JP4763378B2 - Ultrasonic diagnostic equipment - Google Patents

Description

  The present invention relates to an ultrasonic diagnostic apparatus, and more particularly to control of an arithmetic processing unit (processor) that forms an ultrasonic image based on data transferred from an external memory.

  In the ultrasonic diagnostic apparatus, a three-dimensional space (three-dimensional echo data capturing space) is formed by two-dimensional scanning with an ultrasonic beam. As a result, volume data is acquired. A three-dimensional image is formed by rendering processing on the volume data. On the other hand, an arbitrary cross section (cut plane) is specified by a user at an arbitrary position and posture with respect to the three-dimensional space, and a tomographic image (arbitrary tomographic image) corresponding to the arbitrary cross section is formed.

  Generally, volume data is composed of a plurality of frame data, and each frame data is composed of a plurality of beam data. Each beam data is constituted by a data (echo data) sequence existing on the ultrasonic beam. That is, the volume data is a data (echo data) set that exists discretely in the three-dimensional space.

  In general, in rendering processing, a plurality of rays (virtual lines of sight) are set for a three-dimensional space, and voxels are sequentially generated from the viewpoint side with respect to a data string (voxel string) existing on the ray for each ray. The operation is executed. For example, in the volume rendering method, voxel calculation using parameters such as opacity is repeatedly executed. In that case, data that exactly matches the coordinates of each voxel on the ray does not necessarily exist, so data (voxel value) is generated from a plurality of data in the vicinity by interpolation, or the data in the vicinity is the voxel. Is considered to be data (voxel values) (also a kind of interpolation operation).

  Rendering processing, that is, three-dimensional image formation, is generally executed by a general-purpose processor (CPU) or a dedicated processor (DSP) provided in the ultrasonic diagnostic apparatus. These processors have an internal memory (cache memory) that can be accessed at high speed from the processor core, but usually the storage capacity is not large enough to store the entire volume data. Therefore, in the conventional ultrasonic diagnostic apparatus, volume data is temporarily stored in an external memory, and necessary data is cut out from the volume data and transferred to an internal memory in the processor. External memory is generally large in capacity, but access to it is slow. Therefore, when forming 3D images (interpolation and voxel operations), use data cached in the internal memory as much as possible (increase the cache hit rate) to reduce access to the external memory. Is requested. Incidentally, when necessary data is transferred from the external memory to the internal memory in the processor, it is generally transferred together with the surrounding data of the data (transfer in block units), and the surrounding data is also cached. Will be.

Normally, coordinate conversion is necessary due to the difference between the transmission / reception coordinate system (usually, rθφ coordinate system) and the coordinate system for data processing (xyz coordinate system), but the coordinate conversion is performed by writing data to an external memory. Sometimes, or when reading data from external memory. In the former case, the storage space (address space) of the external memory corresponds to a coordinate system for data processing, and coordinate conversion is applied to all data. In the latter case, the storage space (address space) of the external memory corresponds to the transmission / reception coordinate system, and as a result, coordinate transformation is applied to the data to be read. In any case, in order to give a voxel value to each voxel on the ray, the surrounding data is referred to, and the voxel value is obtained by interpolation calculation.
JP 2000-339441 A

  As described above, when a three-dimensional image is formed using volume data, it is required to transfer data from an external memory to an internal memory in the processor as efficiently as possible. Depending on the capacity of the internal memory (cache capacity), for example, if a ray is individually selected according to a raster scan in a ray set and voxel calculation is advanced along the depth direction for each selected ray, scanning The cached data cannot be used effectively at the time when the route turns back (a point at which the route is completely divided and becomes discontinuous). That is, the probability that a plurality of data necessary for the above interpolation calculation is cached is reduced. In addition, when the voxel operation is advanced from the first voxel to the last voxel for each ray and is repeated according to the order of the rays, depending on the capacity of the internal memory (or cache capacity), adjacent data between adjacent rays Even between them, data stored earlier may be expelled from the internal memory prior to subsequent voxel operations. That is, it is required to cache data in consideration of “temporal locality of reference” and “spatial locality of reference”. In the above-mentioned Patent Document 1, it is described that one display screen is divided into a plurality of sections and the processing is performed in units of each section. However, the selection order and three-dimensional sections in the sections are described. It has not been.

  According to the present invention, when image processing is performed by transferring data from an external memory to an internal memory in a processor in an ultrasonic diagnostic apparatus, the data on the internal memory is efficiently used, thereby reducing the image processing time. Is to be able to shorten.

