Organ segmentation method and system

Abstract

The present invention provides a method for identifying a liver in a CT image of a patient. The method includes applying a liver model to the CT image. The method further includes extracting an internal liver region and an external liver region from the CT image based on the applied liver model. The method further includes performing a graph cut algorithm on the CT image based on the internal liver region and the external liver region to generate a liver image. Performing the graph cut algorithm on the CT image to generate the liver image may be further based on internal heart and / or kidney regions and external heart and / or kidney regions. The present invention provides a non-transitory computer-readable storage medium encoded with a program.

Description

Technical Field

[0001] The present disclosure relates to systems and methods for identifying organs and, more particularly, to liver segmentation in clinical applications. Background Art

[0002] When planning a treatment protocol, clinicians typically rely on patient data, including X-ray data, computed tomography (CT) scan data, magnetic resonance imaging (MRI) data, or other imaging data that allows the clinician to view the patient's internal anatomy. Clinicians use the patient data to identify target objects of interest and develop strategies for approaching the target objects of interest for a surgical procedure.

[0003] Using CT images as a diagnostic tool has become routine, and CT results are typically a source of information for clinicians regarding the size and location of lesions, tumors, or other similar target objects of interest. CT images are typically obtained by digitally imaging the patient in slices in each of the axial, coronal, and sagittal directions. Clinicians view the CT image data slice by slice in each direction when attempting to identify or locate the target object.

[0004] Liver segmentation in medical images, particularly CT images, is an important requirement in many clinical applications such as liver transplantation, resection, and ablation. Due to the variation in anatomical shape, manual delineation is a time-consuming and challenging task. Therefore, an automated liver segmentation method for extracting the liver anatomy from 3D CT images is needed. Summary of the Invention

[0005] In one aspect of the present disclosure, a method for identifying a liver in a CT image of a patient is provided. The method includes applying a liver model to the CT image. The method further includes extracting an internal liver region and an external liver region from the CT image based on the applied liver model. The method further includes performing a graph cut algorithm on the CT image based on the internal liver region and the external liver region to generate a liver image.

[0006] In another aspect of the present disclosure, the method further includes applying a heart model to the CT image and extracting an internal heart region and an external heart region from the CT image based on the applied heart model. Performing the graph cut algorithm on the CT image to generate the liver image can be further based on the internal heart region and the external heart region.

[0007] In another aspect of the present disclosure, the method includes applying a kidney model to the CT image and extracting an internal kidney region and an external kidney region from the CT image based on the applied kidney model. Performing the graph cut algorithm on the CT image to generate the liver image may be further based on the internal kidney region and the external kidney region.

[0008] In yet another aspect of the present disclosure, extracting the internal liver region from the CT image is further based on the external kidney region and the external heart region.

[0009] In one aspect of the present disclosure, extracting the external liver region from the CT image is further based on the internal kidney region and the internal heart region.

[0010] In another aspect of the present disclosure, the method further includes extracting a body mask from the CT image. Performing the graph cut algorithm on the CT image to generate the liver image may be further based on the body mask.

[0011] In yet another aspect of the present disclosure, the method further includes extracting a cavity mask from the CT image based in part on the body mask. Performing the graph cut algorithm on the CT image to generate the liver image may be further based on the cavity mask.

[0012] In one aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium encoded with a program. When the program is executed by a processor, the processor is caused to perform the steps of any one or more of the methods described herein.

[0013] Any of the above aspects and embodiments of the present disclosure may be combined without departing from the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The objects and features of the systems and methods disclosed herein will become apparent to those of ordinary skill in the art upon reading the description of its various embodiments with reference to the accompanying drawings, in which:

[0015] Figure 1 Identification of the internal and external regions of the liver according to an exemplary embodiment of the present disclosure is shown;

[0016] Figure 2 A liver mask generated according to an exemplary embodiment of the present disclosure is shown;

[0017] Figure 3 is a flowchart showing a method of generating a liver mask according to an embodiment of the present disclosure;

[0018] Figure 4Illustration of the identification of the internal and external regions of the liver and the regions located between these boundaries according to an exemplary embodiment of the present disclosure;

