Title of invention: attenuation correction of myocardial perfusion single-photon emission computed tomography image data

Abstract

The present invention relates to a method and an apparatus for attenuation correction of myocardial perfusion single-photon emission computed tomography (MPS) image data, the method comprising the steps of: receiving multiple pieces of MPS image data; generating a data set including a first set and a second set by using the MPS image data; applying multiple pieces of image data included in the data set to a deep learning algorithm using multiple loss functions so as to train the deep learning algorithm; generating attenuation-corrected (generated attenuation-corrected (GEN AC )) image data by applying the multiple pieces of image data and the MPS image data to the trained deep learning algorithm; and evaluating the clinical applicability of the generated attenuation-corrected image data. The present invention can also be applied to other embodiments.

Description

Title of the invention: Attenuation correction method and device for myocardial perfusion single-photon tomography image data Technical field [1] The present invention relates to a method and device for attenuation correction of myocardial perfusion single-photon tomography image data. [2] The present invention was derived from research conducted as part of the Basic Research Project in Science and Engineering (Basic Research) of the National Research Foundation of Korea. [3] [Task unique 1711190709 [4] [Task number 2022R1F1A1076500 [5] [Ministry Name] Ministry of Science and ICT [6] [Project Management (Professional) Institution Name] National Research Foundation of Korea [7] [Research Project Name] Basic Research Project in Science and Engineering (Basic Research) [8] [Research Project Name] Development of Deep Learning-Based Myocardial Perfusion Imaging Attenuation Correction Method [9] [Contribution rate] 1 / 1 [1 [Name of the project performing organization] Ajou University

[0011] [Research Period] 2022.06.01~2025.02.28 Background Technology

[0012] Myocardial perfusion single photon emission computed tomography (MPS) imaging is a diagnostic tool for coronary artery disease and is used for coronary artery disease risk stratification. A major concern when performing MPS imaging is soft tissue attenuation artifact due to adjacent organs. Attenuation artifact reduces the specificity of MPS imaging by creating false-positive perfusion defects, which may lead to unnecessary additional testing.

[0013] Therefore, recently, hybrid SPECT / CT examinations that confirm anatomical cross-sectional information of the left ventricle through CT examination and obtain functional cross-sectional information of the left ventricle in a three-dimensional image form through MPS examination are becoming popular, but there is a problem that the equipment for SPECT / CT examination is expensive. In addition, the problem that the patient must be exposed to radiation for CT examination continues to occur. Detailed Description of the Invention Technical Problem [1] Embodiments of the present invention for solving these conventional problems provide a method and device for attenuation correction of myocardial perfusion single-photon tomography image data that can perform deep learning-based attenuation correction on myocardial perfusion single-photon tomography image data acquired using thallium chloride (Tl-201; thallium chloride). It is a means of solving problems.

[0015] A method for attenuation correction of myocardial perfusion single-photon tomography (MPS) image data according to an embodiment of the present invention comprises the steps of: receiving a plurality of myocardial perfusion single-photon tomography (MPS) image data; generating a dataset including a first set and a second set using the MPS image data; applying the plurality of image data included in the dataset to a deep learning algorithm using a plurality of loss functions to learn; and applying the plurality of image data to the learned deep learning algorithm to perform attenuation correction (GEN). AC \ It is characterized by including a step of generating attenuation-corrected image data and a step of evaluating the clinical applicability of the generated attenuation-corrected image data. [1] Also, the step of receiving MPS image data is characterized in that it is a step of receiving the MPS image data acquired in a stress environment and a rest environment. [1] Further, the step of generating the data set is characterized by further comprising the step of receiving computer tomography (CT) image data obtained in the stress environment and the rest environment.

[0018] Further, the step of generating the data set is characterized by generating, as a first set included in the data set, image data for short-axis, horizontal, vertical, and oblique of the gray scale for each item of stress non-attenuation correction (NAC), stress CT-based attenuation correction, rest non-attenuation correction, and rest CT-based attenuation correction. [1] Further, the step of generating the data set is the stress NAC, the stress CT AC , the rest NAC and rest C7 xc Characterized by generating short-axis color image data for the data set as a second set included in the data set. [2] Further, the step of applying and training to a deep learning algorithm using a plurality of loss functions includes applying the NAC image data including the stress NAC and the rest NAC and the stress CTac And the rest 仁~. Included Characterized by including the step of applying the image data to the modified U-Net algorithm using the SSIM (structural similarity index) loss function and the MSE (mean squared error) loss function for training.