(1) An ultrasonic diagnostic apparatus according to the present invention temporarily stores an external memory storing data acquired from a three-dimensional space in a living body by transmitting and receiving ultrasonic waves, and data transferred from the external memory. Having an internal memory to store, while referring to the data on the internal memory, repeatedly executing voxel operations for 3D image formation in the depth direction for each ray set for the 3D space, An arithmetic processing unit that forms a three-dimensional image, and the arithmetic processing unit includes: a ray set setting unit that sets a ray set including a plurality of rays in the three-dimensional space; and A ray group setting unit configured to set a plurality of ray groups, and in the process of sequentially selecting ray groups from the ray set according to a predetermined group selection order, for each selected ray group A progress control unit that sequentially selects rays to be subjected to voxel calculation in accordance with a predetermined ray selection order, and data required for voxel calculation on the selected ray does not exist in the internal memory, A transfer control unit that acquires data from the external memory and stores the data in the internal memory.

  According to the above configuration, a plurality of ray groups are set for a plurality of ray groups. Each ray group is preferably composed of a plurality of rays densely packed together in two dimensions, which corresponds to a processing unit or a two-dimensional section defining it. A plurality of ray groups are selected in accordance with a predetermined group selection order, and voxel calculation proceeds for each ray with the selected ray group as a unit.

  In the above configuration, since the processing can be performed in units of individual ray groups, it is possible to increase the probability that data that are spatially close to each other can be referred to in time proximity. As in the past, when individual rays are selected sequentially in a simple raster scan order, there is a large spatial discontinuity between the rays, especially at multiple turning points on the path, that is, in the past. Although the probability that the stored data is referred to again after being evicted from the cache has increased, according to the above configuration, the probability that such a problem will occur can be reduced.

  In particular, the effect becomes large when referring to a plurality of (for example, eight points) neighboring data existing around the calculation target voxel for the interpolation calculation. Therefore, according to the above configuration, it is possible to increase the cache hit rate and reduce access to the external memory. Thereby, rapid image processing can be realized, and real-time performance of image processing can be improved.

(2) An ultrasonic diagnostic apparatus according to the present invention temporarily stores an external memory storing data acquired from a three-dimensional space in a living body by transmitting and receiving ultrasonic waves, and data transferred from the external memory. Having an internal memory to store, while referring to the data on the internal memory, repeatedly executing voxel operations for 3D image formation in the depth direction for each ray set for the 3D space, An arithmetic processing unit that forms a three-dimensional image, and the arithmetic processing unit includes: a ray set setting unit that sets a ray set including a plurality of rays in the three-dimensional space; and A ray group setting unit for setting a plurality of ray groups, and a process for setting a plurality of processing units for the ray set by setting a plurality of sections in the depth direction for each ray group. In the process of sequentially selecting a ray group from the ray set according to a unit setting unit and a predetermined group selection order, a processing unit is sequentially selected in the depth direction for each selected ray group, and a predetermined processing unit is selected within the selected processing unit. A progress control unit that sequentially selects rays to be subjected to voxel calculation according to the ray selection order, and when the data required for voxel calculation on the selected ray does not exist in the internal memory, the data is And a transfer control unit that acquires the data from the external memory and stores the data in the internal memory.

  According to the above configuration, a plurality of ray groups are set for a plurality of ray groups. Each ray group is preferably composed of a plurality of rays that are two-dimensionally densely packed together, and corresponds to a two-dimensional section that defines a processing unit to be described later. In each ray group, a plurality of sections are set along the depth direction, whereby a plurality of processing units are configured along the depth direction for each ray group. As a result, a plurality of processing units arranged in a three-dimensional manner are configured for the entire ray group. Each processing unit is preferably configured as a three-dimensional subspace that extends in the depth direction, but the thickness in the depth direction can be set to one voxel.

  In image formation, a plurality of ray groups are selected according to a predetermined group selection order, and processing units are sequentially selected along the depth direction for the selected ray groups. In each processing unit, a ray is selected according to a predetermined ray selection order. The voxel calculation proceeds for each ray in each processing unit.

  In the above configuration, since the voxel calculation can be advanced for each processing unit, it is possible to increase the probability that data that are spatially close to each other can be referred to in time proximity. First, as in the past, when individual rays are selected sequentially in a simple raster scan order, a large spatial discontinuity occurs between the rays, particularly at a plurality of turning points on the path, In other words, the probability that data stored in the past is referred to again after being evicted from the cache has increased, but according to the above configuration, the probability that such a problem will occur can be reduced. Secondly, conventionally, after completing the voxel operation along the depth direction for each ray, the voxel operation for the next ray is performed, and it is spatially large when switching from the previous ray to the subsequent ray. There was a discontinuity. On the other hand, according to the said structure, said discontinuity can be eased as a result of having divided | segmented a series of processes on each ray. This increases the cache hit rate.

