A three-dimensional spatial organ medical image feature extraction system and method for reducing false positives

Abstract

This invention provides a three-dimensional spatial organ image feature extraction system that reduces false positives. [Solution] A feature extraction method is developed to serve as the basis for classifying organ positives and false positives. Multiple organ cross-sectional layers of the same type of three-dimensional organ are extracted using MRI. An AI analysis system generates organ cross-sectional layers for each organ, and a specific organ marking image is generated. The organ cross-sectional layers from which the specific organ marking image was generated are imported into the image analysis unit. The centroid point of the specific organ marking image and the organ cross-sectional layer containing the specific organ marking image are imported into the feature extraction unit. The feature extraction unit calculates minimum distance data based on the centroid point and reference point, calculates area data based on the specific organ marking image, generates a layer number data set based on total layer number information, current layer number information, and image code information, and imports the distance data, area data, and layer number data set into the feature calculation unit to generate matching results. The post-processing unit selects whether to delete or retain the specific organ marking image based on the matching results.

Description

Technical Field

[0001] The present invention relates to a feature extraction system, and particularly to a three-dimensional spatial organ medical image feature extraction system and method for reducing false positives.

Background Art

[0002] In current medical image processing, whether it is the result generated by a neural network model or the user's annotation, it is inevitable that incorrect regions are selected and false positives occur. The reason is that the training of the neural network model is relatively complex and requires a large amount of training costs (such as a large number of medical images correctly annotated) and high-quality training data. In the case of user annotation, there may be problems with human labeling errors. As a result, incorrect organ areas may be marked on the medical image, which may affect the diagnosis of the patient.

[0003] For example, magnetic resonance imaging (MRI) technology can obtain multiple organ cross-sectional layers of the same type of three-dimensional organ. When processing medical image graphics of an organ (liver), whether it is the result generated by a neural network or the user's incorrect annotation by mistake, incorrect areas will be selected, and false positives, that is, areas other than the liver will be labeled, which is inevitable.

[0004] ]> The detailed features and advantages of the present invention are described in detail in the following embodiments, and the content is sufficient for those skilled in the related art to understand the technical content of the present invention and implement it accordingly. Furthermore, according to the content disclosed in this specification, the scope of the patent application, and the drawings, those skilled in the art can easily understand the related objectives and advantages of the present invention.

Summary of the Invention

[0005] The main objectives of this invention are to obtain accurate organ segment selection effects without requiring vast amounts of organ cross-sectional layer information or computational costs for judgment and analysis, and to reduce the false positive rate and improve the accuracy rate by effectively eliminating non-specified organ annotations in the cross-sectional layers.

[0006] To achieve the above objectives, feature extraction methods designed based on anatomical and basic physiological features have been developed and are used as a basis for classifying organ-positive and false-positive results. The present invention is a three-dimensional spatial organ image feature extraction system that reduces false positives. Using magnetic resonance imaging, multiple organ cross-sectional layers of the same type of three-dimensional organ are obtained, and an AI analysis system generates at least one specific organ marking image from the organ cross-sectional layers of each organ. The system includes an image analysis unit used to import the organ cross-sectional layers along with the specific organ marking image, and an image analysis unit that generates the centroid point of the specific organ marking image. The feature extraction unit imports the organ cross-sectional layers with a centroid and the specific organ marking image, and the feature extraction unit calculates at least one distance data based on the centroid and at least one reference point, calculates at least one area data based on the specific organ marking image, and generates a set of layer data based on total layer information, current layer information, and image code information. The feature calculation unit imports the distance data, area data and the set of layer data to generate matching results, and a post-processing unit selects whether to delete or retain the specific organ marking image depending on the matching results.

[0007] According to one embodiment of the present invention, a memory unit is also provided, which is used to store reference points and to import the reference points into a feature extraction unit.

[0008] According to one embodiment of the present invention, the organ cross-sectional layer includes total layer number information and current layer number information.

[0009] According to one embodiment of the present invention, before using the system, reference points must be established and stored in a memory unit, and reference true regions are marked using multiple cross-sectional layers of the organ. Next, an image analysis unit is used to generate reference centroids of the reference true regions, and the average value of the reference centroids is calculated to obtain the reference points.