[0021] In addition, the step of evaluating the clinical applicability of the generated attenuation-corrected image data is characterized by using at least one of a quantitative evaluation of the modified U-Net algorithm for which the learning has been completed and a performance evaluation using a polar map.

[0022] In addition, the attenuation correction device for myocardial perfusion single photon emission computed tomography (SPECT) image data according to an embodiment of the present invention receives a plurality of myocardial perfusion SPECT (MPS) image data through a communication unit, generates a data set including a first set and a second set by using the MPS image data, applies the plurality of image data included in the data set to a deep learning algorithm using a plurality of loss functions to perform learning, and evaluates the clinical applicability of the attenuation-corrected (GEN AC \ generated attenuation-corrected) image data generated by applying the plurality of image data to the learned deep learning algorithm. The device is characterized by including a control unit.

[0023] In addition, the MPS image data is characterized by being obtained in a stress environment and a resting environment. [2쉬 In addition, the control unit is characterized by generating the data set by using computer tomography (CT) image data obtained in the stress environment and the resting environment.

[0025] In addition, the data set includes a first set including image data for each item of stress non-attenuation correction (NAC; non-attenuation-corrected), stress CT-based attenuation correction (仁7丄0; CT-based attenuation-corrected), resting non-attenuation correction, and resting CT-based attenuation correction for short-axis, horizontal, vertical, and oblique of the gray scale.

[0026] In addition, the data set includes a second set including the stress NAC, the stress renin, and the rest and is characterized by including the second set. [2] Further, the control unit applies the NAC image data including the stress NAC and the rest NAC and the stress CT AC and the CT AC image data including the rest CT to a modified U-Net algorithm using an SSIM (structural similarity index) loss function and an MSE (mean squared error) loss function.

[0028] Further, the control unit evaluates the clinical applicability of the attenuation correction image data by using at least one of a quantitative evaluation of the modified U-Net algorithm for which the learning has been completed and a performance evaluation using a polar map. Effects of the Invention [2] As described above, the attenuation correction method and apparatus for myocardial perfusion single photon emission computed tomography image data according to the present invention perform deep learning-based attenuation correction on a myocardial perfusion single photon emission computed tomography image obtained by using thallium chloride (Tl-201; thallium chloride), thereby eliminating the occurrence of unnecessary additional examinations, not requiring expensive equipment for SPECT / CT examinations, and having an effect of minimizing the situation in which a patient is exposed to radiation for CT imaging. Brief Description of the Drawings [3] FIG. 1 is a diagram showing the main configuration of an attenuation correction apparatus for myocardial perfusion single photon emission computed tomography image data according to an embodiment of the present invention.

[0031] FIG. 2 is an exemplary screen diagram showing a data set according to an embodiment of the present invention.

[0032] FIG. 3 is a screen example showing a comparison between image data included in a dataset according to an embodiment of the present invention and attenuation-corrected image data generated by a deep learning algorithm.

[0033] FIG. 4 is an example screen showing a screen displaying myocardial perfusion in a rest and stress dataset on a tomographic image and a polar map according to an embodiment of the present invention. [3.5. CT according to an embodiment of the present invention AC , GEA Heart c , is an example screen graph showing the difference in perfusion between polar map segments between NAC image data.

[0035] Figure 6 is a GEN show confirmed based on a rest and stress dataset according to an embodiment of the present invention. c This is an example screen showing the visual evaluation scores obtained by visually evaluating video data. [3 This drawing 7 is a flow chart for explaining a method for attenuation correction of myocardial perfusion single-photon tomography image data according to an embodiment of the present invention. Best mode for carrying out the invention [3. The following detailed description of preferred embodiments of the present invention will now be given with reference to the accompanying drawings. The detailed description set forth below with reference to the accompanying drawings is intended to explain exemplary embodiments of the present invention and is not intended to represent the only embodiments in which the present invention may be practiced. In order to clearly describe the present invention in the drawings, parts that are not related to the description may be omitted, and the same reference numerals may be used throughout the specification for identical or similar components.