  According to the above configuration, the effect is particularly remarkable when referring to a plurality of (for example, eight points) neighboring data existing around the calculation target voxel for the interpolation calculation. Therefore, according to the above configuration, it is possible to increase the cache hit rate and reduce the access to the external memory. Thereby, rapid image processing can be realized, and real-time performance of image processing can be improved.

(3) The above voxel operation is preferably a product-sum operation using opacity (opacity) based on the volume rendering method, but may be other than that as long as the operation proceeds along the ray. Absent. The voxel operation may be executed only in an effective space where data actually exists. That is, an incident point where each ray crosses the three-dimensional space may be set as a calculation start point, and an exit point exiting from the three-dimensional space may be set as a calculation end point. According to the configuration, unnecessary calculation can be omitted, and a quicker image forming process can be achieved. Further, when a predetermined condition is satisfied on each ray, the voxel calculation may be terminated there for that ray. As the predetermined condition, for example, in the volume rendering method, the condition that the result (output light amount) of each voxel calculation reaches the maximum, or the cumulative added value of the opacity determined by each voxel calculation has reached a predetermined value. Condition, etc. The above-mentioned ray group may be constituted by a plurality of rays that are in parallel with each other, or may be constituted by a plurality of rays that are non-parallel to each other (for example, radially spread or converge).

  Preferably, the calculation processing unit further includes an interpolation calculation unit that obtains a voxel value of each voxel on the selected ray from a plurality of neighboring data existing in the vicinity thereof, and an interpolation calculation unit on the selected ray. A voxel operation unit that performs a voxel operation using a voxel value of each voxel.

  In the above configuration, the interpolation calculation unit immediately refers to the data necessary for the interpolation calculation in the internal memory. On the other hand, if the data necessary for the interpolation operation is not held in the internal memory, it is acquired (transferred) from the external memory and stored in the internal memory. The transferred data is referred (utilized) by the interpolation calculation unit. In the present invention, since the temporal locality and spatial locality of the reference can be enhanced, the probability that the data once stored in the internal memory is referenced again before it is evicted (overwritten) is increased. The result of the interpolation calculation (interpolation value as a voxel value) may be temporarily stored in the internal memory, or may be used immediately in the next voxel calculation.

  Preferably, the predetermined ray selection order constitutes a ray selection route corresponding to one stroke. Preferably, the ray selection route has a zigzag shape, a spiral shape, or a Peano curve shape. According to this configuration, it is possible to reduce spatial discontinuity between sequentially referred data. Preferably, the predetermined group selection order constitutes a group selection route corresponding to one stroke. Preferably, the group selection route has a zigzag shape, a spiral shape, or a Peano curve shape. According to this configuration, it is possible to reduce spatial discontinuity between sequentially referred data. The last ray of the previous ray group and the first ray of the subsequent ray group may be spatially adjacent between the ray groups in the group selection order. All rays may be sequentially selected by a single stroke route. In addition, it is desirable to perform conditioning so that at least one of the group selection route and the ray selection route has a form corresponding to one stroke.

  Preferably, each processing unit has a size corresponding to one or more voxels in the depth direction. Preferably, each processing unit has a size corresponding to a plurality of voxels in the depth direction from the start surface to the stop surface, and voxel calculation is performed from the start surface to the stop surface for each ray in each processing unit. proceed. Preferably, a voxel calculation result is inherited for each ray between two processing units adjacent in the depth direction in each ray group. As a result, the same result as that obtained when the voxel calculation is continuously performed along the depth direction for each ray can be obtained.

  As described above, according to the present invention, when image processing is performed by transferring data from an external memory to an internal memory in the processor in the ultrasonic diagnostic apparatus, the number of accesses to the external memory is reduced, and the image Processing time can be shortened.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a preferred embodiment of an ultrasonic diagnostic apparatus according to the present invention. The ultrasonic diagnostic apparatus shown in FIG. 1 is an apparatus that forms an ultrasonic image by transmitting / receiving ultrasonic waves to / from a living body.