[0010] To achieve the above objective, the present invention provides a feature extraction method for 3D organ images that reduces false positives, comprising acquiring multiple cross-sectional layers of the same type of 3D organ by magnetic resonance imaging, and generating at least one specific organ-marked image from the cross-sectional layers of each organ using an AI analysis system. This method includes the following: a. A step of importing cross-sectional layers of organs having images indicating specific organs into an image analysis unit, b. Analyze images with specific organ markings and generate the centroid point. c. A step of importing the cross-sectional layer of the organ with the centroid point and the specific organ labeling image into the feature extraction unit, d. Calculate distance data between the centroid and at least one reference point, calculate at least one area data for the specific organ-marking image, and generate a set of layer data based on the total number of layers information, the current number of layers information, and the image code information. e. Distance data, area data, and layer count data sets are imported into the feature calculation unit, which then generates matching results. f. The matching results are imported into the post-processing unit, which then selects whether to delete or retain the images labeled with specific organs based on the matching results.

[0011] According to one embodiment of the present invention, the reference point is imported from the storage unit to the feature extraction unit.

[0012] According to one embodiment of the present invention, the organ cross-sectional layers include information on the total number of layers and information on the current number of layers.

[0013] According to one embodiment of the present invention, before using the system, it is necessary to establish and store reference points in a memory unit, marking reference true regions using multiple cross-sectional layers of the organ. Next, an image analysis unit is used to generate reference centroids of the reference true regions, and the average value of the reference centroids is calculated to obtain the reference points. [Brief explanation of the drawing]

[0014] [Figure 1] Figure 1 is a schematic diagram of the three-dimensional spatial organ medical image feature extraction system for reducing false positives according to the present invention. [Figure 2] Figure 2 is a schematic diagram of the organ cross-sectional layer and specific organ labeling image of the present invention. [Figure 3] Figure 3 is a schematic diagram of the data processing of the three-dimensional spatial organ medical image feature extraction system that reduces false positives according to the present invention. [Figure 4] Figure 4 is a schematic diagram of the center of gravity of the present invention. [Figure 5] Figure 5 is a schematic diagram of the organ cross-sectional layer, reference true region, and reference centroid point of the present invention. [Figure 6] Figure 6 is a schematic diagram showing how the feature extraction unit of the present invention calculates distance data based on one of the centroid points and a reference point. [Figure 7] Figure 7 is a schematic diagram of the area data generated by the present invention. [Figure 8] Figure 8 is a schematic diagram of the layer number data set of the present invention. [Figure 9] Figure 9 is a schematic diagram of the layer number data set, distance data, area data, and matching results of the present invention. [Figure 10] Figure 10 is a schematic diagram illustrating the removal of false positive marking images according to the present invention. [Figure 11] Figure 11 is a flowchart of the method for extracting features from 3D organ medical images that reduce false positives according to the present invention. [Modes for carrying out the invention]

[0015] The present invention will be illustrated below with specific embodiments. Those familiar with this art will be able to easily understand other advantages and effects of the present invention from the contents disclosed herein.

[0016] The structures, proportions, sizes, etc., shown in the drawings herein are used solely to ensure consistency with the disclosed content for the understanding and interpretation of those skilled in the art, and are not intended to limit the conditions under which the invention can be implemented. Therefore, they have no substantial technical significance. Modifications to the structure, changes in proportions, adjustments to size, etc., remain within the scope of the technical content disclosed herein without affecting the effects or purposes that the invention can achieve. At the same time, terms such as "I," "II," and "above" used herein are for illustrative purposes only and do not limit the scope of the invention. Changes or adjustments to their relative relationships should be considered within the scope of the invention without substantially altering the technical content.