[0019] Figure 5 Shows the identification of the internal and external regions of the liver, kidney, and heart, and the right and left lung lobes according to an exemplary embodiment of the present disclosure;

[0020] Figures 6A to 6D Shows a coronal view of CT images (including enhanced and plain CT images) of the same patient according to an exemplary embodiment of the present disclosure;

[0021] Figure 7A and Figure 7B Shows the body mask and cavity mask in a CT image according to an exemplary embodiment of the present disclosure;

[0022] Figure 8 Is a flowchart showing a method for extracting the abdominal cavity and repositioning the model according to an embodiment of the present disclosure;

[0023] Figure 9A and Figure 9B Is a flowchart showing a method for extracting the internal and / or external regions of the kidney and heart according to an embodiment of the present disclosure;

[0024] Figure 10 Is a flowchart showing a method for initializing liver data according to an embodiment of the present disclosure;

[0025] Figure 11 Is a flowchart showing a method for refining the liver region according to an embodiment of the present disclosure;

[0026] Figure 12 Is a flowchart showing a method for generating a final liver mask according to an embodiment of the present disclosure;

[0027] Figure 13 Is a flowchart showing a method for generating a trained heart, kidney, and / or liver model according to an embodiment of the present disclosure; and

[0028] Figure 14 Is a schematic diagram of a computing device used according to an exemplary embodiment of the present disclosure. Detailed Description

[0029] Although the present disclosure will be described in accordance with specific exemplary embodiments, it will be apparent to those skilled in the art that various modifications, rearrangements, and substitutions can be made without departing from the essence of the present disclosure. The scope of the present disclosure is defined by the claims appended hereto.

[0030] The present disclosure provides an automatic liver segmentation method for extracting liver anatomical structures from 3D CT images. Conventional liver segmentation tools are user-interactive and time-consuming, may require contrast-enhanced CT images, and may not be able to extract the liver from plain CT scans without contrast.

[0031] Segmentation is a processing algorithm commonly applied to medical images that attempts to identify the boundaries of various types of tissue by comparing the value of each data element of the CT image data or the generated 3D reconstruction with a series of thresholds or other similar criteria. The segmentation algorithm groups similar types of tissue (e.g., lungs, airways, lung lobes, nodules, vasculature, liver, ribs, heart, or other key structures) based on the comparison results. Each group can then be processed separately for presentation to the clinician. For example, since the intensity of each pixel in a CT image corresponds to the actual density of the scanned tissue material, segmentation can be used to separate tissue materials with different densities by analyzing the intensity values in the CT image.

[0032] One benefit of segmentation is the ability to present each key structure of a patient's anatomy to the clinician in a visual form with different colors and / or transparencies. This provides the clinician with an easy way to identify different tissue types in the same image. For example, once segmented into groups, the lungs, airways, bones, etc. can each be presented with different colors or different transparency settings that the clinician can adjust.

[0033] The technology disclosed in the present invention provides significant benefits over the prior art. For example, the liver segmentation method according to the present disclosure operates automatically without user interaction. This technology is applicable to both contrast CT images and plain CT images. This technology uses a pre-trained liver model to improve accuracy and acceleration, and also uses pre-trained kidney and heart models to label non-liver regions to improve accuracy.

[0034] Figure 1 The identification of the internal and external regions of the liver is shown, which can be used in a graph-based method (e.g., graph cut algorithm) to obtain Figure 2 the liver mask shown. In Figure 1 Image 100 includes a liver boundary 130, an external liver region boundary 110, and an internal liver region boundary 120. The boundary of the liver boundary 130 lies between the external liver region boundary 110 and the internal liver region boundary 120. The liver is a large organ in the abdomen, but it may be difficult to extract directly visually. The algorithm according to this technology first identifies two liver regions, namely the internal region and the external region, as Figure 1As shown. The internal liver region boundary 120 is completely within the liver boundary 130, and the external liver region boundary 110 includes all of the liver boundary 130. Then, an improved graph cut method will be applied to obtain the final liver mask result. The final mask result includes all of the internal liver region boundary 120 and does not exceed the external liver region boundary 110.