[0038] FIG. 1 is a drawing showing the main configuration of an attenuation correction device for myocardial perfusion single-photon tomography image data according to an embodiment of the present invention. [3 Referring to FIG. 1, an electronic device (100) according to the present invention may include a communication unit (110), an input unit (120), a display unit (130), a memory (140), and a control unit (150).

[0040] The communication unit (110) can receive myocardial perfusion single photon emission computed tomography (MPS) image data and computed tomography scan (CT) image data of the subject through communication with an external device (not shown). To this end, the communication unit (110) can receive 5G (5 th It can perform wireless communications such as LTE (long term evolution), LTE-A (long term evolution-advanced), and Wi-Fi (wireless fidelity), and can perform wired communications using cables.

[0041] The input unit (120) generates input data in response to input from a user of the electronic device (100). To this end, the input unit (120) may include input devices such as a keyboard, a mouse, a keypad, a dome switch, a touch panel, a touch key, and a button.

[0042] The display unit (130) outputs output data according to the operation of the electronic device (100). To this end, the display unit (130) may include a display device such as a liquid crystal display (LCD), a light emitting diode (LED) display, or an organic light emitting diode (OLED) display. In addition, The display unit (130) can be combined with the input unit (120) and implemented in the form of a touch screen.

[0043] The memory (140) stores operation programs of the electronic device (100). In particular, the memory (140) can store MPS image data and CT image data received from the communication unit (110) and store a plurality of image data included in a data set set by the control unit (150). In addition, the memory (140) can store a deep learning algorithm, such as a modified U-Net algorithm including a loss function. [The 4-hour control unit (150) receives myocardial perfusion single photon emission computed tomography (MPS) image data and CT image data through the communication unit (110). At this time, the MPS image data may include image data acquired in a stress environment of the test subject, for example, the test subject, and image data acquired in a rest environment. In order to establish a stress environment when acquiring the MPS image data, the tester injects 140 ng / kg / min of adenosine into the test subject for 6 minutes, and intravenously injects 11 IMBp of Tl-2()1 during the adenosine injection. Tl-2()1 has a higher attenuation phenomenon than other injections because its radiation energy is lower.

[0045] In this way, the investigator acquires MPS image data in a stress environment immediately after inducing pharmacological stress in the subject, and acquires MPS image data in a resting environment 4 hours later.

[0046] The control unit (150) generates a data set including a first set and a second set by user input using MPS image data. This will be described in more detail using FIG. 2 below. The control unit (150) can receive CT image data acquired from a CT imaging device to generate the data set.

[0047] The control unit (150) applies at least a part of the dataset to a modified U-Net algorithm, which is a deep learning algorithm using an SSIM (structural similarity index) loss function and an MSE (mean squared error) loss function, to reduce the inconsistency in the image distribution between the stress non-attenuation-corrected (NAC) image data and the stress CT-based attenuation-corrected AC CT-based attenuation-corrected) image data in the dataset, and trains the modified U-Net algorithm. At this time, the NAC image data may be MPS image data, and the C ◎。image data may be image data obtained by attenuation-correcting CT image data based on MPS image data.

[0048] Then, the control unit (150) applies the result of the training to another part of the dataset, evaluates it, and finally modifies the modified U-Net algorithm. At this time, at least a part of the dataset means training data extracted from the dataset, and the other part means verification data.

[0049] The control unit (150) extracts a plurality of image data from the image data included in the dataset, applies the extracted plurality of image data to the modified U-Net algorithm, and generates new attenuation-corrected (c구互分心仁; generated attenuation-corrected) image data based on the training result obtained by the application. [5This control unit (150) is GEN AC performs a clinical applicability evaluation to evaluate whether the image data is applicable to clinical applications. At this time, the clinical applicability evaluation may include a quantitative evaluation of the modified U-Net algorithm, a performance evaluation using a polar map, and a visual transverse diaphragm attenuation correction effect evaluation.​

[0051] The quantitative evaluation of the modified U-Net algorithm is to evaluate the attenuation correction performance of the modified U-Net algorithm by comparing it with other deep learning algorithms. To quantitatively evaluate the performance of the modified U-Net, the control unit (150) can compare the results of applying the first set to the U-Net, MMTrans, Reg-GAN, Palette, and the modified U-Net respectively. In addition, the control unit (150) can evaluate the reproducibility of the modified U-Net algorithm through the evaluation of horizontal, vertical, and oblique image data.