[0017] Referring to FIGS. 1 and 2, the present invention is a three-dimensional organ image feature extraction system 1 for reducing false positives. By magnetic resonance imaging (MRI), a three-dimensional multi-layer cross-sectional image of an organ, that is, a "slice-type" two-dimensional image of the organ, is obtained. For example, when the organ is the liver, a plurality of organ cross-sectional layers J (for example, 50 layers) are generated for the liver organ, and one of the organ cross-sectional layers J is imported into the AI analysis system 10. That is, one layer is selected from the 50 organ cross-sectional layers J and imported into the AI analysis system 10. Next, taking the third layer as an example, an organ cross-sectional layer J having a specific organ marked image K is generated, and then the organ cross-sectional layer J having the specific organ marked image K is imported into the AI analysis system 10. The organ cross-sectional layer J is imported into the feature determination system 20, and it is determined whether the specific organ marked image K is a false positive organ area. The feature determination system 20 includes an image analysis unit 201, a feature extraction unit 202, a feature calculation unit 203, a post-processing unit 204, and a storage unit 205. The organ cross-sectional layer J further includes total layer number information Q1 and current layer number information Q11. The total layer number information Q1 is the total cross-sectional layer number information of the three-dimensional organ. In the above example, it is the 50 layers. The current layer number information Q11 is the layer number of the organ cross-sectional layer J imported into the AI analysis system 10. In the above example, it is the third layer.

[0018] Referring to FIGS. 2 and 3, when the organ cross-sectional layer J is imported into the AI analysis system 10, the AI analysis system 10 generates an organ cross-sectional layer J having at least one specific organ marked image K. Here, the AI analysis system 10 is an artificial intelligence image recognition model and has models such as a residual network model (ResNet), a U-shaped network model (U-Net), and a masked region convolutional neural network model (Mask R-CNN). In addition, the AI analysis system 10 is applied in the field of medical image analysis and is used to mark a specific organ area in the organ cross-sectional layer J, that is, a specific organ marked image K. In this embodiment, the specific organ marked images of the liver organ cross-sectional layer J have three, namely K1, K2, and K3.

[0019] Refer to FIGS. 3 and 4. The organ cross-sectional layer J having the specific organ labeled image K is imported into the image analysis unit 201, and the image analysis unit 201 generates the centroid point P of the specific organ labeled image K. The image analysis unit 201 is software, a device or a system capable of executing data processing, logical operations, and mathematical calculations, such as a CPU or an MCU, and particularly refers to a centroid computer. In this embodiment, the centroid points P1, P2, and P3 are the centroids of the specific organ labeled images K1, K2, and K3, and according to the shape of the specific organ labeled image, the image analysis unit has various calculation methods for generating the centroid point, such as weighted average, integration, centroid formula of a polygon, and image moment operation of each coordinate.

[0020] Referring to FIG. 3, the organ cross-sectional layer J, the specific organ labeled image K, the centroid point P, and at least one reference point A are imported into the feature extraction unit 202. The feature extraction unit 202 calculates at least one distance data S according to the centroid point P and the reference point A. The feature extraction unit 202 calculates at least one area data R according to the specific organ labeled image K. The feature extraction unit 202 generates a layer number data group Q according to whether the total layer number information Q1 has a specific organ labeled image. Here, the feature extraction unit 202 is executable by software, a device or a system (such as a CPU, an MCU, etc.) capable of performing data processing, logical operations, and mathematical calculations. The feature extraction unit 202 is used to obtain feature data including the organ cross-sectional layer J, the specific organ labeled image K, the centroid point P, and the reference point A, at least one distance data S, at least one area data R, and at least one layer number data group Q. The storage unit 205 can include volatile and / or non-volatile storage units such as random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), and solid state drive (SSD). The storage unit is used to store the reference point information, and the reference point information includes at least one reference point A of each layer of the organ cross-sectional layer J.

[0021] Referring to Figures 5 and 6, an organ cross-section layer J containing the centroid P and a specific organ-marking image K is imported into the feature extraction unit 202. The feature extraction unit 202 uses the centroid P to read at least one reference point A from the memory unit 205 and calculates at least one distance data S. The reference point A is calculated and generated using multiple reference centroid points N. Before using the system, the reference point A must be established and stored in the memory unit 205. In this embodiment, the liver is used as an example. First, organ cross-sectional layers of the same layer of the liver are extracted, and the image analysis unit 201 marks the reference true regions M1 to M9, which are correct liver segment images. That is, the reference true regions M1 to M9 are marked in the same layer of organ cross-sectional layers J of different livers. Using an artificial intelligence image recognition model (such as the AI ​​analysis system 10) or expert artificial marking, the reference true regions M1 to M9 marked in organ cross-sectional layers J1 to J3 are manually marked, and then the image analysis unit 201 generates reference centroid points N1, N11, N12, N2, N3, N22, N3, N31, and N32 of the reference true regions M1 to M9, and the average value of these reference centroid points is calculated to obtain the average value, which becomes the reference points A1, A2, and A3. To further explain, when averaging the reference centroid points, the average value is calculated based on the reference centroid points of the corresponding reference true regions. Taking Figure 5 as an example, N1, N11, and N12 are groups, N2, N21, and N22 are groups, and N3, N31, and N32 are groups. Since the average values ​​are calculated separately, three reference points A1, A2, and A3 are obtained, and the data for these reference points are stored in the storage unit 205. In addition, both the centroid point P and the reference points A are coordinate data. Therefore, the distance between two coordinates can be calculated based on the two sets of coordinate data to obtain distance data S. See Figure 6 for details. In this embodiment, distance data is calculated using one of the centroid points and a reference point. Since there are three reference points, the centroid point P1 and reference points A1, A2, and A3 are calculated to obtain three sets of distance data S1, S2, and S3. Similarly, three sets of distance data will be obtained for centroid points P2 and P3, but these will not be repeated herein.

