WO2017193477A1 - 三维医学影像数据处理方法及装置 - Google Patents
三维医学影像数据处理方法及装置 Download PDFInfo
- Publication number
- WO2017193477A1 WO2017193477A1 PCT/CN2016/091706 CN2016091706W WO2017193477A1 WO 2017193477 A1 WO2017193477 A1 WO 2017193477A1 CN 2016091706 W CN2016091706 W CN 2016091706W WO 2017193477 A1 WO2017193477 A1 WO 2017193477A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image data
- medical image
- processing
- dimensional medical
- cpu
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30068—Mammography; Breast
Definitions
- the present invention relates to the field of medical image processing technologies, and in particular, to a method and an apparatus for processing three-dimensional medical image data.
- Medical image processing including but not limited to interpolation, denoising, segmentation, and analysis, is closely related to clinical diagnosis. It is designed to enhance image quality, or to separate areas of interest, or to separate potential disease areas, to provide doctors with better first-hand materials for more focused analysis, judgment and identification, thereby improving clinical diagnostic accuracy. Due to the complexity of image processing algorithms and the lack of acceleration strategies, many image processing algorithm processes cannot meet the real-time analysis requirements of 3D medical images.
- Three-dimensional medical imaging is the digitization of human tissues and organs. As software and hardware continue to upgrade, the scanned organs will be more clear and the amount of data generated will be even larger. For example, a magnetic resonance data of a brain, if the resolution is [256, 256, 256], and the gray value is stored in 8 bytes, the data amount reaches 120 megabytes, which causes many image processing processes to fail to meet the clinical real-time requirements.
- common acceleration strategies are mainly hardware acceleration, software acceleration, and parallel acceleration.
- Hardware acceleration is the use of hardware modules to replace software algorithms to take advantage of the fast features inherent in hardware.
- the drawback is that (1) turning on hardware acceleration may have negative effects; (2) designing specific hardware or components, such as chips or processors, for specific tasks, adding additional hardware and software design, time consumption, or additional expenses.
- Software acceleration is aimed at the intrinsic properties of the algorithm, such as multi-layer loop, parameter optimization, etc., designing the corresponding algorithm flow, avoiding repeated operations in the software implementation to reduce time consumption.
- the drawback is that (1) the speedup ratio is limited, for example, the multi-layer loop of many algorithms cannot be avoided; (2) it needs to go deep into the core of the algorithm, thus performing algorithm redesign and code reconstruction, increasing time consumption, and not having generalization. .
- Parallel acceleration is to make full use of the hardware properties of the machine.
- the existing parallel acceleration generally enhances the parallel processing capability of the machine by directly purchasing a Graphics Processing Unit (GPU).
- GPU Graphics Processing Unit
- Parallel acceleration takes into account the decomposability of algorithms and data, as well as the inherent properties of hardware platforms, and is generally more efficient than independent hardware acceleration or software acceleration.
- current GPU-based parallel acceleration requires redesigning and rewriting the algorithm, as well as additional purchases of GPU hardware.
- Embodiments of the present invention provide a three-dimensional medical image data processing method for processing a three-dimensional medical image in real time and efficiently.
- the method is performed by an image processing apparatus including a plurality of CPUs, the method comprising:
- the image processing device groups the three-dimensional medical image data according to the number of CPUs included, wherein each CPU corresponds to a set of three-dimensional medical image data;
- the image processing apparatus runs a two-dimensional image processing algorithm on each CPU to process each set of three-dimensional medical image data.
- the embodiment of the invention further provides a three-dimensional medical image data processing device for processing a three-dimensional medical image in real time and efficiently, the device comprising a plurality of CPUs, the device comprising:
- a packet processing module configured to group three-dimensional medical image data according to the number of CPUs included in the device, wherein each CPU corresponds to a set of three-dimensional medical image data;
- the algorithm running module is configured to run a two-dimensional image processing algorithm on each CPU to process each group of three-dimensional medical image data.
- the embodiment of the invention further provides a three-dimensional medical image data processing device for processing a three-dimensional medical image in real time and efficiently, the device comprising a plurality of CPUs, wherein at least one CPU is configured according to: a CPU included in the device The number is grouped into three-dimensional medical image data, wherein each CPU corresponds to a set of three-dimensional medical image data;
- each CPU in the device is configured to: run a two-dimensional image processing algorithm, and process the three-dimensional medical image data corresponding to the CPU.
- the embodiment of the present invention does not have a negative effect, and does not need to design special hardware for the task; compared with the software acceleration, the embodiment of the present invention can greatly improve the operation speed after selecting the two-dimensional image processing algorithm; Compared with the GPU-based parallel acceleration, the embodiment of the present invention does not need to redesign and rewrite the algorithm, and does not need to purchase any hardware device.
- the embodiments of the present invention have broad application prospects, do not require additional funds and time expenditure, and do not require extensive rewriting or process design of the algorithm. It can greatly reduce running time consumption on ordinary multi-core CPU machines, and can process 3D medical images in real time and efficiently based on existing machines (hardware) and 2D image processing algorithms (software).