[0052] For the performance evaluation using the polar map, the user can randomly select the image data for a plurality of subjects from the short-axis image data (NAC and CT AC ) and the MPS image data that have been learned and generated by the modified U-Net to generate a polar map for each. c At this time, the polar map can calculate the average intensity of the MPS image data for each segment and display the calculated average intensity as a percentage of the maximum segment. Through this, the user can compare the perfusion of the inferior wall segment in the GEA heart

[0053] image data and the <三7*三 image data, and compare the perfusion of the inferior wall segment in the GEN c image data and the NAC image data. The statistical significance of the perfusion difference in the inferior wall segment can be set to 0.05. AC The evaluation of the visual septal attenuation correction effect is that nuclear medicine doctors use the stress NAC stress data set and A method of visually checking a resting dataset included and selecting and evaluating image data for a subject with diaphragmatic attenuation. At this time, nuclear medicine doctors can select image data for a subject with diaphragmatic attenuation to give clinical significance during visual evaluation.

[0055] More specifically, nuclear medicine doctors can GEN AC image data to CT AC Compare the image data with NAC image data and input scores from 1 to 5 as the evaluation result. If it is indistinguishable from the NAC image data, it is 1 point, NAC If it is similar to the image data, it is 2 points, if it is in the middle between the NAC image data and the 《image data, it is 3 points, if it is similar to the (three 7丄0 image data, it is 4 points, and if it is indistinguishable from the CT AC image data, it can be evaluated as 5 points. The evaluation result can be displayed as the mean + standard deviation and the mean - standard deviation. [5These three methods of performing the clinical applicability evaluation as described above can be more specifically explained using FIGS. 3 to 6 below. [5FIG. 2 is a screen example diagram showing a dataset according to an embodiment of the present invention.

[0058] Referring to FIG. 2, it shows a dataset for training a deep learning algorithm by applying it to the deep learning algorithm to generate attenuation-corrected image data. The dataset may include a first set and a second set, and stress non-attenuation correction (NAC; non-attenuation-corrected), stress CT-based attenuation correction (仁 / 丄 c; A first set of grayscale (DIC0M format) short-axis, horizontal, vertical and oblique image data for each item of CT-based attenuation-corrected, rest unattenuated correction and rest CT-based attenuation-corrected were generated, and stress NAC, stress A second set of color image (Short-axis (capture)) data (Jpeg format) can be generated.

[0059] FIG. 3 is a screen example showing a comparison between image data included in a dataset according to an embodiment of the present invention and attenuation-corrected image data generated by a deep learning algorithm.

[0060] Referring to Figure 3, the first and second sets include Data, NAC image data, and GZTA heart image data generated using a deep learning algorithm are shown. As can be seen in Fig. 3, the first set is grayscale digital image data and the second set is short-cut color image data. At this time, the arrows shown in Fig. 3 are marks for comparing the difference in hypoperfusion.

[0061] In addition, as shown in Fig. 3, both the first and second sets of GEA heart c image data are CT AC You can see that it has similar quality to video data.

[0062] FIG. 4 is an example screen showing a screen displaying myocardial perfusion in a rest and stress dataset on a tomographic image and a polar map according to an embodiment of the present invention.

[0063] Referring to Figure 4, the resting dataset and stress dataset are based on the polar map. It shows the comparison results. As a result of the comparison, it can be confirmed that myocardial perfusion in the GEA heart c image data is similar to that in the image data based on the rest dataset and the stress dataset, and it can be confirmed that perfusion reduction has occurred in the inferior wall segment in the NAC image data. [6 Figure 5 is a CT according to an embodiment of the present invention AC , between GEA heart., NAC image data It is a screen example diagram showing the perfusion difference by polar map segment as a graph.