[0035] According to human anatomy, the liver is close to the heart, kidneys, and ribs. The intensities in the CT images of the heart and kidneys can be similar to that of the liver, especially in conventional CT scans. Thus, it may be difficult to separate them without some limitations. In most of the available software for this purpose, user interaction is required to specify the initial liver region and exclude the connected non-liver organs. In the present technique, pre-trained models of the liver, heart, and kidneys are used to localize the initial internal and external regions of the liver. Then, based on the detected internal and external regions of the liver, heart, and kidneys, the liver result can be calculated by a graph cut method to produce Figure 2 the liver image 200 as shown.

[0036] Figure 3 FIG. 300 is a flow chart outlining the process, which may include some or all of the following operations. A CT image is obtained in operation 310, and in operation 320, the input CT image is smoothed slice by slice using Gaussian smoothing or another suitable smoothing technique. In operation 325, the smoothed CT image is resampled to a spacing of 4 mm × 4 mm × 4 mm or any other suitable size along the x-axis, y-axis, and z-axis. Then, in operation 330, a patient body mask is extracted to limit the search region for the liver. A cavity mask is used in operation 335 to further limit the search region for the liver, and the trained model mask is repositioned onto the input CT image. The difference between the body and the cavity is shown in Figure 7A and Figure 7B where the body mask 700 is shown in Figure 7A and the cavity mask 710 is shown in Figure 7B .

[0037] As mentioned above, the heart is an organ that is often connected to the liver in CT images, especially in conventional CT images. Based on the cavity information, the pre-trained heart model obtained in operation 340 of Figure 3 is placed on the specified patient data. Then, the heart internal / external region is extracted in operation 350 based on the model position.

[0038] Similarly, the kidneys are another organ that is often connected to the liver in CT images, especially in conventional CT images. Based on the cavity and spine information, the kidneys will be in Figure 3The pre-trained kidney model obtained in operation 345 is placed on the designated patient data. Then, in operation 355, the internal / external regions of the kidney are extracted based on the model location.

[0039] Similar to the extraction in operations 350 and 355 for the heart and kidney respectively, the pre-trained liver model is placed on the patient data. Based on the model location and the associated heart and kidney regions, the internal / external regions of the liver are extracted in operation 370.

[0040] The central step in the final liver extraction involves using a graph cut method. This algorithm is used for image segmentation based on the foreground information and background information of the object. According to the present technology, the foreground information and background information are generated from the internal and external regions of the liver. Thus, a liver mask is obtained in operation 380.

[0041] Reference Figure 4 which extends the recognition shown in Figure 1 Image 400 includes a liver boundary 130, an external liver region boundary 110, and an internal liver region boundary 120. Figure 4 The regions between the boundaries are also shown in

[0042] including a foreground region 420, which is bounded by the internal liver region boundary 120 and represents the internal part of the liver. The background region 410 is bounded outside by the external liver region boundary 110 and represents a region that is definitely not the liver. The background region 410 is bounded outside by the external liver region boundary 110 and represents a region that is definitely not the liver. The uncertain region 430 is bounded outside by the external liver region boundary 110 and inside by the internal liver region boundary 120, and represents a region where it has not been determined whether it is liver substance and includes the liver boundary 130. Figure 5 The anatomical relationship between the liver and adjacent organs is shown in the diagram 500 in Figure 5 Some parts of the heart and kidney regions should be excluded from the external liver region boundary 110 in order to obtain good liver segmentation results. Otherwise, the final graph cut result may include parts of the heart and / or kidney because they are connected together in the CT image.