- FIG. 1 is a schematic diagram of a method for processing three-dimensional medical image data according to an embodiment of the present invention
- FIG. 2 is a schematic diagram of processing a three-dimensional medical image data by using a two-dimensional image segmentation algorithm according to an embodiment of the present invention
- 3 is a diagram showing an example of segmentation accuracy of three-dimensional medical image data in an embodiment of the present invention.
- FIG. 4 is a schematic diagram of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention.
- FIG. 5 is a view showing a specific example of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention.
- FIG. 6 is a diagram showing another specific example of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention.
- the embodiment of the invention provides a three-dimensional medical image data processing method.
- This method belongs to the category of parallel acceleration.
- the main features are as follows: (1) The method needs to be arranged on a multi-CPU machine. Multiple CPUs can enhance the parallel processing power of the machine. In many cases, the speed of calculation increases linearly with the number of CPUs. (2) The method processes the three-dimensional medical image data, and the larger the amount of data, the more obvious the acceleration ratio. (3) The method can introduce any two-dimensional image processing algorithm, which is not limited to image segmentation, image interpolation, image denoising, and the like.
- the three-dimensional medical image data processing method of the embodiment of the present invention is executed by an image processing apparatus including a plurality of CPUs. As shown in FIG. 1, the method may include:
- Step 101 The image processing apparatus groups the three-dimensional medical image data according to the number of CPUs included, wherein each CPU corresponds to a set of three-dimensional medical image data;
- Step 102 The image processing apparatus runs a two-dimensional image processing algorithm on each CPU to process each group of three-dimensional medical image data.
- the three-dimensional medical image data processing method is directed to three-dimensional medical image data, and the premise is that the image processing apparatus that executes the method has a plurality of central processing units (CPUs).
- CPUs central processing units
- This requirement is very easy to satisfy in actual life or work, and thus the embodiment of the present invention has a large application range.
- the embodiments of the present invention particularly indicate that the embodiments of the present invention can be integrated in The medical device, the computer, the personal computer, the machine and the like can be implemented on any medical device and a personal computer, and the image processing device can be a real-time efficient use of the existing two-dimensional image processing algorithm. Processing 3D medical image data.
- the image processing apparatus including the plurality of CPUs first groups the three-dimensional medical image data according to the number of CPUs included, wherein each CPU corresponds to a set of three-dimensional medical image data; that is, the subsequent image processing apparatus is in each CPU.
- the two-dimensional image processing algorithm is run on, and each group of three-dimensional medical image data is processed by each CPU.
- the three-dimensional medical image data may be grouped by using a continuous grouping mode or a skipping grouping mode.
- the continuous grouping mode or the skipping grouping mode is only an example. In the specific implementation, those skilled in the art may also adopt other grouping modes according to actual needs.
- the continuous grouping mode may be: the first CPU processing Images, the ith CPU processing Images, and so on;
- the skip grouping mode can be: the first CPU processing Images, the ith CPU processing Images, And so on;
- the size of the three-dimensional medical image data is [m, n, l], and the number of CPUs included in the image processing device is c, and the number of images in each group is ⁇ > indicates the rounding operation.
- the three-dimensional medical image data may be grouped by using a continuous grouping mode or a skipping grouping mode according to the biological structure involved in the three-dimensional medical image data.
- a continuous grouping mode or a skipping grouping mode can be used depending on the biological structure. For example, if the biological structure involved in the three-dimensional medical image data is the brain or the lung, since the size of the start position and the end position of the data acquisition does not change much, the continuous grouping mode or the skip grouping mode may be selected for the three-dimensional medicine.
- the image data is grouped; if the biological structure involved in the three-dimensional medical image data is a breast, since the difference in chest and nipple size is very obvious, the three-dimensional medical image data can be grouped by using the skip grouping mode, which can improve the operation efficiency. .
- the image processing apparatus runs a two-dimensional image processing algorithm on each CPU, and before processing each group of three-dimensional medical image data, the image processing apparatus may further include: performing initialization processing on each group of three-dimensional medical image data; This can include manual interactions and/or incomplete annotations, and the like.
- the image processing apparatus runs a two-dimensional image processing algorithm on each CPU, and after processing each group of three-dimensional medical image data, the image processing apparatus may further include: evaluating the processing result, and storing the processing result if the processing result reaches a target state; If the processing result does not reach the target state, the two-dimensional image processing algorithm is re-run after the initialization process is re-executed, or an image editing operation is performed.
- the achievement of the target state here means that the target image effect is achieved, and the effect can be characterized by some image parameters, and some indicators can be preset, and the image parameters are compared to determine whether the processing result reaches the target state.
- FIG. 2 is a schematic diagram of processing a three-dimensional medical image data by using a two-dimensional image segmentation algorithm in the present example. As shown in FIG. 2, the process may include:
- the software is implemented as Visual Studio 2010, which is accelerated by OpenMP.