[0022] Referring to Figure 7, the organ cross-section layer J with the centroid P and the specific organ marking image K are imported into the feature extraction unit 202, and the feature extraction unit 202 calculates the area of ​​the specific organ marking image K to generate at least one area data R. To explain further, the specific organ marking images K1, K2, and K3 are closed figures, and the feature extraction unit 202 uses mathematical formulas and programs to calculate the area of ​​the closed figures to generate at least one area data R. As shown in the inclined surface range in Figure 7, the inclined surface range of the specific organ marking images K1, K2, and K3 is calculated to generate area data R1, R2, and R3, respectively.

[0023] As shown in Figure 8, an organ cross-section layer J containing the centroid P and a specific organ labeling image K is introduced into the feature extraction unit 202. The feature extraction unit 202 generates a group of layer data Q based on the total number of layers information Q1, the current layer number information Q11, and the image code information Q2. To explain further, the total number of layers information Q1 is the total number of layers of the liver organ cross-section layers. For example, there are a total of 50 layers of liver organ cross-section layers, and the total number of layers information is 1 to 50. The current layer number information Q11 is the layer number to which organ cross-section layer J belongs. For example, organ cross-section layer J is a slice image of the third layer of the liver organ, and the current layer number information Q11 is 3. The image code information Q2 is a numerical value, data, parameter, etc., indicating the state of the specific organ labeling image K in the current layer of organ cross-section layer J, that is, whether the specific organ labeling image K exists in the current layer of organ cross-section layer J. For example, if a specific organ labeling image K exists in the organ cross-section layer J, the image code information Q2 will be 1; otherwise, the image code information Q2 will be 0. The image code information Q2 can be adjusted according to the actual application and is not limited to the above 1 or 0. The feature extraction unit 202 acquires the total number of layers information Q1, the current number of layers information Q11, and the image code information Q2, and then integrates them into the layer number data group Q. Furthermore, as shown in Figure 8, if the layer number data group Q includes the total number of layers information Q1, the current number of layers information Q11, and the image code information Q2, the total number of layers information Q1 for the third layer of the liver is 1 to 50, the current number of layers information Q11 is 3, and since the third layer contains specific organ labeling images K1, K2, and K3, the image code information Q2 is 1. In addition, the image code information Q2 can also include data from other layers depending on the usage requirements. For example, in Figure 8, since there are no other layers in the organ cross-section layer J in the current layer, the image code information Q2 for other layers is 0.