[0065] Referring to Figure 5, the y-axis represents the average intensity of myocardial perfusion for each segment and is expressed as a percentage of the maximum value. At this time, Figure 5 may be a table in which the control unit (150) statistically evaluates myocardial perfusion using a polar map for a randomly selected patient by a nuclear medicine doctor. [6 Thus, it can be confirmed that the myocardial perfusion in segments 4 and 10 of the G4V*; image data is higher than that of the NAC image data in both the rest dataset and the stress dataset. The myocardial perfusion in segment 10 of the 04入心0 image data is CTa ^ It can be confirmed that it is not different from the data. In addition, the myocardial perfusion in segment 4 is than the image data It can be confirmed that it is slightly lower in the GET heart. image data, but the difference is not statistically significant. [6 Figure 6 is a screen example diagram showing the visual evaluation score obtained by visually evaluating the <%시 image data confirmed based on the rest and stress datasets according to an embodiment of the present invention.

[0068] Referring to Figure 6, GEN calculated based on the evaluation result input from the outside ACThe average score of the visual evaluation for the image data was found to be 4.45+0.24 for the stress dataset and 4.38±0.49 for the rest dataset.

[0069] In the stress dataset, all 34 subjects had an average score of 4 or more, while in the rest dataset, 3 subjects had an average score of less than 4. All 3 nuclear medicine physicians had a GEN score of less than 4 in the stress dataset. AC It can be seen that the image data was given a score of 4 or 5. However, the nuclear medicine physician provided a score of 4 or 5 in the resting data set. G£Wac , it can be confirmed that the video data also provided scores less than 4.

[0070] All three nuclear medicine physicians provided a score of 4 or 5 to the shared input image data of the stress dataset, confirming that the attenuation correction of the myocardial perfusion single-photon computed tomography image data according to the present invention was properly performed.

[0071] In this way, the present invention has the effect of eliminating unnecessary additional examinations of subjects by replacing CT image data with a deep learning technique, minimizing the cost of purchasing expensive equipment for SPECT / CT examinations, and minimizing the situation in which subjects are exposed to radiation for CT scanning.

[0072] FIG. 7 is a flowchart for explaining a method for attenuation correction of myocardial perfusion single-photon tomography image data according to an embodiment of the present invention.

[0073] Referring to Fig. 7, in step 7()1, the control unit (150) communicates with an external A device (not shown), for example, receives myocardial perfusion single photon emission computed tomography (MPS) image data obtained from a SPECT (single photon emission computed tomography) device. At this time, the MPS image data may include image data obtained in a stress environment of a test subject, for example, a patient undergoing a procedure, and image data obtained in a rest environment. To construct a stress environment when acquiring MPS image data, the examiner injects 140 ng / kg / min of adenosine into the test subject for 6 minutes, and injects Tl-201 of 111 MBq intravenously while injecting adenosine. In this way, the examiner acquires MPS image data in a stress environment immediately after inducing pharmacological stress in the test subject, and acquires MPS image data in a rest environment after 4 hours have elapsed.

[0074] In step 703, the control unit (150) generates a data set including a first set and a second set using the MPS image data as shown in FIG. 2 above according to the input of the input unit (120). More specifically, the first set may include short-axis, horizontal, vertical, and oblique image data that are grayscale (DICOM format) digital images. And the first set may include stress non-attenuation correction (NAC), stress CT-based attenuation correction, rest non-attenuation correction, and rest CT-based attenuation correction image data. The second set includes stress NAC, stress CT show c , rest NAC, rest CC cIt may include short color image data (in Jpeg format). In addition, the control unit (150) may receive CT image data acquired from a CT imaging device to generate a data set.

[0075] In step 705, the control unit (150) applies the image data included in the dataset to the deep learning algorithm to perform learning of the deep learning algorithm. More specifically, the control unit (150) applies at least a part of the dataset to the modified U-Net algorithm, which is a deep learning algorithm using the structural similarity index (SSIM) loss function and the mean squared error (MSE) loss function to reduce the mismatch of the image distribution between the NAC image data included in the dataset and the In-Gi image data, to learn the modified U-Net algorithm. Then, the control unit (150) applies and evaluates the result according to the learning to another part of the dataset to finally modify the modified U-Net algorithm. At this time, at least a part of the dataset means learning data extracted from the dataset, and the other part means verification data.