- the experimental machine is 8 cores Cores (TM), clocked at 3.7GHz and 8G memory.
- tc is the time required for each tomographic image segmentation
- n is the number of tomographic images.
- the image segmentation accuracy parameter (Dice) is calculated as follows:
- Table 1 compares the manual split time, the split run time of the method of the embodiment of the present invention, and the split run time of the embodiment of the present invention. It is found that the method of the embodiment of the present invention can segment the image with a single resolution of [512, 512] within 0.78 seconds, which only accounts for 1.8% of the manual segmentation time, which is 15.9% of the method not introduced in the embodiment of the present invention. The amplitude increases the segmentation speed and meets the real-time requirements.
- Figure 3 shows the segmentation accuracy of 32 sets of data in this example. Overall, the average accuracy is 90%. Of these, 28 cases exceeded 80%. Since the accuracy of the segmentation result is related to the segmentation algorithm selected, and is independent of the acceleration algorithm proposed by the embodiment of the present invention, the comment on the segmentation algorithm is omitted here.
- a three-dimensional medical image data processing apparatus is also provided in the embodiment of the present invention, as described in the following embodiments. Since the principle of solving the problem of the device is similar to that of the three-dimensional medical image data processing method, the implementation of the device can be referred to the implementation of the three-dimensional medical image data processing method, and the repeated description is not repeated.
- FIG. 4 is a schematic diagram of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention.
- the apparatus includes a plurality of CPUs. As shown in FIG. 4, the apparatus may include:
- the packet processing module 401 is configured to group the three-dimensional medical image data according to the number of CPUs included in the device, where each CPU corresponds to a set of three-dimensional medical image data;
- the algorithm running module 402 is configured to run a two-dimensional image processing algorithm on each CPU to process each group of three-dimensional medical image data.
- the packet processing module 401 can be specifically configured to:
- the three-dimensional medical image data is grouped by a continuous grouping mode or a skip grouping mode, wherein:
- the continuous grouping mode is: the first CPU processing Images, the ith CPU processing Images, and so on;
- the skip grouping mode is: the first CPU processing Images, the ith CPU processing Images, And so on;
- the size of the three-dimensional medical image data is [m, n, l], and the number of CPUs included in the device is c, and the number of images in each group is ⁇ > indicates the rounding operation.
- the packet processing module 401 can be specifically configured to:
- the three-dimensional medical image data is selected to be grouped by using a continuous grouping mode or a skipping grouping mode.
- the packet processing module 401 can be specifically configured to:
- the three-dimensional medical image data is selected to be grouped by using a continuous grouping mode or a skipping grouping mode;
- the three-dimensional medical image data is selected to be grouped by the skip grouping mode.
- FIG. 5 is a specific example of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention. As shown in FIG. 5, the apparatus shown in FIG. 4 may further include:
- the initialization processing module 501 is configured to run a two-dimensional image processing algorithm on each CPU in the algorithm running module 402, and perform initialization processing on each group of three-dimensional medical image data before processing each group of three-dimensional medical image data; the initialization processing includes Manual interaction and / or incomplete annotation.
- FIG. 6 is a schematic diagram of another embodiment of a three-dimensional medical image data processing apparatus according to an embodiment of the present invention. As shown in FIG. 6, the apparatus shown in FIG. 4 may further include:
- the post-processing module 601 is configured to run a two-dimensional image processing algorithm on each CPU in the algorithm running module 501. After processing each group of three-dimensional medical image data, the processing result is evaluated, and if the processing result reaches the target state, the processing result is stored. If the processing result does not reach the target state, re-run the 2D image processing algorithm or perform the image editing operation after re-initializing the processing.
- the apparatus shown in FIG. 4 in the embodiment may also include an initialization processing module 501 and a post-processing module 601.
- the embodiment of the present invention does not have a negative effect with respect to hardware acceleration, and does not need to design special hardware for the task.
- the embodiment of the present invention can be large after selecting the two-dimensional image processing algorithm.
- the amplitude increases the operation speed; compared with the GPU-based parallel acceleration, the embodiment of the present invention does not need to redesign and rewrite the algorithm, and does not need to purchase any hardware device.
- the embodiments of the present invention have broad application prospects, do not require additional funds and time expenditure, and do not require extensive rewriting or process design of the algorithm. It can greatly reduce running time consumption on ordinary multi-core CPU machines, and can process 3D medical images in real time and efficiently based on existing machines (hardware) and 2D image processing algorithms (software).