[0043] Similar internal and external regions of the heart and kidney are found using this algorithm. Figure 5 The heart boundary 560, the external heart region boundary 540, and the internal heart region boundary 550 are also shown in Figure 5 The kidney boundary 530, the external kidney region boundary 510, and the internal kidney region boundary 520 are further shown in

[0044] According to the present technology, the external liver region boundary 110 will not touch the internal heart region boundary 550 and the internal kidney region boundary 520. Additionally, according to Figure 5 , the internal liver region boundary 120 will not collapse together with the external heart region boundary 540 and the external kidney region boundary 510. The lung image density is constant in both plain CT and contrast-enhanced CT and may be easier to extract, so the left lung lobe 570 and the right lung lobe 580 (which may alternatively be reversed in orientation) can be referenced to confirm the liver position. There are three training models for the liver, heart, and kidney, which cover all possible regions of each organ. Based on the corresponding training models, the initial constraint of the algorithm is the maximum external region of each organ. Based on the corresponding training models, small internal regions will be discovered. Then the algorithm operates to expand the internal regions of each corresponding organ and reduce its external regions. The input to the algorithm includes liver CT image data (contrast-enhanced CT or plain CT data) and trained liver, heart, and / or kidney models. The output includes a liver mask image and constraints. The CT image data includes all liver regions and may also include at least a portion of one or two lung lobes, but does not need to include the entire heart or kidney.

[0045] Figures 6A to 6D A coronal view of CT images (including contrast-enhanced and plain CT images) of the same patient is shown. Figure 6A A contrast-enhanced CT image 600 is shown, which shows the liver in combination with the heart. Figure 6B A contrast-enhanced CT image 610 is shown, which shows the liver in combination with the kidney. Figure 6C A plain CT image 620 is shown, which shows the liver in combination with the heart. Figure 6B A plain CT image 630 is shown, which shows the liver in combination with the kidney. The densities of the liver, heart, and kidney are different between plain CT and contrast-enhanced CT, and the density is similar in plain CT.

[0046] Figure 8 It is a flowchart of a method 800 for extracting the abdominal cavity and repositioning the model. The process proceeds from the start ellipse to operation 810, which indicates obtaining a CT body image. When extracting the internal and external regions of the liver, the input CT image can be downsampled at a resolution of 4 mm × 4 mm × 4 mm to reduce CPU usage. Before the downsampling action, a 5×5 Gaussian kernel or a similar smoothing filter can be applied to process the CT image slice by slice. In the CT data, there may be noise, which can be removed through preprocessing. This can be achieved through two region growings. The first region growing may include growing from seeds outside the body to connect all pixels outside the body. Based on seeds inside the body, the second region growing can connect pixels from the center of the (body's) image. This solution can remove objects outside the body.

[0047] The process in method 800 proceeds from operation 810 to operation 820, which indicates calculating the right lung area. The right lung area is calculated based on air density using a region growing method. Then, the lung / body size ratio is calculated, assuming that the lung area size is the largest at the liver top slice position. The lung / body area ratio may decrease downward in progressive slices. The slice where the ratio is less than 0.15 may be approximately where the liver region size is the largest.

[0048] The process in method 800 proceeds from operation 820 to operation 830, which indicates obtaining a cavity mask. The process in method 800 proceeds from operation 830 to operation 840, which indicates obtaining liver abdominal information. When the clean body mask is ready, the algorithm extracts the abdominal mask and key slice positions. Based on the abdominal information from operation 840, the trained liver model from input 660 can be repositioned to the appropriate location in operation 860. Additionally, based on the abdominal information from operation 840, the trained heart model from input 640 can be repositioned to the appropriate location in operation 850. Further, based on the abdominal information from operation 840, the trained kidney model from input 645 can be repositioned to the appropriate location in operation 870. Repositioning the trained models may include translating and scaling the trained models based on real CT data (including cavity center and size information). Repositioning may also include removing the model parts that are air in the CT image. The repositioned model masks can be regarded as the initial organ external regions.

[0049] Figure 9AIt is a flowchart showing a method 900 for extracting internal and / or external regions of a kidney. The process proceeds from the start ellipse to operation 905, which indicates repositioning the kidney model. The flow in method 900 proceeds from operation 905 to operation 910, which indicates removing the spine and the parts outside the cavity. The flow in method 900 proceeds from operation 910 to operation 915, which indicates detecting the circular roundness of the kidney slice by slice. The flow in method 900 proceeds from operation 915 to operation 920, which indicates finding the central slice. The flow in method 900 proceeds from operation 920 to operation 925, which indicates finding the internal seed mask. The internal seed position of the kidney is calculated based on the regional circular roundness. This algorithm selects the seed points where the possible kidney shape in the slice is most like a circle. The flow in method 900 proceeds from operation 925 to operation 930, which indicates performing threshold growth cutting. An improved threshold growth cutting method is used instead of a simple region growing method to expand the internal region. The flow in method 900 proceeds from operation 930 to operation 935, which indicates performing morphology to fill the open holes. The flow in method 900 proceeds from operation 935 to operation 940, which indicates shrinking the mask. The flow in method 900 proceeds from operation 940 to operation 945, which indicates obtaining the internal region of the kidney. The method for extracting the internal and / or external region of the kidney is similar to the method for extracting the internal and / or external region of the heart. The repositioned kidney model can be regarded as the external region, and then the process locates the internal region. The flow in method 900 proceeds from operation 945 to the end ellipse.