[0076] In step 707, the control unit (150) generates a new attenuation correction (GEN) based on the learning results from step 705. A More specifically, the modified U-Net algorithm has an encoder-decoder architecture and uses skip connections to transfer high-level feature maps given in the encoder layer to the decoder layer to generate the original image, for example, The structure of the MPS image data acquired in step 7()1 can be preserved. [The 7th revised U-Net consists of a contextual pathway and a localization pathway, and each pathway consists of four levels of convolutional blocks. In the contextual pathway, feature map extraction is performed by increasing the stride of the convolutional layer, and the spatial resolution of the image data is reduced by half at each stage while extracting overall contextual information. The localization pathway connects the feature map extracted in the contextual pathway to the original image. In the localization pathway, the spatial resolution of the image data is increased by half at each stage.

[0078] The modified U-Net is a different algorithm from the original U-Net in that it integrates dropout, group normalization, and leaky rectified linear unit (leaky ReLU) activation. The modified U-Net is more powerful and effective for various image segmentation tasks.

[0079] This modified U-Net is intended to minimize the total loss between SSIM and MSE, and is used with the generated GEAheart c-image data. To maintain the similarity between image data, SSIM loss function and MSE loss function are applied to the conversion method between image data.

[0080] More specifically, the control unit (150) can divide the data set into training data and verification data, segment the first set and the second set among the data sets included in the training data, and train the modified U-Net using each segmentation. The control unit (150) performs resizing, rotation, and flipping of the video data included in the data set, which is the training data, to improve the learning result, and can use MAE (mean absolute error), SSIM, and PSNR (peak signal-to-noise ratio) metrics to quantitatively evaluate the learning performance. In addition, the control unit (150) can derive a mathematical formula for calculating a final loss value that can obtain an optimal result as shown in Equation 5 by combining the loss functions of SSIM and MSE based on the following Equations 1 to 4.

[0081] [Equation 1]

[0082] At this time, and GEN AC Video data and CT AC means the average value of all pixel values of the video data, and c heart and ◎y are respectively GEN AC Video data and mean the variance of the CTAC video data. [8 is [Equation 2] ! n MAE = — 乞[少厂니 n «=i [8 쉬 At this time, n means the total number of pixel values, and i means the index that repeats each pixel value of the MPS video data obtained in the 7()1 step.

[0085] [Equation

[0086] [Equation 쉬

[0087] At this time, I is GEN AC video data and CT AC represents the maximum pixel value of the video data.

[0088] [Equation 5] 刀—= :( 1 - SSIM + MSE)

[0089] Subsequently, at step 709, the control unit (150) GEN AC performs a clinical applicability evaluation to evaluate whether the video data is applicable to clinical use. At this time, the clinical applicability evaluation may include a quantitative evaluation of the modified U-Net algorithm, a performance evaluation using a polar map, and an evaluation of the effect of visual horizontal diaphragm attenuation correction. [9This quantitative evaluation of the modified U-Net algorithm is to evaluate the attenuation correction performance of the modified U-Net algorithm by comparing it with other deep learning algorithms. More specifically, the control unit (150) compared the results of applying the first set to U-Net, MMTrans, Reg-GAN, Palette, and the modified U-Net respectively to quantitatively evaluate the performance of the modified U-Net algorithm. This is shown in Table 1 below.

[0091] [Table 1]

[0092] As such, it can be confirmed that other algorithms except Reg-GAN show good performance in video data conversion. In addition, it can be confirmed that the modified U-Net achieved the highest performance. [9In addition, as shown in Table 2 below, it can be confirmed that the modified U-Net algorithm also shows high performance in horizontal, vertical, and oblique video data for evaluating the reproducibility of the modified U-Net algorithm.

[0094] [Table 2]

