CN101520817A - Massive medical image three-dimensional visualization processing system - Google Patents

Abstract

A massive medical image three-dimensional visualization processing system, comprising a sequentially connected medical image import module database, a medical image three-dimensional visualization processing module and a three-dimensional model display module; the medical image three-dimensional visualization processing module includes a client, a server and a hard disk cache , the client is connected to the hard disk buffer through the server, and the hard disk buffer is connected to the client and the three-dimensional model display module. The present invention utilizes the hard disk cache to connect each data processor through the hard disk cache, so that the active image data in the pipeline can be kept to a minimum. The invention can solve the problem of three-dimensional visualization of massive medical image data that cannot be processed at present, so that the general image processing algorithm can support DICOM image processing of more than GB without modification. It is an innovative breakthrough for the small data processing that can only process part of the organs.

Description

A 3D visualization processing system for massive medical images

technical field

The invention relates to a visualization processing system for massive images, in particular to a three-dimensional visualization processing system for massive medical images.

technical background

With the development of digital image technology and multimedia information technology, the application of massive image data is becoming more and more common. For example, in the field of medical imaging, massive image data has been widely used. As the degree of refinement increases, the amount of data that needs to be processed is also increasing. However, corresponding to this is that the increase rate of computer memory in terms of hardware can never catch up with the increase rate of data volume; the design of computer systems in terms of software enables 32-bit software to have a maximum data processing capacity of 3GB. Even if 64-bit design software is used, it is unrealistic to read all the data of massive medical images into memory for processing.

In the field of medical image processing, extracting regions of interest with special meanings in medical images and performing 3D reconstruction is of special significance in medical applications. In the various image processing mentioned above, massive medical image data will be processed directly or indirectly, and corresponding intermediate results will be generated. Therefore, how to scientifically and effectively represent and store medical image data to make it suitable for medical image processing operations becomes a solution. The key to massive medical image processing.

The key to realizing massive image processing is how to solve the data communication between memory and hard disk, make up for the lack of memory space with large hard disk space, and at the same time ensure that the software can still meet user needs under a large amount of data communication. Currently there are four main methods:

Method 1: Allow the program to access memory addresses exceeding 3GB. For example, in the prior art, the medical 3D reconstruction research software VG Studio Max uses a processing model to develop software based on a 64-bit processor. Under a 64-bit processor, the address that the program can access is tens of gigabytes.

Method 2: Use memory image data to process in blocks and summarize the results. Guangdong Weichuang Rixin Electronics Co., Ltd. proposed a method for compressing, storing and displaying massive image data. Patent application number: 200710027386.5. This solution divides massive image data into blocks, reads the block data and stores it in the memory cache, generates layered data of various resolution levels after compression, and saves them as intermediate image files, and then releases the memory cache; When the image is displayed, the intermediate image file stored in the disk is read into the memory buffer, and then copied to the display buffer, and the image within the range of the display area is displayed.

Method 3: Use the method of immediately releasing the memory space for each processing;

Method 4: Use the image size reduction method to reduce the amount of data to be processed.

The above methods all have the following disadvantages: Method 1 requires independent memory management for different operating systems, which is not easy to transplant, is complicated, and cannot make full use of the efficient memory management of existing operating systems. In particular, the medical 3D reconstruction research software VG Studio Max has requirements for processors, memory and other hardware and operating systems, but the current mainstream is still dominated by 32-bit processors, and VG Studio Max cannot process massive image data on 32-bit processors. deal with.

Method 2 needs to carry out corresponding block and merge operations for different existing image processing algorithms, but different image processing algorithms are different, and the workload is heavy. At the same time, some image processing algorithms cannot perform block processing, which means using In this way, some data processing that does not support image segmentation cannot be performed. The patent application number is 200710027386.5, and the patent application titled a mass image data compression, storage and display method is mainly used for image display, and does not involve digital image processing, such as smoothing, segmentation, 3D reconstruction, etc. At the same time, the block method is adopted. As mentioned above, some algorithms that involve three-dimensional volume data processing in similar medical image processing cannot be performed.

Method 3 only frees up space by temporarily releasing data, and does not really solve the problem, and dynamically releasing memory will reduce processing efficiency.

Method 4 is the most commonly used method, but the reduction means that the image is distorted by different procedures, and medical images often require high fineness, this method cannot meet the fine requirements of medical images.