- embodiments of the present invention can be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or a combination of software and hardware. Moreover, the invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
- computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
- the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
- the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
- These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
- the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Processing Or Creating Images (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
一种三维医学影像数据处理方法及装置，其中该方法由包含多个CPU的图像处理装置执行，该方法包括：图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据（101）；图像处理装置在各CPU上运行二维图像处理算法（102）；最终将处理好的结果存储到相应的三维数据内。采用所述方法和装置可以实时和高效地对三维医学影像进行处理。
Description
本发明涉及医学图像处理技术领域，尤其涉及三维医学影像数据处理方法及装置。
医学图像处理，包括但不限于插值、去噪、分割以及分析，与临床诊断息息相关。它旨在增强图像质量，或将感兴趣区域，或将潜在疾病区域分离出来，便于为医生提供更优质的第一手材料，以进行更专注的分析、判断和识别，从而提高临床诊断精度。受限于图像处理算法的复杂度和加速策略的缺失，目前很多图像处理算法流程还无法满足三维医学影像的实时性分析需求。
三维医学影像是对人体组织器官的数字化。随着软硬件的不断升级，被扫描的器官会更加清晰，产生的数据量也就更加庞大。比如一个大脑的磁共振数据，若分辨率为[256,256,256]，灰度值以8字节进行存储，则其数据量达到120兆字节左右，导致很多图像处理流程无法满足临床的实时性要求。根据加速类型的不同，常见的加速策略主要分硬件加速、软件加速和并行加速。
硬件加速是利用硬件模块来替代软件算法以充分利用硬件所固有的快速特性。其缺陷在于，(1)开启硬件加速可能会带来负面效果；(2)需要为特定任务设计特定的硬件或元件，如芯片或处理器，增加额外软硬件设计、时间消耗或额外经费支出。
软件加速是针对算法的内在性质，如多层循环、参数优化等，设计相应的算法流程，避免在软件实现中重复运行，以降低时间消耗。其缺陷在于，(1)加速比有限，比如很多算法的多层循环无法避免；(2)需要深入到算法核心，从而进行算法重新设计和代码重构，增加时间消耗，而且不具有可推广性。
并行加速是充分利用机器的硬件属性，现有的并行加速一般通过直接购买图形处理器(Graphics Processing Unit，GPU)来增强机器的并行处理能力。并行加速会充分考虑算法和数据的可分解性以及硬件平台的固有属性，一般比独立的硬件加速或软件加速的效率要高。然而，目前基于GPU的并行加速，需要对算法进行重新设计和改写，也需要额外购买GPU硬件设备。
总之，现有的图像处理算法流程无法实时和高效地对三维医学影像进行处理。
发明内容
本发明实施例提供一种三维医学影像数据处理方法，用以实时和高效地对三维医学影像进行处理，该方法由包含多个CPU的图像处理装置执行，该方法包括：
图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；
图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
本发明实施例还提供一种三维医学影像数据处理装置，用以实时和高效地对三维医学影像进行处理，该装置包含多个CPU，该装置包括：
分组处理模块，用于根据该装置所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；
算法运行模块，用于在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
本发明实施例还提供一种三维医学影像数据处理装置，用以实时和高效地对三维医学影像进行处理，该装置包含多个CPU，其中至少一个CPU被配置为：根据该装置所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；
并且，该装置中每一CPU被配置为：运行二维图像处理算法，对该CPU对应组的三维医学影像数据进行处理。
相对于硬件加速，本发明实施例不会有负面效果，不需要为任务设计特殊的硬件；相对于软件加速，本发明实施例在选定二维图像处理算法后，能够大幅度提高运算速度；相对于基于GPU的并行加速，本发明实施例不需要对算法进行重新设计和改写，更不需要购买任何的硬件设备。总之，本发明实施例具有广泛的应用前景，不需要额外经费和时间支出，不需要对算法进行大幅度改写或流程设计。它能够在普通的多核CPU机器上，大幅度降低运行时间消耗，能够在现有的机器(硬件)和二维图像处理算法(软件)基础上，实时和高效地对三维医学影像进行处理。
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本
发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。在附图中：
图1为本发明实施例中三维医学影像数据处理方法的示意图；
图2为本发明实施例中利用二维图像分割算法对三维医学影像数据进行处理的示意图；
图3为本发明实施例中三维医学影像数据的分割精度示例图；
图4为本发明实施例中三维医学影像数据处理装置的示意图；
图5为本发明实施例中三维医学影像数据处理装置的具体实例图；
图6为本发明实施例中三维医学影像数据处理装置的另一具体实例图。
为使本发明实施例的目的、技术方案和优点更加清楚明白，下面结合附图对本发明实施例做进一步详细说明。在此，本发明的示意性实施例及其说明用于解释本发明，但并不作为对本发明的限定。
为了在现有的机器(硬件)和二维图像处理算法(软件)基础上，实时和高效地对三维医学影像进行处理，本发明实施例提供了一种三维医学影像数据处理方法。该方法属于并行加速范畴，主要特点有：(1)该方法需要布置在多CPU的机器上。多CPU可以增强机器的并行处理能力。在许多情况下，计算速度随CPU数量呈线性提高。(2)该方法针对三维医学影像数据进行处理，数据量越大，加速比越明显。(3)该方法可以引入任意的二维图像处理算法，不局限于图像分割、图像插值、图像去噪等。
本发明实施例的三维医学影像数据处理方法由包含多个CPU的图像处理装置执行，如图1所示，该方法可以包括：
步骤101、图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；
步骤102、图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