  The 3D probe 10 is a transmission / reception unit that transmits and receives ultrasonic waves, and the 3D probe 10 forms a three-dimensional space (three-dimensional data capturing space) 40. Specifically, the 3D probe 10 has a 2D array transducer not shown in the present embodiment. The 2D array transducer is constituted by a plurality of vibration elements arranged two-dimensionally. An ultrasonic beam B is formed by the 2D array transducer, and the scanning surface S is formed by scanning the ultrasonic beam B in the θ direction. By scanning the scanning surface S in the φ direction, three-dimensional A space 40 is formed. Incidentally, r represents the depth direction.

  As described above, the three-dimensional space 40 may be formed by mechanically moving the 1D array transducer instead of electronically scanning the ultrasonic beam two-dimensionally. The 3D probe 10 is used in contact with the surface of a living body, but it may be inserted into a body cavity. As an electronic scanning method, electronic linear scanning, electronic sector scanning, and the like are known.

  The transmission / reception unit 12 functions as a transmission beamformer and a reception beamformer. That is, the transmission / reception unit 12 outputs a plurality of transmission signals to the 2D array transducer with a predetermined delay relationship. As a result, a transmission beam is formed. A plurality of reception signals output from the 2D array transducer are input to the transmission / reception unit 12, and a phasing addition process is performed on the plurality of reception signals. As a result, a reception beam is electronically formed, and a reception signal after phasing addition representing it is output to the signal processing unit 14.

  The signal processing unit 14 is a module that executes various necessary processes such as a detection process and a logarithmic compression process on the received signal. In the present embodiment, a three-dimensional image of a tissue is constructed based on echo data, but a three-dimensional image of a moving body such as a blood flow may be constructed based on Doppler information.

  The 3D memory 16 functions as an external memory in relation to the 3D image forming processor 20 described in detail later. The 3D memory 16 has a three-dimensional storage space, and a data group acquired in the three-dimensional space 40 is stored at each address on the 3D memory 16. A plurality of echo data is obtained by performing A / D conversion (sampling) on the reception signal corresponding to one ultrasonic beam in the transmission / reception unit 12 or the signal processing unit 14, and the echo data is converted into 3D. Stored in the memory 16.

  The three-dimensional space 40 is defined by a transmission / reception coordinate system, while the 3D image forming processor 20 needs to perform coordinate transformation at any stage in order to perform calculations according to the orthogonal coordinate system. Such coordinate conversion can be performed at the time of writing each data in the 3D memory 16 or can be performed at the time of reading each data from the 3D memory 16. That is, the storage space of the 3D memory 16 corresponds to the transmission / reception coordinate system or corresponds to the orthogonal coordinate system after coordinate conversion.

  As shown in FIG. 1, a 2D image forming processor 18 and a 3D image forming processor 20 are connected to the 3D memory 16. The processors 18 and 20 can be constituted by a single processor. Each of the processors 18 and 20 is configured by a digital signal processor (DSP) that operates according to a program in the present embodiment. The 2D image forming processor 18 forms a tomographic image corresponding to a cut surface arbitrarily set by the user in the three-dimensional space 40. Data of each coordinate on the cut surface is read out from the 3D memory 16, and a tomographic image is formed according to the data read out as such. In that case, an interpolation calculation is performed as necessary.

  The 3D image forming processor 20 is a processor that constructs a three-dimensional image that spatially represents the three-dimensional space 40. In this embodiment, as will be described in detail later, a ray group composed of a plurality of rays is set in the three-dimensional space 40, and by sequentially executing voxel operations according to the volume rendering method for each ray, A pixel value is obtained for each ray. A three-dimensional image is constructed by mapping these pixel values on a plane. As described later, the 3D image forming processor 20 has various functions such as an interpolation calculation function, a voxel calculation function, and a data transfer control function.

  The data of the 3D image formed by the 3D image forming processor 20 is output to the display processing unit 22. Data of tomographic images formed by the 2D image forming processor 18 is also output to the display processing unit 22. The display processing unit 22 has a function of combining image data and graphic data, and an image to be displayed on the display unit 24 is generated by the display processing unit 22. In FIG. 1, a module for forming a normal B-mode image (tomographic image) or the like is not shown.

  The control unit 28 performs operation control of each component shown in FIG. An operation panel 26 having a keyboard, a trackball, and the like is connected to the control unit 28, and the user can select an operation mode and set operation conditions using the operation panel 26.

