dimensional CT slice And hyperparameters , build Two-dimensional curve of The area under the curve of the two-dimensional curve is recorded as the nodule progression detection result of the t-th two-dimensional CT slice; based on the nodule progression detection results of all two-dimensional CT slices, the final nodule progression detection result is output.

2. The pulmonary nodule auxiliary detection device according to claim 1, characterized in that: The two-dimensional CT slice sequence is a continuous CT slice sequence that only contains the target lung nodule; wherein, the two-dimensional CT slice sequence is obtained through the following steps: first, a lung CT image of the patient is obtained through tomography technology; then, the lung CT image is reviewed and annotated, and continuous two-dimensional CT slices containing the target lung nodule are selected to form a sequence.

3. The pulmonary nodule auxiliary detection device according to claim 1, characterized in that The (n1, n2)th sub-block matrix of the t-th 2D CT slice , generated by: Using a three-dimensional tensor The first CT image is defined, wherein represents the t-th two-dimensional CT slice, T is the total number of CT slices, p represents the dimension of the two-dimensional CT; R represents the real number field; The tth CT slice of the first CT image middle The logarithmic transformation of the HU value of the pixel at the position is recorded as ; The tth CT slice Divide into sub-block matrices, recorded as ,in for matrix, ; Defining an index collection , to re- Recorded as .

Images