如上所述，本发明实施例的三维医学影像数据处理方法针对三维医学影像数据，前提要求是执行该方法的图像处理装置有多个中央处理器(Central Processing Unit，CPU)。这个要求在实际生活或工作中非常容易满足，因此本发明实施例具有很大的应用范围。为避免不必要的技术纠纷，本发明实施例特别指出：本发明实施例可以集成在
任何医学设备和个人电脑上，即上述图像处理装置可以是能够实现其功能的医学设备、计算机、个人电脑、机器等装置；本发明实施例可以利用现有的二维图像处理算法，来实时高效的处理三维医学影像数据。
具体实施时，包含多个CPU的图像处理装置先根据所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；也就是说，后续图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理时，每一CPU处理对应组的三维医学影像数据。实施时对三维医学影像数据进行分组可以有多种方式，例如可以采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，此处的连续性分组模式或跳跃式分组模式仅为举例，具体实施时本领域技术人员也可以根据实际需要采用其它的分组模式。
具体的，连续性分组模式可以是：第1个CPU处理第
个图像，第i个CPU处理第
个图像，以此类推；
跳跃式分组模式可以是：第1个CPU处理第
个图像，第i个CPU处理第
个图像，
以此类推；
其中，三维医学影像数据大小为[m,n,l]，图像处理装置所包含的CPU数量为c，每组图像个数为
<·>表示向上取整操作。
具体实施时，可以根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组。根据生物结构的不同，可以采用不同的分组模式。例如，若三维医学影像数据所涉及的生物结构为大脑或肺部，由于数据采集的起始位置和结束位置的大小变化不大，则可以选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组；若三维医学影像数据所涉及的生物结构为乳房，由于其胸部和乳头大小差异非常明显，则可以选择采用跳跃式分组模式对三维医学影像数据进行分组，这样更能够提高运行效率。
此外，实施例中，图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之前，还可以包括：对各组三维医学影像数据进行初始化处理；其中的初始化处理可以包括人工交互和/或不完全标注等。
实施例中，图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之后，还可以包括：对处理结果进行评价，若处理结果达到目标状态则存储处理结果；若处理结果未达到目标状态则在重新进行初始化处理后重新运行二维图像处理算法，或进行图像编辑操作。此处的达到目标状态是指达到目标图像效果，该效果可以由一些图像参数来表征，可以预先设定一些指标，通过比较图像参数来确定处理结果是否达到目标状态。
下面仅以个人电脑以及某二维图像分割算法来验证本发明实施例在一种三维医学影像数据上分割的加速比以及实时性，从而探讨本发明实施例的可行性、有效性和优越性。图2为本例中利用二维图像分割算法对三维医学影像数据进行处理的示意图。如图2所示，处理过程可以包括：
1)读入三维医学影像数据，根据机器CPU的个数进行分组；
2)根据算法需要，进行一定的初始化工作，如人工交互、不完全标注等，然后在各CPU(CPU_1,……,CPU_i,……，CPU_n)运行二维图像分割算法；
3)将分割结果进行显示；若分割结果不理想，则进行后处理操作，如重新人工标记和算法运行，或者是图像编辑操作；若分割结果可行，则写入体数据，并保存。
本例中经过32组临床医学影像数据(三维乳房影像，分辨率为[512,512]，平均断层图像个数为18)进行二维图像分割算法的试验。与手动分割时间，以及没有引入本发明实施例方法的分割时间进行对比，可以发现本发明实施例方法在准确分割三维医学影像的同时，能够大幅度提升时间效率。机器的CPU越多，加速比越高，越能减少时间消耗。
本例中软件实现为Visual Studio 2010，采用OpenMP进行加速实现。实验机器为8核
Cores(TM)，主频3.7GHz，内存8G。
每个断层图像的平均消耗时间(TC)公式如下：
图像分割准确率参数(Dice)计算公式如下：
其中|·|用来统计三维数据内的点个数，G为手动分割的金标准，而S为分割结果。
表1比较了手动分割时间，没有引入本发明实施例方法的分割运行时间，以及引入本发明实施例的分割运行时间。通过分析发现，本发明实施例方法能在0.78秒内对单张分辨率为[512,512]的图像进行分割，仅占手动分割时间的1.8％，是未引入本发明实施例方法的15.9％，大幅度提升了分割速度，能够满足实时性要求。
表1三维数据手动分割和加速后的平均时间消耗
| 手动分割 | 未引入本发明实施例方法 | 引入本发明实施例方法 | |
| 时间消耗(秒) | 43.83 | 4.92 | 0.78 |
图3展示了本例中32组数据的分割精度。整体上来看，平均精度达到90％。其中28例结果超过80％。由于分割结果的精度与所选用的分割算法相关，而与本发明实施例提出的加速算法无关，此处略去对分割算法的评论。
基于同一发明构思，本发明实施例中还提供了一种三维医学影像数据处理装置，如下面的实施例所述。由于该装置解决问题的原理与三维医学影像数据处理方法相似，因此该装置的实施可以参见三维医学影像数据处理方法的实施，重复之处不再赘述。
图4为本发明实施例中三维医学影像数据处理装置的示意图，该装置包含多个CPU，如图4所示，该装置可以包括：
分组处理模块401，用于根据该装置所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；
算法运行模块402，用于在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
具体实施时，分组处理模块401具体可以用于：
采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，其中：
连续性分组模式为：第1个CPU处理第
个图像，第i个CPU处理第
个图像，以此类推；
跳跃式分组模式为：第1个CPU处理第
个图像，第i个CPU处理第
个图像，
以此类推；
其中，三维医学影像数据大小为[m,n,l]，该装置所包含的CPU数量为c，每组图像个数为
<·>表示向上取整操作。
具体实施时，分组处理模块401具体可以用于：
根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组。
具体实施时，分组处理模块401具体可以用于：
在三维医学影像数据所涉及的生物结构为大脑或肺部时，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组；
在三维医学影像数据所涉及的生物结构为乳房时，选择采用跳跃式分组模式对三维医学影像数据进行分组。
图5为本发明实施例中三维医学影像数据处理装置的具体实例图，如图5所示，图4所示装置还可以包括：
初始化处理模块501，用于在算法运行模块402在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之前，对各组三维医学影像数据进行初始化处理；所述初始化处理包括人工交互和/或不完全标注。