  The 3D image forming processor 20 includes a processor core 30 and an internal memory 32. The processor core 30 functions as a calculation unit that executes interpolation calculation, voxel calculation, and the like. The internal memory 32 has an input memory 34, a cache memory 36, and an output memory 38 in the example shown in FIG. The input memory 34 functions as an input buffer, and data transferred from the 3D memory 16 is temporarily stored therein. The cache memory 36 is a memory for facilitating the later data reference by caching the data transferred as described above. In general, data stored in the input memory 34 is provided to the processor core 30 via the cache memory 36. As the cache memory 36, a memory device that can be accessed at high speed is used. It should be noted that a multi-stage cache structure can be adopted as the cache memory 36. That is, it is possible to provide a primary cache memory and a secondary cache memory. The output memory 38 functions as an output buffer, and stores the result of the volume rendering operation executed for each ray in the processor core 30. That is, the pixel value for each coordinate on the display plane is mapped on the output memory 38. Thereby, a three-dimensional image representing the three-dimensional space 40 is constructed on the output memory 38.

  As described above, the processor core 30 performs an interpolation operation and a voxel operation. That is, the voxel calculation is sequentially advanced for each ray in the order as described in detail later, and the data (voxel value) of each sample point on the ray required at that time is obtained by the interpolation calculation. In this embodiment, for each voxel, that is, each sample point, for example, eight data (neighboring data) existing around the voxel are referred to, and an interpolation value, that is, a voxel value is obtained by executing an interpolation operation based on them. It has been demanded.

  In this case, if the neighborhood data required for the interpolation calculation is held in the cache memory 36, the neighborhood data is referred to by the processor core 30. On the other hand, if the data required for the interpolation calculation does not exist on the cache memory 36, a transfer process for acquiring the data from the 3D memory 16 is executed. The transferred data is stored on the input memory 34 and then stored on the cache memory 36, and the data is referred to by the processor core 30. In the present embodiment, reference and transfer of data as described above are controlled by a cache controller (not shown).

  When acquiring certain data from the 3D memory 16, the cache controller performs control so that data transfer is performed in units of blocks including the data. In this embodiment, the block is constituted by a predetermined number of echo data aligned on the same ultrasonic beam. That is, one block corresponds to one echo data string. The length of the echo data string is set in advance by the cache controller. Therefore, in addition to the data that is actually referred to as obtained by the control of the cache controller as described above, a plurality of other data existing in the block including the cache memory 36 are also stored on the cache memory 36. . Therefore, when the processor core 30 performs an interpolation operation or the like, in order to make more effective use of such cached data and further reduce access to the external memory, an operation order as described in detail below. Control is being performed.

  FIG. 2 shows a ray set 42 set for the three-dimensional space 40. Specifically, the ray set 42 includes a plurality of rays 46 extending from a viewpoint (upward in FIG. 2) set outside the three-dimensional space 40. The plurality of rays 46 may be parallel to each other or non-parallel to each other. FIG. 2 shows an example of the relationship between the three-dimensional space 40 and the ray set 42. In FIG. 2, the three-dimensional space 40 is surrounded by the ray set 42. The set 42 may be set so as to wrap around a part of the three-dimensional space 40. In that case, the partial three-dimensional image is constructed. Alternatively, a three-dimensional region of interest may be set for the three-dimensional space 40, and a ray set 42 may be set for the region of interest (which also corresponds to the three-dimensional space).

  The ray set 42 described above is composed of a plurality of rays 46, and voxel operations are sequentially executed for each voxel from the viewpoint side to the mapping plane 52 side for each ray. As a result, a pixel value as a final result value of the voxel calculation is obtained for each ray, and the pixel value is mapped to an address on the mapping plane 52 associated with the ray. This is indicated by P (x, y) in FIG. The mapping plane 52 has an xy coordinate system, and the ray set 42 has an xyz coordinate system. Incidentally, in addition to such a coordinate system, an XYZ coordinate system is defined as a coordinate system after coordinate conversion from the above-described transmission / reception coordinate system, but this is not shown.

  In this embodiment, in order to increase the hit rate of the cache, some measures are taken. First, the ray set 42 is divided into a plurality of ray groups 44 as shown in FIG. Each ray group 44 is configured as a plurality of ray arrays that are two-dimensionally densely packed together. Second, a plurality of sections 50 are set in the depth direction, that is, a plurality of z directions, for each ray group 44, and a processing unit 48 is defined by each section 50. That is, a plurality of processing units 48 are set in one ray group 44 along the depth direction. A plurality of processing units 48 are defined three-dimensionally as a whole of the three-dimensional ray set 42. The thickness of each processing unit 48 in the z direction, that is, the size of the section 50 is set to be equal to or greater than at least one voxel, and is preferably defined as corresponding to a plurality of voxels.