Contents of the invention

The purpose of the present invention is to address the above-mentioned deficiencies in the prior art, to design a mass medical image that can perform three-dimensional visualization processing on massive medical images, and can effectively use memory space, in order to achieve the best balance between memory space and memory processing efficiency. Image 3D visualization processing system.

In order to solve the above-mentioned technical problems, the present invention includes the following technical solutions: a massive medical image three-dimensional visualization processing system, including a picture import module, a database, an image three-dimensional visualization processing module and a three-dimensional model display module connected in sequence; the image three-dimensional visualization processing The module includes a client, a server and a hard disk buffer, the client is connected to the hard disk buffer through the server, and the hard disk buffer is connected to the client and the three-dimensional model display module.

The server includes an image smoothing module, an image segmentation module, and a three-dimensional reconstruction module; the hard disk cache includes a smoothing result hard disk cache, a segmentation result hard disk cache, and a three-dimensional reconstruction result hard disk cache; the image smoothing module, smoothing result hard disk cache, The image segmentation module, the hard disk cache of the segmentation result, the three-dimensional reconstruction module, and the hard disk cache of the three-dimensional reconstruction result are connected in sequence.

Due to the need for a large amount of memory and hard disk cache operations, the present invention uses run-length coding to compress the results of image segmentation to reduce the demand for hard disk caching. A compression unit, a hard disk buffer and a run-length encoding decompression unit.

For the convenience of subsequent processing, the same series of tomographic images are merged into 3D volume data and stored in a single DICOM file format. The image import module includes a file header information reading unit and an image sorting unit connected in sequence.

In order to allow the user to choose whether to reduce the size of the image so that the processing result and efficiency can meet the user's requirements, the image importing module includes an image size scaling selection unit.

The image size scaling selection unit includes a scaling command executor and a file information modifier connected in sequence.

In order to support data management and avoid repeated processing, the database includes a sequentially connected data management unit and data storage unit.

The data management unit includes a record queryer and a record generator connected with the record queryer, and the record generator is connected with the data storage unit.

Compared with the prior art, the present invention has the following advantages: through the present invention, the problem of three-dimensional visualization of massive medical image data that cannot be processed currently can be solved. The general image processing algorithm can support DICOM image processing of more than GB without modification. It is an innovative breakthrough for small data processing that can only process part of the organs.

Description of drawings

Fig. 1 is a schematic flow chart of processing massive image data in the prior art;

Fig. 2 is the massive image data processing model intention of the present invention;

Fig. 3 is the massive image processing flowchart of the present invention;

Fig. 4 is the flow chart of massive medical image data import of the present invention;

Fig. 5 is the database management flowchart of the present invention;

Fig. 6 is a mass image intermediate data format generated by the file header information generation unit of the present invention.

Detailed ways

The present invention is based on the principle of the image data processing pipeline, that is, various image data processing can be abstracted as a data processor, and the data is input and output in the form of results. In this way, we have to regard the task of image data processing as sending data into the pipeline, and connect different processes with input and output methods, and finally we can get the results we want.

Figure 1 is a schematic diagram of the processing flow of massive image data in the prior art. In order to meet the needs of medical image processing, it is necessary to collect raw DICOM data, then obtain volume data through the data import module, and pass the volume data through the image denoising and smoothing module in turn. 1. Processing by the image segmentation module and the three-dimensional reconstruction module, the result obtained enters the three-dimensional model processing module, and the result processed by the three-dimensional model processing module is sent to the model library or displayed by the display module. However, assuming that the original collected DICOM data is 600 pieces, and the size of each picture is 512*512, it is about 400MB. It is close to 2GB, and the general 32-bit program cannot support such a large memory operation. The memory cannot load multiple massive image data at one time. Therefore, by connecting the hard disk cache and connecting each data processor through the hard disk buffer, This keeps the amount of active image data in the pipeline to a minimum. It makes it possible to process massive image data.