图6为本发明实施例中三维医学影像数据处理装置的另一具体实例图，如图6所示，图4所示装置还可以包括：
后处理模块601，用于在算法运行模块501在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之后，对处理结果进行评价，若处理结果达到目标状态则存储处理结果；若处理结果未达到目标状态则在重新进行初始化处理后重新运行二维图像处理算法，或进行图像编辑操作。实施例中图4所示装置还可以同进包括初始化处理模块501和后处理模块601。
综上所述，相对于硬件加速，本发明实施例不会有负面效果，不需要为任务设计特殊的硬件；相对于软件加速，本发明实施例在选定二维图像处理算法后，能够大幅度提高运算速度；相对于基于GPU的并行加速，本发明实施例不需要对算法进行重新设计和改写，更不需要购买任何的硬件设备。总之，本发明实施例具有广泛的应用前景，不需要额外经费和时间支出，不需要对算法进行大幅度改写或流程设计。它能够在普通的多核CPU机器上，大幅度降低运行时间消耗，能够在现有的机器(硬件)和二维图像处理算法(软件)基础上，实时和高效地对三维医学影像进行处理。
本领域内的技术人员应明白，本发明的实施例可提供为方法、系统、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上所述的具体实施例，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施例而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。
Claims (18)
- 一种三维医学影像数据处理方法，其特征在于，该方法由包含多个CPU的图像处理装置执行，该方法包括：图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
- 如权利要求1所述的方法，其特征在于，图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，包括采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，其中：连续性分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，以此类推；
跳跃式分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，
以此类推；
其中，三维医学影像数据大小为[m,n,l]，图像处理装置所包含的CPU数量为c，每组图像个数为<·>表示向上取整操作。
- 如权利要求2所述的方法，其特征在于，图像处理装置根据所包含的CPU数量对三维医学影像数据进行分组，包括根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求3所述的方法，其特征在于，所述根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，包括：若三维医学影像数据所涉及的生物结构为大脑或肺部，则选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组；若三维医学影像数据所涉及的生物结构为乳房，则选择采用跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求1所述的方法，其特征在于，图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之前，还包括：对各组三维医学影像数据进行初始化处理；所述初始化处理包括人工交互和/或不完全标注。
- 如权利要求5所述的方法，其特征在于，图像处理装置在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之后，还包括：对处理结果进行评价，若处理结果达到目标状态则存储处理结果；若处理结果未达到目标状态则在重新进行初始化处理后重新运行二维图像处理算法，或进行图像编辑操作。
- 一种三维医学影像数据处理装置，其特征在于，该装置包含多个CPU，该装置包括：分组处理模块，用于根据该装置所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；算法运行模块，用于在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理。
- 如权利要求7所述的装置，其特征在于，所述分组处理模块具体用于：采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，其中：连续性分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，以此类推；
跳跃式分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，
以此类推；
其中，三维医学影像数据大小为[m,n,l]，该装置所包含的CPU数量为c，每组图像个数为<·>表示向上取整操作。
- 如权利要求8所述的装置，其特征在于，所述分组处理模块具体用于：根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求9所述的装置，其特征在于，所述分组处理模块具体用于：在三维医学影像数据所涉及的生物结构为大脑或肺部时，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组；在三维医学影像数据所涉及的生物结构为乳房时，选择采用跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求7所述的装置，其特征在于，该装置还包括：初始化处理模块，用于在所述算法运行模块在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之前，对各组三维医学影像数据进行初始化处理；所述初始化处理包括人工交互和/或不完全标注。
- 如权利要求11所述的装置，其特征在于，该装置还包括：后处理模块，用于在所述算法运行模块在各CPU上运行二维图像处理算法，对各组三维医学影像数据进行处理之后，对处理结果进行评价，若处理结果达到目标状态则存储处理结果；若处理结果未达到目标状态则在重新进行初始化处理后重新运行二维图像处理算法，或进行图像编辑操作。
- 一种三维医学影像数据处理装置，其特征在于，该装置包含多个CPU，其中至少一个CPU被配置为：根据该装置所包含的CPU数量对三维医学影像数据进行分组，其中每一CPU对应一组三维医学影像数据；并且，该装置中每一CPU被配置为：运行二维图像处理算法，对该CPU对应组的三维医学影像数据进行处理。
- 如权利要求13所述的装置，其特征在于，所述至少一个CPU进一步被配置为：采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组，其中：连续性分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，以此类推；
跳跃式分组模式为：第1个CPU处理第个图像，第i个CPU处理第
个图像，
以此类推；
其中，三维医学影像数据大小为[m,n,l]，该装置所包含的CPU数量为c，每组图像个数为<·>表示向上取整操作。