[0050] The spine mask is used for liver and kidney internal / external region extraction. The spine mask is segmented by a simple threshold method, and then three-dimensional morphology is performed after processing to shrink the mask. The repositioned heart model can be regarded as the initial heart external region covering all possible heart regions. The internal heart region can be found in the repositioned model mask. In many CT images, the heart is connected to the liver, so the segmentation result of simple region growing in the heart will touch the liver. This technique solves this problem by morphological erosion operation to reduce the mask with an adaptive radius. In the slices that result in touching the liver, the radius should be larger. This technique provides for performing region growing twice. The first region growing is performed in the whole cavity, and the second is performed in the heart external region.

[0051] Figure 9Bis a flowchart showing a method 950 for extracting internal and / or external regions of the heart. The process proceeds from the start oval to operation 955, which indicates repositioning the heart model. The flow in method 950 proceeds from operation 955 to operation 960, which indicates removing portions outside the cavity. The flow in method 950 proceeds from operation 960 to operation 965, which indicates detecting mask shrinkage slice by slice and cleaning the mask below the slice. In the shrinkage method of operation 965, the central slice is assumed to be the slice with the maximum area, and the masks of other slices should not extend beyond their regions. The mask becomes smaller, and the process recursively removes the excess of the mask slice by slice. The flow in method 950 proceeds from operation 965 to operation 970, which indicates identifying the external region of the heart. The flow in method 950 proceeds from operation 970 to operation 975, which indicates finding the internal heart seeds and thresholds. The seed point is the central mask position of the top slice of the liver. The flow in method 950 proceeds from operation 975 to operation 980, which indicates performing three-dimensional growth twice in different regions with the same threshold and seeds. The flow in method 950 proceeds from operation 980 to operation 985, which indicates comparing the mask areas slice by slice to calculate the morphological radius. The flow in method 950 proceeds from operation 985 to operation 990, which indicates performing an erosion operation with the radius and finding the largest connected mask in the slice. The flow in method 950 proceeds from operation 990 to operation 995, which indicates obtaining the internal region of the heart. The flow in method 950 proceeds from operation 995 to the end oval.

[0052] Figure 10It is a flowchart showing method 1000 for initializing liver data. The process proceeds from the start ellipse to operation 1010, which indicates receiving a cavity image. The flow in method 1000 also proceeds from the start ellipse to operation 1020, which indicates receiving a liver model mask. The flow in method 1000 proceeds from operations 1010 and 1020 to operation 1025, which indicates obtaining an original liver mask. The flow in method 1000 proceeds from operation 1025 to operation 1040, which also receives the input of external kidney data in operation 1030 and indicates correcting the liver data. The flow in method 1000 proceeds from operation 1040 to operations 1050 and 1055, both of which indicate region growing. The flow proceeds from operation 1050 to operation 1060, which indicates obtaining the liver data after region growing. Similarly, the flow proceeds from operation 1055 to operation 1065, which also indicates obtaining the liver data after region growing. The flow proceeds from operation 1060 to operation 1080, which also receives the input of external kidney data in operation 1030 and external heart data in operation 1070. Operation 1080 indicates obtaining original internal liver data. The flow proceeds from operation 1080 to operation 1090, which indicates obtaining a clean internal mask. The flow proceeds from operation 1090 to operation 1097, which indicates obtaining internal liver data.