[0095] Through this, it can be confirmed that the ranges of MAE, SSIM, and PSNR are 0.003 - 0.004, 0.988 - 0.994, and 33.658 - 38.669 respectively, and it can be confirmed that the gradient video data achieves the highest performance in both the rest dataset and the stress dataset. At this time, the rest dataset can include rest NAC video data and the stress dataset can include stress NAC video data and stress C~ c video data. In addition, as shown in FIG. 3, the GEA heart c video data generated using the deep learning algorithm can be confirmed to have a quality similar to that of the C ◎ video data in both the first set and the second set. [9For the performance evaluation using the polar map, the control unit (150) generates a polar map according to the user's input. The control unit (150) short-axis video data (NAC and CT AC ) and the MPS video data in the first set are learned by the modified U-Net to generate the GEA heart c Among the video data, the video data for a plurality of subjects can be randomly selected by the user, and a polar map can be generated for each of the selected video data. [9At this time, the polar map can calculate the average intensity of the MPS video data for each segment and display the calculated average intensity as a percentage of the maximum segment. Through this, the user can compare the perfusion of the inferior wall segment in the <3 five 7 heart 0 video data and the 07丄0 video data, and compare the perfusion of the inferior wall segment in the stone mutual heart. video data and the NAC video data.

[0098] More specifically, in the segment-by-segment analysis, the control unit (150) can G EN AC compare the video data in the polar map with the 仁7 three < video data and the NAC video data. As shown in FIG. 4, GEA Changc Myocardial perfusion in the image data is CT AC Similar to the image data, it can be confirmed that perfusion reduction occurred in the inferior wall segment in the NAC image data. In addition, the control unit (150) can statistically evaluate myocardial perfusion using the polar map for the subject randomly selected by the user. The statistical significance for the perfusion difference in the inferior wall segment can be set to 0.05. [As shown in Fig. 5, 。It can be confirmed that the myocardial perfusion in segments 4 and 10 of the image data is higher than that in the NAC image data in both the rest dataset and the stress dataset. GEA heart. The myocardial perfusion in segment 10 of the image data is in the rest dataset or the stress dataset CTac ^It can be confirmed that it is not different from the upper data. In addition, the myocardial perfusion in segment 4 is仁〜。Lower than the GEA heart c It can be confirmed that it is slightly lower than that in the image data, but the difference is not statistically significant.

[0100] Visual evaluation of the diaphragmatic attenuation correction effect is that nuclear medicine doctors use a stress dataset including stress NAC image data and stress C ◎. image data, and rest NAC image data and rest C〜 c The rest dataset including the image data This is a method of visually checking and selecting the image data for the subject with diaphragmatic attenuation for evaluation. More specifically, nuclear medicine doctors use GEN AC The image data is CT AC ^The upper data and the NAC image data can be compared and scores from 1 to 5 can be input as the evaluation result. If it is indistinguishable from the NAC image data, it is 1 point, if it is similar to the NAC image data, it is 2 points, and if it is similar to the NAC image data and CT ACIf the image data is intermediate, it can be evaluated as 3 points, if it is similar to the image data, it can be evaluated as 4 points, and if it is indistinguishable from the image data, it can be evaluated as 5 points. At this time, nuclear medicine doctors can select image data for subjects with diaphragmatic attenuation to give clinical meaning during visual evaluation.

[0101] More specifically, referring to Figure 6, the GE7V produced the evaluation results evaluated by nuclear medicine doctors. AC The average visual evaluation scores for the image data were found to be 4.45 ± 0.24 for the stress dataset and 4.38 ± 0.49 for the rest dataset.

[0102] In the stress dataset, all 34 subjects had an average score of 4 or more, while in the rest dataset, 3 subjects had an average score of less than 4. All 3 nuclear medicine physicians had a GEN score of less than 4 in the stress dataset. AC It can be seen that the image data provided a score of 4 or 5. However, the nuclear medicine physician can see that the GE7V* of the resting data set; •It can be seen that the image data also provided a score of less than 4.

[0103] In this way, the control unit (150) is GEA-heart c Once the clinical applicability assessment of the image data is completed, the process is terminated.

[10] The embodiments of the present invention disclosed in the specification and drawings are only specific examples presented to easily explain the technical content of the present invention and to help understand the present invention, and are not intended to limit the scope of the present invention. Therefore, the scope of the present invention should be interpreted to include all changes or modified forms derived based on the technical idea of ​​the present invention in addition to the embodiments disclosed herein.