- 如权利要求14所述的装置，其特征在于，所述至少一个CPU进一步被配置为：根据三维医学影像数据所涉及的生物结构，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求15所述的装置，其特征在于，所述至少一个CPU进一步被配置为：在三维医学影像数据所涉及的生物结构为大脑或肺部时，选择采用连续性分组模式或跳跃式分组模式对三维医学影像数据进行分组；在三维医学影像数据所涉及的生物结构为乳房时，选择采用跳跃式分组模式对三维医学影像数据进行分组。
- 如权利要求13所述的装置，其特征在于，该装置中每一CPU进一步被配置为：在运行二维图像处理算法，对该CPU对应组的三维医学影像数据进行处理之前，对该CPU对应组的三维医学影像数据进行初始化处理；所述初始化处理包括人工交互和/或不完全标注。
- 如权利要求17所述的装置，其特征在于，该装置中每一CPU进一步被配置为：在运行二维图像处理算法，对该CPU对应组的三维医学影像数据进行处理之后，对处理结果进行评价，若处理结果达到目标状态则存储处理结果；若处理结果未达到目标状态则在重新进行初始化处理后重新运行二维图像处理算法，或进行图像编辑操作。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610300185.7 | 2016-05-09 | ||
| CN201610300185.7A CN105957085A (zh) | 2016-05-09 | 2016-05-09 | 三维医学影像数据处理方法及装置 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2017193477A1 true WO2017193477A1 (zh) | 2017-11-16 |
Family
ID=56914560
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2016/091706 WO2017193477A1 (zh) | 2016-05-09 | 2016-07-26 | 三维医学影像数据处理方法及装置 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN105957085A (zh) |
| WO (1) | WO2017193477A1 (zh) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108172275B (zh) * | 2016-12-05 | 2022-02-11 | 北京东软医疗设备有限公司 | 一种医学影像处理方法及装置 |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1691069A (zh) * | 2004-04-20 | 2005-11-02 | 香港中文大学 | 采用多图形处理器加速方格片元过滤的实时体绘制方法 |
| CN1833258A (zh) * | 2003-08-07 | 2006-09-13 | 皇家飞利浦电子股份有限公司 | 图像对象处理 |
| JP2008076231A (ja) * | 2006-09-21 | 2008-04-03 | Bridgestone Corp | タイヤ外観検査装置 |
| CN101201722A (zh) * | 2006-12-13 | 2008-06-18 | 索尼株式会社 | 显示及提供图像的设备、方法及计算机程序、和记录介质 |
| CN101271582A (zh) * | 2008-04-10 | 2008-09-24 | 清华大学 | 基于多视角二维图像并结合sift算法的三维重建方法 |
| CN101401128A (zh) * | 2005-02-07 | 2009-04-01 | 索尼计算机娱乐公司 | 使用图形处理进行粒子操作的方法和装置 |
| CN101849224A (zh) * | 2007-10-20 | 2010-09-29 | 思杰系统有限公司 | 用于远程处理三维图像数据的方法和系统 |
| CN102538709A (zh) * | 2012-01-09 | 2012-07-04 | 黑龙江科技学院 | 一种基于结构光的三维测量系统中利用gpu并行计算的方法 |
| US9245358B2 (en) * | 2014-05-30 | 2016-01-26 | Apple Inc. | Systems and methods for generating refined, high fidelity normal maps for 2D and 3D textures |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1736907A3 (de) * | 2005-06-10 | 2016-07-06 | Siemens Healthcare GmbH | Verbesserung von Messdaten-Erfassung- und Bild-Rekonstruktion bei MR-Bildern |
| CN100543663C (zh) * | 2006-09-11 | 2009-09-23 | 北京大学 | 一种并行光栅图像处理方法及系统 |
| CN103916668A (zh) * | 2013-01-04 | 2014-07-09 | 云联（北京）信息技术有限公司 | 一种图像处理方法及系统 |
| CN105389853B (zh) * | 2015-11-02 | 2018-01-19 | 北京航空航天大学 | 一种基于多gpu的人脑变形仿真方法 |
-
2016
- 2016-05-09 CN CN201610300185.7A patent/CN105957085A/zh active Pending
- 2016-07-26 WO PCT/CN2016/091706 patent/WO2017193477A1/zh active Application Filing
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1833258A (zh) * | 2003-08-07 | 2006-09-13 | 皇家飞利浦电子股份有限公司 | 图像对象处理 |
| CN1691069A (zh) * | 2004-04-20 | 2005-11-02 | 香港中文大学 | 采用多图形处理器加速方格片元过滤的实时体绘制方法 |
| CN101401128A (zh) * | 2005-02-07 | 2009-04-01 | 索尼计算机娱乐公司 | 使用图形处理进行粒子操作的方法和装置 |
| JP2008076231A (ja) * | 2006-09-21 | 2008-04-03 | Bridgestone Corp | タイヤ外観検査装置 |
| CN101201722A (zh) * | 2006-12-13 | 2008-06-18 | 索尼株式会社 | 显示及提供图像的设备、方法及计算机程序、和记录介质 |
| CN101849224A (zh) * | 2007-10-20 | 2010-09-29 | 思杰系统有限公司 | 用于远程处理三维图像数据的方法和系统 |
| CN101271582A (zh) * | 2008-04-10 | 2008-09-24 | 清华大学 | 基于多视角二维图像并结合sift算法的三维重建方法 |