  The order of voxel operations is determined according to the following conditions in this embodiment. First, a predetermined group selection order is determined, and each ray group 44 is sequentially selected according to the predetermined group selection order. As described above, a plurality of processing units 48 are set for each ray group 44, and the processing units 48 are sequentially selected along the depth direction for each ray group 44. Further, in the present embodiment, a predetermined ray selection order is set in the ray group 44, and in the processing of each processing unit 48, rays are sequentially selected according to the predetermined ray selection order. That is, by sequentially proceeding with voxel operations according to such a plurality of conditions, it is possible to effectively increase the probability of referring to cached data, that is, it is possible to reduce access to external memory. Become. In other words, it is possible to improve the temporal locality of the reference and the spatial locality of the reference in the cache.

  FIG. 3 shows one processing unit 48 as a representative. The processing unit 48 is a space sandwiched between the i-th start surface corresponding to the stop surface (i-1th stop surface) in the previous processing unit and the i-th stop surface as the stop surface for the processing unit. It corresponds to. A plurality of rays constituting the ray group in which the processing unit 48 exists penetrates the processing unit 48. Specifically, the processing unit 48 is constituted by a plurality of sections 50 as ray fragments. In the process of advancing the voxel calculation with respect to the processing unit 48, the sections 50 are selected in order according to the predetermined ray selection order described above, and the depth direction is determined with respect to the plurality of voxels 54 on each ray included in the section. The voxel operations are sequentially executed along When the last voxel operation on a certain ray, that is, on a certain section 50 is completed, the next ray, that is, the next section 50 is selected, and the voxel operation is sequentially executed for each voxel along the depth direction. This is repeated until the last section 50 in the processing unit 48. When all of the voxel operations for one processing unit 48 are completed, the next processing unit shifted by one in the depth direction from the processing unit becomes the target of processing, and processing according to the predetermined ray selection order as described above is performed. Will be executed. When the processing for the last processing unit in a certain ray group 44 is completed, the next ray group according to the ray selection order is selected, and the processing is started for the first processing unit of the ray group. When such a process is completed for the last processing unit in the last ray group, a three-dimensional image is finally constructed on the mapping plane 52 shown in FIG.

  However, in the present embodiment, several calculation end conditions are defined for volume rendering for each ray. First, if the voxel operation reaches the last voxel for each ray, the voxel operation for that ray ends there. Secondly, when each voxel calculation proceeds in sequence, if the result (output light amount) of the voxel calculation reaches the maximum value, the voxel calculation for the ray ends there. Third, when the voxel calculation proceeds for each ray, the opacity value (opacity value) used in each voxel calculation is cumulatively added and reaches a predetermined value (for example, 1). At the time, the voxel operation for the ray ends.

  Therefore, as described above, the voxel calculation proceeds in accordance with the group selection order, the arrangement of the processing units in the depth direction, and the ray selection order. When the end condition or the opacity end condition is satisfied, the voxel calculation is ended only for the ray. Of course, such conditioning is merely an example, and various modifications may be employed according to the requirements of the actual device. For example, the end condition as described above may not be set, or another end condition may be set.

Next, the above-described group selection order and ray selection order will be described. In FIG. 4, the ray set 42 is shown as a conceptual diagram. As described above, a plurality of ray groups are set two-dimensionally for the ray set 42, and in this embodiment, n ray groups from G ₁ to G _n are set as shown in FIG. ing. Each ray group has a quadrangular shape, that is, a plurality of rays in a two-dimensional matrix.

  FIG. 5 shows some examples of the ray selection order set for each ray group. In both examples, a one-stroke ray selection route is configured. Specifically, ray selection routes 58 and 60 having a zigzag form are shown in (A1) and (A2). Ray selection routes 62 and 64 having a spiral shape are shown in (B1) and (B2). (C) shows a ray selection route 66 having a Peano curve form. In each ray selection route, there is no large spatial gap between adjacent rays on the route, so that the cache hit rate can be increased.

  FIG. 6 shows an example of the group selection order. In the illustrated example, a group selection route 68 having a zigzag form is shown. Individual ray groups existing on the route along the positive direction in the group selection route 68 are sequentially selected. According to such a configuration, the spatial discontinuity or gap between the ray groups can be reduced, so that the cache hit rate can be increased as described above. Incidentally, as a form of the group selection route, a spiral form or a Peano curve form as shown in FIG. 5 can be adopted in addition to the zigzag form shown in FIG. Note that all the rays may be selected with a single stroke route.

  Next, an operation example of the 3D image forming processor illustrated in FIG. 1 will be described with reference to FIG.