[0024] Referring to Figure 3, distance data S, area data R, and layer number data set Q are imported into the feature calculation unit 203, and the feature calculation unit 203 generates a matching result T. The feature calculation unit 203 is an AI classifier, a trained AI classifier that can perform pattern recognition based on input data and output a corresponding classification result. This classifier can consist of one or more neural network models (convolutional neural network CNN, long short-term memory network LSTM, transformer, etc.) and is used to implement this method. Furthermore, feature data of specific organ labeling images K1, K2, and K3, namely distance data S, area data R, and layer number data set Q, are imported into the feature calculation unit 203, and the feature calculation unit 203 generates a matching result T. The matching result T is used to indicate whether the organ cross-section layer J is the correct organ region or a false positive, and can be numerical, data, parameters, etc. For example, see Figure 9, which is a schematic diagram of the distance data S, area data R, layer number data group Q, and matching result T for three different specific organ marking images K1, K2, and K3 of the third layer liver organ cross-section layer J. The above feature data is imported into the feature calculation unit, which generates three matching result groups T1, T2, and T3, indicated by the numbers 0, 1, and 2 in Figure 9. The numbers 0, 1, and 2 are shown in Figure 9. Furthermore, the feature calculation unit 203 needs to be trained before use. The feature data used is distance data S, area data R, and layer number data group Q obtained after confirmation from the correct organ region in the organ cross-section layer J. The corresponding result data (result, i.e., matching result T) is a numerical value, data, parameter, etc., representing the correct data. Conversely, in order to confirm the distance data S, area data R, and layer number data group Q obtained after confirmation from the false-positive organ region in the organ cross-section layer J, the corresponding result data of the matching result represents the false-positive data. The result data is set to 0 and 1 as correct data and 2 as false-detection data. Referring to Figure 9, matching result 2 is false-positive data, and the corresponding specific organ labeling image K3 is a false-positive region. Furthermore, since the specific organ labeling image K is different for each layer, the matching result T or result data can be adjusted according to the number of specific organ labeling images K in the current layer.

[0025] The post-processing unit selects whether to delete or retain the specific organ-marking image based on the matching results. The post-processing unit 204 is an image processing unit (IPU) that can receive, process, and output image data and can perform various image processing operations such as image enhancement, noise reduction, edge detection, and color adjustment. It can be a standalone module or integrated into a processing unit (CPU, GPU, NPU, etc.) or a dedicated image processing chip (ISP, Image Signal Processor, etc.). To further explain, the matching result T is imported into the post-processing unit 204. If the matching result T is false positive data, the specific organ-marking image is deleted; conversely, if the matching result T is correct data, the specific organ-marking image K is retained. See Figure 10. In this embodiment, the specific organ-marking image K3 (shaded area) is deleted from the organ cross-section layer J.

[0026] Thus, a large number of samples (organ cross-sectional layers) are not required. This is because obtaining a large number of samples is not easy in medical image analysis, and personal data issues arise. Furthermore, even if a large number of samples are obtained, the training and computation costs increase with the number of samples. Therefore, the present invention reduces the false positive rate and improves the accuracy rate by obtaining distance data S, area data R, and layer number data group Q of organ regions based on a feature extraction method designed based on the anatomy of organs, and further using these as classification criteria to determine whether an organ region is positive or false positive. The advantage of this method is that it does not require a very large amount of data and cost.

[0027] Referring to Figure 11, the present invention is a method for extracting three-dimensional spatial organ medical image features that reduce false positives. The method for extracting three-dimensional organ image features that reduce false positives involves acquiring multiple cross-sectional layers of the same type of three-dimensional organ using magnetic resonance imaging, and generating at least one specific organ-marked image from the cross-sectional layers of each organ using an AI analysis system. This method includes: a. Import the cross-sectional layers of organs that have images indicating specific organs into the image analysis unit. b. Analyze images with specific organ markings and generate the centroid point. c. Import the cross-sectional layers of organs with a centroid point and the specific organ-marked images into the feature extraction unit. d. Calculate distance data between the centroid and at least one reference point, calculate at least one area data for the specific organ-marking image, and generate a set of layer data based on the total number of layers information, the current number of layers information, and the image code information. e. Distance data, area data, and layer count data sets are imported into the feature calculation unit, which then generates matching results. f. The matching results are imported into the post-processing unit, which then selects whether to delete or retain the images indicating specific organs based on the matching results.

[0028] The embodiments described above are illustrative of the principles and effects of the present invention and do not limit it. Those familiar with the art can modify the embodiments without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention should be described in the same way as the scope of the patent application below. [Explanation of symbols]

[0029] 1. Three-dimensional spatial organ medical image feature extraction system that reduces false positives 10 AI Analysis Systems J, J1, J2, J3 Organ cross-sectional layers K, K1, K2, K3 Specific Organ Labeling Images M1, M2, M3, M4, M5, M6, M7, M8, M9 Reference truth area 20 Feature Recognition Systems 201 Image Analysis Unit 202 Feature Extraction Unit 203 Feature Calculation Unit 204 Post-processing unit 205 Memory Unit P, P1, P2, P3 Center of gravity N, N1, N11, N12, N2, N21, N22, N3, N31, N32 Reference center of gravity point A, A1, A2, A3 reference points S, S1, S2, S3 Distance Data Area data for R, R1, R2, and R3. Q: Number of layers in the data set Q1 Total number of layers information Q11 Current number of layers information Q2 Image code information T, T1, T2, T3 Matching Results