| CN102538709A (zh) * | 2012-01-09 | 2012-07-04 | 黑龙江科技学院 | 一种基于结构光的三维测量系统中利用gpu并行计算的方法 |
| US9245358B2 (en) * | 2014-05-30 | 2016-01-26 | Apple Inc. | Systems and methods for generating refined, high fidelity normal maps for 2D and 3D textures |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105957085A (zh) | 2016-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Fast and low-GPU-memory abdomen CT organ segmentation: the flare challenge | |
| Shamonin et al. | Fast parallel image registration on CPU and GPU for diagnostic classification of Alzheimer's disease | |
| Yang et al. | Intra: 3d intracranial aneurysm dataset for deep learning | |
| Cao et al. | Multi-task neural networks for joint hippocampus segmentation and clinical score regression | |
| Fluck et al. | A survey of medical image registration on graphics hardware | |
| Lee et al. | CUDA optimization strategies for compute-and memory-bound neuroimaging algorithms | |
| Jiménez et al. | A Web platform for the interactive visualization and analysis of the 3D fractal dimension of MRI data | |
| Gulo et al. | Techniques of medical image processing and analysis accelerated by high-performance computing: A systematic literature review | |
| Chen et al. | TarGAN: Target-aware generative adversarial networks for multi-modality medical image translation | |
| Zhuge et al. | Parallel fuzzy connected image segmentation on GPU | |
| Zhang et al. | Brain tumor segmentation from multi-modal MR images via ensembling UNets | |
| Van Aart et al. | CUDA-accelerated geodesic ray-tracing for fiber tracking | |
| Govyadinov et al. | Robust tracing and visualization of heterogeneous microvascular networks | |
| US20100186017A1 (en) | System and method for medical image processing | |
| Liu et al. | MimicNet: mimicking manual delineation of human expert for brain tumor segmentation from multimodal MRIs | |
| US10621720B2 (en) | Deformable registration of magnetic resonance and ultrasound images using biomechanical models | |
| WO2017193477A1 (zh) | 三维医学影像数据处理方法及装置 | |
| JP6735340B2 (ja) | 核医学画像からの生理的集積の自動除去及びｃｔ画像の自動セグメンテーション | |
| Ciżnicki et al. | Efficient isosurface extraction using marching tetrahedra and histogram pyramids on multiple GPUs | |
| Valero-Lara | A GPU approach for accelerating 3D deformable registration (DARTEL) on brain biomedical images | |
| Loewe et al. | Fast construction of voxel-level functional connectivity graphs | |
| Smistad et al. | GPU-based airway segmentation and centerline extraction for image guided bronchoscopy | |
| Zhang et al. | Exploiting hpc resources for the 3d-time series analysis of caries lesion activity | |
| Plishker et al. | Utilizing hierarchical multiprocessing for medical image registration | |
| Plishker et al. | Towards a heterogeneous medical image registration acceleration platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 16901432 Country of ref document: EP Kind code of ref document: A1 |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 16901432 Country of ref document: EP Kind code of ref document: A1 |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 29.03.2019) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 16901432 Country of ref document: EP Kind code of ref document: A1 |