  In S101, a ray group is set for the three-dimensional space. Specifically, for example, a viewpoint is arbitrarily set by a user for a three-dimensional space, and a ray group including a plurality of rays is defined by the viewpoint. The plurality of rays may be in a parallel relationship or non-parallel relationship. In S102, a plurality of ray groups are set for the ray group. Each ray group is configured as a matrix array having a rectangular shape as described above. In S102, a plurality of sections are set along the depth direction for each ray group, that is, a plurality of processing units are set. In S103, one ray group is selected according to a predetermined group selection order. In S104, one processing unit is selected in the order along the depth direction in the selected ray group. In S105, one ray is selected according to the ray selection order within the selected processing unit. In S106, one voxel is selected in the order along the depth direction in the selected ray.

  In S107, an interpolation operation is executed for the selected voxel. In that case, a neighborhood data set necessary for the interpolation calculation is acquired from the cache memory or the external memory. That is, if neighboring data necessary for the interpolation calculation exists on the cache memory, the operation of acquiring it from the external memory is not performed, and the neighboring data on the cache memory is referred to as it is. On the other hand, if the neighboring data does not exist on the cache memory, the block including the neighboring data is taken into the 3D image forming processor from the external memory by the action of the cache controller described above. The above-mentioned neighboring data included in the block is referred to by the interpolation calculation, and a plurality of data constituting the block are stored on the cache memory. In S107, the voxel value that is the result value of the interpolation operation is stored in the internal memory.

  In S108, the voxel calculation according to the volume rendering method is executed for the voxel selected in S106 using the voxel value calculated as described above. In S108, the calculation result is stored in the internal memory.

  In S109, it is determined whether or not the voxel calculation has reached the stop plane determined for the processing unit currently being calculated. If not, each step from S106 is repeatedly executed. In S109, the termination condition of the voxel calculation is also considered, and the process moves to S110 not only when the voxel calculation reaches the stop surface but also when the predetermined termination condition is satisfied.

  In S110, it is determined whether or not the voxel calculation is completed for all the rays in the processing unit that is the current calculation target. If not completed, the steps after S105 are repeatedly executed. If the voxel calculation has been completed for all the rays, it is determined in S111 whether or not the processing for the final processing unit in the currently calculated ray group has been completed. Each process after S104 is repeatedly executed.

  In S112, it is determined whether or not the processing for all the ray groups has been completed. If the processing has not been completed, each step after S103 is repeatedly executed. When processing for all the processing units is completed, a three-dimensional image is constructed with the pixel values obtained for each ray.

  In the above-described embodiment, the 3D image is constructed according to the volume rendering method. However, the 3D image may be constructed using other methods, and in any case, the voxel operation is sequentially performed along the ray. It is desirable to apply the inherent control in the present invention when applied processing is applied.

  In the above embodiment, the voxel density on each ray, that is, the sample interval, according to the capacity of the cache memory provided in the image forming processor, the ray density constituting the ray group, or the like. Depending on the above, the shape of each processing unit may be variably set as appropriate. In the above embodiment, a plurality of processing units are defined along the depth direction for each ray group. However, it is also possible to proceed with processing by ray groups without defining such processing units. is there. Even in this case, since the locality of the reference can be increased in comparison with the selection of the ray according to the conventional simple raster scan, the probability of being able to refer to the cached data is increased, so that rapid image formation is possible. There is an advantage that can be done. Of course, combining the above-described setting of a plurality of processing units in addition to such conditioning has the advantage that the locality of reference can be further improved and control with high practical value can be realized.

  Note that, in a plurality of processing units arranged in the depth direction, a calculation result is passed for each ray between adjacent processing units. That is, in the past, the voxel calculation has progressed continuously along the depth direction for each ray, but according to the above embodiment, the processing interval is set fragmentarily for each ray, and for each interval. It is possible to improve the locality of the reference as described above by performing the calculation. In the above embodiment, the unit of transfer from the external memory to the 3D image forming processor is a block. However, the same effect as described above can be obtained even when data is individually transferred. .

1 is a block diagram showing a preferred embodiment of an ultrasonic diagnostic apparatus according to the present invention. It is a figure which shows the relationship between a three-dimensional space and a ray set. It is a figure for demonstrating a processing unit. It is a figure for demonstrating the setting of a some ray group. It is a figure for demonstrating the various variation of ray selection order. It is a figure for demonstrating a group selection order. 3 is a flowchart illustrating an operation example of the apparatus illustrated in FIG. 1.