[0053] The flow proceeds from operation 1065 to operation 1085, which also receives the input of internal kidney data in operation 1035 and internal heart data in operation 1075. Operation 1085 indicates obtaining original external liver data. The flow proceeds from operation 1085 to operation 1095, which indicates obtaining a clean external mask. The flow proceeds from operation 1095 to operation 1099, which indicates obtaining external liver data. The flow in method 1000 proceeds from operations 1097 and 1099 to the end ellipse.

[0054] Method 1000 includes two sub-steps: first, liver region initialization, and second, liver region refinement. In the liver region initialization step, the goal is to extract the internal and external data of the liver and ensure that the internal data is included in the liver and the liver is included in the external data. The program corrects the liver data according to the liver model data and the external data of the kidney, and then the program extracts the internal / external data of the liver.

[0055] In the process of extracting the internal data of the liver, the program first erodes the liver model data. Then, the region growing algorithm executes seeds from the base slice mask of the liver model. The actions of removing the external data of the spine, heart, and / or kidney are also performed. Next, the internal mask data of the liver is cleaned to ensure that all the internal liver data is within the range of the liver.

[0056] The method of extracting the external data of the liver is similar to the method of extracting the internal data of the liver. The program first performs a region growing algorithm on the seeds of the base slice mask from the liver model data. The data of the spine, heart, and / or kidneys is also removed. The external mask data of the liver is cleaned to ensure that the liver data is within the range of the external liver data. Finally, a dilation operation is performed on the external liver data.

[0057] Figure 11 is a flowchart showing a method 1100 for refining the liver region. The process proceeds from the start ellipse to operation 1110, which indicates receiving the original internal liver mask. The flow in method 1100 also proceeds from the start ellipse to operation 1120, which indicates receiving the original external liver mask. The flow in method 1100 proceeds from operations 1110 and 1120 to operation 1130, which indicates performing a growth cut. The flow proceeds from operation 1130 to operation 1025, which indicates obtaining the original liver mask. The flow proceeds from operation 1025 to operation 1140, which indicates performing erosion, removing the spine, and cleaning the internal mask. The flow proceeds from operation 1140 to operation 1150, which indicates obtaining the internal liver mask. The flow proceeds from operation 1150 to operation 1160, which also receives the input from operation 1120 and indicates performing a growth cut. The flow proceeds from operation 1160 to operation 1170, which indicates obtaining the liver mask. The flow in method 1100 proceeds from operation 1170 to operation 1180, which indicates cleaning the external mask and performing dilation. The flow proceeds from operation 1180 to operation 1190, which indicates obtaining the external liver mask. The flow in method 1100 proceeds from operation 1190 to the end ellipse.

[0058] Method 1100 refines the internal and / or external liver data to make the internal data larger and the external data smaller. Based on the original internal / external liver data, the algorithm performs a growth cut method to obtain the original liver data mask. After the erosion operation on the original liver data mask, the external spine is removed and the internal mask is cleaned to obtain the internal liver data.

[0059] Based on the internal liver data and the original external liver data, the algorithm performs a threshold growth cut method to obtain the liver data mask. Then the external mask is cleaned, and a dilation operation is performed. Then the program obtains the external liver data.

[0060] Figure 12FIG. 1200 is a flow chart showing a method 1200 for generating a final liver mask. The process proceeds from the start ellipse to operations 1210, 1220, and 610. Operation 1210 indicates obtaining an internal liver region, and operation 1220 indicates obtaining an external liver region. Operation 610 indicates obtaining an input CT image. The flow in method 1200 proceeds from operations 1210 and 1220 to operation 1230, which indicates upsampling a synthesized VOI (volume of interest) label image. The flow proceeds from operation 610 to operation 1240, which indicates downsampling a VOI CT image. The flow in method 1200 proceeds from operations 1230 and 1240 to operation 1250, which indicates identifying a possible background image and a foreground image. The flow proceeds from operation 1250 to operation 1260, which indicates obtaining a liver graphical image. The flow proceeds from operation 1260 to operation 1270, which indicates performing a graphical cut to obtain a resulting mask. The flow proceeds from operation 1270 to operation 1280, which indicates performing post-processing on the mask. The flow proceeds from operation 1280 to operation 1290, which indicates upsampling the final liver mask. The flow in method 1200 proceeds from operation 1290 to the end ellipse.