Claims

Scope of claim

1. A step for receiving a plurality of myocardial perfusion single photon scan (MPS) image data; a step for generating a data set including a first set and a second set using the MPS image data; a step for applying the plurality of image data included in the data set to a deep learning algorithm using a plurality of loss functions to learn; and a step for applying the plurality of image data to the learned deep learning algorithm. A method for attenuation correction of myocardial perfusion single-photon tomography image data, comprising: a step of generating image data; and a step of evaluating the clinical applicability of the generated attenuation-corrected image data. ［

2. A method for attenuation correction of myocardial perfusion single-photon tomography image data, characterized in that in claim 1, the step of receiving the MPS image data is a step of receiving the MPS image data acquired in a stress environment and a rest environment. ［ A method for attenuation correction of myocardial perfusion single-photon tomography image data, characterized in that in claim 2, the step of generating the dataset comprises the step of receiving computed tomography (CT) image data acquired in the stress environment and the rest environment; [In claim 3, the step of generating the dataset comprises: stress non-attenuation correction (NAC), stress CT-based attenuation correction (CT c ; CT-based attenuation-corrected), rest attenuation correction ■ and rest CT-based attenuation correction. A method for attenuation correction of myocardial perfusion single-photon tomography image data, characterized in that it is a step of generating image data for short-axis, horizontal, vertical and oblique in grayscale for each item as a first set included in the dataset. ［ Claim 5. The method according to claim 4, wherein the step of generating the data set includes the stress NAC, the stress CT show c , the short color image data for the rest NAC and rest 07丄0 is included in the second of the data set A method for attenuation correction of myocardial perfusion single-photon tomography image data, characterized in that it is a step of generating it as a set. [In claim 4, the step of learning by applying the plurality of loss functions to a deep learning algorithm is as follows: The modified U-Net algorithm using the SSIM (structural similarity index) loss function and the MSE (mean squared error) loss function was applied to the NAC image data including the stressed NAC and the resting NAC and the stressed NAC. A method for attenuation correction of myocardial perfusion single-photon tomography image data, characterized by including a step of learning by applying In770 image data including the above-mentioned In770. [A method for attenuation correction of myocardial perfusion single-photon computed tomography image data, characterized in that in claim 6, the step of evaluating the clinical applicability of the generated attenuation-corrected image data uses at least one of a quantitative evaluation of the modified U-Net algorithm for which learning has been completed and a performance evaluation using a polar map.] ［ Claim 8] A communication unit receiving a plurality of myocardial perfusion single-photon computed tomography (MPS) image data; and generating a data set including a first set and a second set using the MPS image data, applying the plurality of image data included in the data set to a deep learning algorithm using a plurality of loss functions to learn, and applying the plurality of image data to the learned deep learning algorithm to generate the data set. An attenuation correction device for myocardial perfusion single-photon tomography image data, characterized by including a control unit for evaluating the clinical applicability of generated attenuation-corrected image data. ［ An attenuation correction device for myocardial perfusion single-photon tomography image data, characterized in that the MPS image data in claim 8 is acquired in a stress environment and a resting environment. ［

10. The myocardial perfusion single photon emission computed tomography (SPECT) image data attenuation correction apparatus according to claim 9, wherein the control unit generates the data set using computer tomography (CT) image data obtained in the stress environment and the rest environment.

11. The myocardial perfusion single photon emission computed tomography (SPECT) image data attenuation correction apparatus according to claim 10, wherein the data set includes a first set including image data for short-axis, horizontal, vertical, and oblique of grayscale for each item of stress non-attenuation correction (NAC), stress CT-based attenuation correction, rest non-attenuation correction, and rest CT-based attenuation correction. [

12. In claim 11, the dataset comprises: the stress NAC, the stress CT AC . Above rest NAC and rest The myocardial perfusion single photon emission computed tomography (SPECT) image data attenuation correction apparatus, characterized by including a second set including data.

13. The myocardial perfusion single photon emission computed tomography (SPECT) image data attenuation correction apparatus according to claim 12, wherein the control unit The modified U-Net algorithm using the SSIM (structural similarity index) loss function and the MSE (mean squared error) loss function was applied to the NAC image data including the stressed NAC and the resting NAC and the stressed NAC. applies the image data including the rest CT to perform attenuation correction.

14. The myocardial perfusion single photon emission computed tomography (SPECT) image data attenuation correction apparatus according to claim 13, wherein the control unit evaluates the clinical applicability of the attenuation-corrected image data using at least one of quantitative evaluation of the modified U-Net algorithm for which the learning has been completed and performance evaluation using a polar map.

Images