Explanation of symbols

  10 3D probe, 12 transmission / reception unit, 16 3D memory (external memory), 20 3D image forming processor, 22 display processing unit, 24 display unit, 40 three-dimensional (3D) space, 42 ray set, 44 ray group, 48 processing units .

Claims (10)

An external memory for storing data acquired from a three-dimensional space in the living body by transmitting and receiving ultrasonic waves;
Having an internal memory for temporarily storing data transferred from the external memory, and referring to the data on the internal memory, three-dimensionally in the depth direction for each ray set for the three-dimensional space An arithmetic processing unit that repeatedly executes voxel arithmetic for image formation, thereby forming a three-dimensional image,
Including
The arithmetic processing unit includes:
A ray set setting unit for setting a two-dimensional ray set composed of a plurality of rays with respect to the three-dimensional space;
A ray group setting unit that sets a plurality of two-dimensional ray groups that are two-dimensionally aligned with respect to the two- dimensional ray set;
In the process of sequentially selecting a two-dimensional ray group from the ray set in accordance with a predetermined group selection order, a progress control unit that sequentially selects rays to be subjected to voxel calculation according to a predetermined ray selection order for each selected two-dimensional ray group When,
A transfer control unit that obtains the data from the external memory and stores the data in the internal memory when the data required for the voxel operation on the selected ray does not exist in the internal memory;
An ultrasonic diagnostic apparatus comprising: An external memory for storing data acquired from a three-dimensional space in the living body by transmitting and receiving ultrasonic waves;
Having an internal memory for temporarily storing data transferred from the external memory, and referring to the data on the internal memory, three-dimensionally in the depth direction for each ray set for the three-dimensional space An arithmetic processing unit that repeatedly executes voxel arithmetic for image formation, thereby forming a three-dimensional image,
Including
The arithmetic processing unit includes:
A ray set setting unit for setting a two-dimensional ray set composed of a plurality of rays with respect to the three-dimensional space;
A ray group setting unit that sets a plurality of two-dimensional ray groups that are two-dimensionally aligned with respect to the two- dimensional ray set;
A processing unit setting unit that sets a plurality of three-dimensional processing units that are three-dimensionally aligned with respect to the two-dimensional ray set by setting a plurality of sections in the depth direction for each two-dimensional ray group;
In the process for sequentially selecting a two-dimensional ray group from said two-dimensional ray set according to a predetermined group selection order, the three-dimensional processing selected with sequentially selecting the three-dimensional processing unit in the depth direction for each two-dimensional ray group selected A progress control unit that sequentially selects rays to be subjected to voxel calculation in a unit according to a predetermined ray selection order;
A transfer control unit that obtains the data from the external memory and stores the data in the internal memory when the data required for the voxel operation on the selected ray does not exist in the internal memory;
An ultrasonic diagnostic apparatus comprising: The apparatus according to claim 1 or 2,
The arithmetic processing unit further includes:
An interpolation calculation unit that obtains a voxel value of each voxel on the selected ray from a plurality of neighboring data existing in the vicinity thereof by interpolation calculation;
A voxel operation unit that performs a voxel operation using a voxel value of each voxel on the selected ray;
An ultrasonic diagnostic apparatus comprising: The apparatus according to claim 1 or 2,
The ultrasonic diagnostic apparatus according to claim 1, wherein the predetermined ray selection order constitutes a ray selection route corresponding to one stroke. The apparatus of claim 4.
The ultrasonic diagnostic apparatus according to claim 1, wherein the ray selection route has a zigzag form, a spiral form, or a Peano curve form. The apparatus according to claim 1 or 2,
The ultrasonic diagnostic apparatus according to claim 1, wherein the predetermined group selection order constitutes a group selection route corresponding to one stroke. The apparatus of claim 5.
The ultrasonic diagnostic apparatus according to claim 1, wherein the group selection route has a zigzag form, a spiral form, or a Peano curve form. The apparatus of claim 2.
Ultrasonic diagnostic apparatus wherein the three-dimensional processing unit, characterized in that it has a size equivalent to multiple of voxels in the depth direction. The apparatus of claim 8.
Each of the three-dimensional processing units has a size corresponding to a plurality of voxels in the depth direction from the start surface to the stop surface,
An ultrasonic diagnostic apparatus, wherein a voxel calculation proceeds from the start surface to the stop surface for each ray in each three-dimensional processing unit. The apparatus of claim 9.
An ultrasonic diagnostic apparatus, wherein a voxel calculation result is inherited for each ray between two three-dimensional processing units adjacent in the depth direction in each two-dimensional ray group.

Images