[0061] Graph cut methods (also known as maximum flow algorithms) have been employed to effectively solve various computer vision problems such as image segmentation. Based on the internal and external liver regions, an improved graph cut can be applied for liver segmentation.

[0062] Extract the minimum volume of interest (VOI) region to reduce CPU usage. Then, the algorithm labels the liver internal region pixels as foreground and the pixels outside the external region as background. The regions located within the external boundary and outside the internal boundary are considered unknown and remain segmented and / or labeled (see Figure 4 ). A Gaussian mixture model (GMM) can be used to analyze foreground and background statistics. Then a foreground / background likelihood image will be constructed based on the statistics.

[0063] Then a graph is created, and an augmented path maximum flow algorithm is applied to compute the label (foreground and background) for each pixel. When creating the graph, only the unknown regions and their nearest neighboring pixels are considered in order to reduce memory and CPU usage. Finally, the algorithm performs morphological post-processing and up-samples the label mask result to the original image resolution.

[0064] Figure 13FIG. 1300 is a flow chart showing a method 1300 for generating trained heart, kidney, and / or liver models. The process proceeds from the start oval to operation 1310, which indicates inputting CT images. The flow proceeds from operation 1310 to operation 1315, which indicates downsampling the images. The flow proceeds from operation 1315 to operation 1320, which indicates cleaning the body images. The flow proceeds from operation 1320 to operation 1330, which indicates extracting an abdominal cavity mask and identifying key slices of the CT image. The information from operation 1330 is output to operations 1350, 1370, and 1390, which represent heart, kidney, and liver models, respectively. The heart model also has as input a heart ground truth mask from operation 1340. Similarly, the kidney model also has as input a kidney ground truth mask from operation 1360. The flow proceeds from operation 1390 to the end oval.

[0065] The model training protocol is similar to the protocol for the input image process. After downsampling, cleaning the body images, and extracting cavity and key slice information, the user-drawn ground truth masks are translated and scaled to generate the models. The ground truth data is imported and accumulated together to generate the final trained model.

[0066] Reference Figure 14 , the present disclosure can be used by or executed on a computing device 1400, such as a laptop computer, desktop computer, tablet computer, or other similar device, which has a display 1406, a memory 1402, one or more processors 1404, and / or other components of the type typically present in a computing device. The display 1406 can be touch-sensitive and / or voice-activated, enabling the display 1406 to act as both an input and output device. Alternatively, a keyboard (not shown), a mouse (not shown), or other data input devices can be employed.

[0067] Memory 1402 includes any non-transitory computer-readable storage medium for storing data and / or software that can be executed by processor 1404 and control the operation of computing device 1400. In one embodiment, memory 1402 may include one or more solid-state storage devices, such as flash memory chips. Alternatively, or in addition to one or more solid-state storage devices, memory 1402 may include one or more mass storage devices connected to processor 1404 via a mass storage controller (not shown) and a communication bus (not shown). Although the description of computer-readable media included herein refers to solid-state memory, those skilled in the art should understand that computer-readable storage media can be any available medium accessible to processor 1404. That is, computer-readable storage media includes non-transitory, volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. For example, computer-readable storage media may include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state memory technologies, CD-ROM, DVD, Blu-ray or other optical storage devices, magnetic tape cartridges, tapes, magnetic disk storage devices or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible by computing device 1400.

[0068] Memory 1402 may store CT data 1414, which may be raw data or processed data. Additionally, memory 1402 may store application program 1416, which can be executed by processor 1404 to run any program described herein. Application program 1416 may include instructions for operating user interface 1418, which may utilize input device 1410.

[0069] Computing device 1400 may also include network interface 1408, which is connected to a distributed network or the Internet via a wired or wireless connection for sending data to and receiving data from other sources. For example, computing device 1400 may receive computer tomography (CT) image data of a patient from a server (e.g., a hospital server, an Internet server or other similar server) for use during surgical ablation planning. The patient CT image data may also be provided to computing device 1400 via removable memory 1402.