Claims

1. A three-dimensional organ image feature extraction system that reduces false positives, comprising: acquiring multiple cross-sectional layers of the same type of three-dimensional organ using magnetic resonance imaging; and generating at least one specific organ-marked image by marking specific organ regions in the cross-sectional layers using an AI analysis system; An image analysis unit that imports the organ cross-sectional layer having the image of the specific organ and generates the centroid of the image of the specific organ, A feature extraction unit that imports the organ cross-section layer having the centroid point and the specific organ marking image, the feature extraction unit that calculates at least one distance data based on the centroid point and at least one reference point, calculates at least one area data based on the specific organ marking image, and generates a set of layer data based on total layer information, current layer information, and image code information indicating whether or not the specific organ marking image currently exists in the organ cross-section layer, A feature calculation unit that imports the distance data, area data, and layer number data set and generates matching results indicating whether or not the data is a false positive, The system includes a post-processing unit that selects whether to delete or retain the specific organ labeling image according to the matching result, The aforementioned reference points were obtained by taking multiple organ cross-sectional layers of the same type of organ and the same layer, using the correct organ region as the reference true region, the centroid of the reference true region as the reference centroid point, and then performing an average calculation of the multiple reference centroid points. A three-dimensional organ image feature extraction system that reduces false positives.

2. Furthermore, the system includes a storage unit used to store the reference points and to import the reference points into the feature extraction unit. A three-dimensional organ image feature extraction system for reducing false positives as described in claim 1.

3. The organ cross-sectional layer includes the total number of layers information and the current number of layers information. A three-dimensional organ image feature extraction system for reducing false positives as described in claim 1.

4. Before using the system, the image analysis unit generates the reference centroid of the reference true region for multiple organ cross-sectional layers of the same type of organ and the same layer, where the reference true region is marked. The reference point is determined and established by calculating the average of the multiple reference centroids and stored in the storage unit. A three-dimensional organ image feature extraction system for reducing false positives, as described in claim 2.

5. A three-dimensional organ image feature extraction method that reduces false positives, comprising: acquiring multiple cross-sectional layers of the same type of three-dimensional organ by magnetic resonance imaging; and generating at least one specific organ-marked image by marking specific organ regions in the cross-sectional layers using an AI analysis system. a. Import the organ cross-sectional layer having the image of the specific organ into the image analysis unit. b. Analyze the image of the specified organ and generate the centroid point, c. The organ cross-sectional layer having the centroid point and the image indicating the specific organ is imported into the feature extraction unit. d. Calculate distance data between the centroid point and at least one reference point, calculate at least one area data for the specific organ marking image, generate a set of layer data based on the total number of layers information, the current number of layers information, and image code information indicating whether or not a specific organ marking image currently exists in the organ cross-sectional layer, e. The distance data, area data, and layer number data set are imported into the feature calculation unit, and the feature calculation unit generates a matching result indicating whether or not the data is a false positive. f. The matching results are imported into the post-processing unit, and the post-processing unit selects whether to delete or retain the specific organ labeling image according to the matching results. The aforementioned reference points were obtained by taking multiple organ cross-sectional layers of the same type of organ and the same layer, using the correct organ region as the reference true region, the centroid of the reference true region as the reference centroid point, and then performing an average calculation of the multiple reference centroid points. A three-dimensional organ image feature extraction method that reduces false positives.

6. The reference points are imported from the storage unit to the feature extraction unit. A method for extracting three-dimensional organ image features that reduces false positives, as described in claim 5.

7. The organ cross-sectional layer includes the total number of layers information and the current number of layers information. A method for extracting three-dimensional organ image features that reduces false positives, as described in claim 5.

8. Before using the system, the image analysis unit generates the reference centroid of the reference true region for multiple organ cross-sectional layers of the same type of organ and the same layer, where the reference true region is marked. The reference point is determined and established by calculating the average of the multiple reference centroids and stored in the memory unit. A method for extracting three-dimensional organ image features that reduces false positives, as described in claim 5.

Images