[0070] The liver segmentation module may include a software program stored in the memory 1402 and executed by the processor 1404 of the computing device 1400. The liver segmentation module may communicate with a user interface 1418, which may generate a user interface for presenting visual interaction features to a clinician, for example, on the display 1406 and for receiving clinician input, for example, via the input device 1410. For example, the user interface module 1418 may generate a graphical user interface (GUI) and output the GUI to the display 1406 for viewing by the clinician.

[0071] Although the various embodiments have been described in detail with reference to the drawings for purposes of illustration and description, it should be understood that the methods and apparatus of the present invention should not be considered limited. It will be apparent to those of ordinary skill in the art that various modifications can be made to the foregoing embodiments without departing from the scope of the present disclosure.

Claims

1. A method for identifying the liver in a CT image of a patient, comprising: applying a liver model to the CT image; extracting from the CT image an internal liver region boundary, a foreground region surrounded by the internal liver region boundary and representing the internal part of the liver, an external liver region boundary surrounding the internal liver region and the foreground region, and an uncertain region defined between the external liver region boundary and the internal liver region boundary, the uncertain region representing a region of the CT image of a part that has not been determined to be the liver; and performing a graph cut algorithm on the CT image based on the internal liver region boundary, the foreground region, the external liver region boundary, and the uncertain region to generate a liver image.

2. The method according to claim 1, further comprising: applying a heart model to the CT image; and extracting from the CT image an internal heart region and an external heart region based on the applied heart model; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the internal heart region and the external heart region.

3. The method according to claim 2, further comprising: applying a kidney model to the CT image; and extracting from the CT image an internal kidney region and an external kidney region based on the applied kidney model; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the internal kidney region and the external kidney region.

4. The method according to claim 3, wherein extracting the internal liver region boundary from the CT image is further based on the external kidney region and the external heart region.

5. The method according to claim 3, wherein extracting the external liver region boundary from the CT image is further based on the internal kidney region and the internal heart region.

6. The method according to claim 1, further comprising: extracting a body mask from the CT image; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the body mask.

7. The method according to claim 6, further comprising: extracting a cavity mask from the CT image based in part on the body mask; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the cavity mask.

8. A non-transitory computer-readable storage medium encoded with a program that, when executed by a processor, causes the processor to perform the following steps: applying a liver model to a CT image; extracting from the CT image an internal liver region boundary, a foreground region surrounded by the internal liver region boundary and representing the internal part of the liver, an external liver region boundary surrounding the internal liver region and the foreground region, and an uncertain region defined between the external liver region boundary and the internal liver region boundary, the uncertain region representing a region of the CT image of a part that has not been determined to be the liver; and Perform a graph cut algorithm on the CT image based on the internal liver region boundary, the foreground region, the external liver region boundary, and the uncertain region to generate a liver image.

9. The non-transitory computer-readable storage medium according to claim 8, wherein when the program is executed, it further causes the processor to perform the following steps: Apply a heart model to the CT image ； and Extract an internal heart region and an external heart region from the CT image based on the applied heart model; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the internal heart region and the external heart region.

10. The non-transitory computer-readable storage medium according to claim 9, wherein when the program is executed, it further causes the processor to perform the following steps: Apply a kidney model to the CT image ； and Extract an internal kidney region and an external kidney region from the CT image based on the applied kidney model; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the internal kidney region and the external kidney region.

11. The non-transitory computer-readable storage medium according to claim 10, wherein extracting the internal liver region boundary from the CT image is further based on the external kidney region and the external heart region.

12. The non-transitory computer-readable storage medium according to claim 10, wherein extracting the external liver region boundary from the CT image is further based on the internal kidney region and the internal heart region.

13. The non-transitory computer-readable storage medium according to claim 8, wherein when the program is executed, it further causes the processor to perform the following steps: Extract a body mask from the CT image; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the body mask.

14. The non-transitory computer-readable storage medium according to claim 13, wherein when the program is executed, it further causes the processor to perform the following steps: Extract a cavity mask from the CT image based in part on the body mask; wherein performing the graph cut algorithm on the CT image to generate the liver image is further based on the cavity mask.

Images