Fig. 2 is a schematic diagram of a massive image data processing model of the present invention. The medical image three-dimensional visualization processing module 24 includes a client 25, a server 26 and a hard disk buffer 27, the client 25 is connected to the hard disk buffer 27 through the server 26, and the hard disk buffer 27 is connected to the client and the three-dimensional model display Module 9 is connected. The present invention mainly adopts a method similar to providing services to refine each image processing execution flow:

(1) Different image processing is used as the server 26, and each processing is sent to the corresponding server 26 by the program;

(2) The server 26 receives the command and executes the corresponding image processing request;

(3) order is finished, and service end 26 compresses result and writes hard disk buffer 27, and generates corresponding record, returns client 25;

(4) The client 25 queries the records returned by the server 26, manages the processing results, and executes subsequent image processing operations.

Among them, the main job of the client 25 is to manage the imported data and the operation results generated by the server 26 . The server 26 mainly refines and completes the image processing flow.

Fig. 3 is a flow chart of massive image processing in the present invention. Among them, the pictures in the database are connected sequentially through image smoothing module 3, smoothing result hard disk cache 4, image segmentation module 5, segmentation result hard disk cache 6, 3D reconstruction module 7, 3D reconstruction result hard disk cache 8 and 3D model display module 9. Get the processing result. Therefore, unlike the general image processing process, the memory used here is only the maximum memory required by different processing modules, not the sum of memory required by all modules. Here, the hard disk cache is mainly used to solve the problem of limited memory addressing. For example, the general image processing process is assumed to include image smoothing, segmentation and 3D reconstruction modules. The existing original collected DICOM data is 600 pieces, and the size of each picture is 512*512, which is about 400MB. It is required to include original data, smoothed data, segmentation results, and 3D reconstruction. The memory requirement is greater than 4*400MB, which is close to 2GB. The operating system cannot bear such a large memory access. By adopting hard disk cache and module division, each module is independent and can address 2GB. In this way, the memory required by each module is only data input and data output. If the above is reversed, it is only 2*400MB, which indirectly improves The efficiency with which memory can process data.

Fig. 4 is a flow chart of importing massive medical image data according to the present invention. The present invention realizes the import of a large amount of medical image data through the image import module 1, and the medical image import module 1 includes a sequentially connected file header information read-in unit 14, a picture sorting unit 15 and an image size scaling selection unit 16, the image The size scaling selection unit 16 includes a scaling command executor 17 and a file information modifier 18 connected in sequence.

Image data import is the first step in data processing. Because medical image data is stored in DICOM format, an individual generally contains up to several hundred slice images. To facilitate subsequent processing, the main task of data import is to combine the same series of slice images The images are merged into 3D volume data and saved in a single DICOM file format. Each DICOM file header contains relevant information, such as patient name, series ID, image spatial position, image accuracy, etc. The invention uses DICOM header information to solve the problems of automatically dividing series and sorting fault files. At the same time, the user can choose whether the program should reduce the image size so that the processing results and efficiency can meet the user's requirements, and the reduction process should simultaneously modify the spatial information of the DICOM header file so that the data will not be unreadable due to the change of the image size.

Among them, the file header information read-in unit 14 reads in and analyzes the DICOM header file information (such as patient name, series ID, image spatial position, image accuracy, etc.), and the picture sorting unit 15 uses the DICOM header information to solve automatic division of series and sort faults. file problem.

According to the needs of the user, the image size zoom selection unit 16 zooms the picture according to the user's selection. When receiving the user's order to select the zoom picture, the zoom command executor 17 zooms the picture according to the user's request, and the file information modifier 18 At the same time, modify the spatial information of the DICOM header file, so that the image will not be unreadable due to the size change of the image. After sorting and zooming selected processed images, DICOM volume data is generated and imported into the database.

Fig. 5 is a flow chart of database management in the present invention. Generally speaking, the import and processing of massive data is very time-consuming, and users frequently operate on the data of the same individual. Therefore, how to effectively manage data is the primary problem to be solved in massive data processing. The present invention uses XML files as a simple database to record the locations and processing results of all current massive medical image data, so that the processing can be persisted, that is, the program can automatically record the user's previous operations and avoid repeated processing. In order to support data management, it is necessary to provide relevant interfaces to automatically generate record marks, query record items, insert record items, and delete record items.

The database includes a data management unit 19 and a data storage unit 20 connected in sequence. The data management unit 19 includes a record queryer 21 and a record generator 22 connected to the record queryer 21 , and the record generator 22 is connected to the data storage unit 20 .

The picture processed by the image import module 1 first generates a record mark, and the record query device 21 inquires about the record mark, and if there is data, the operation is ended; if the existing data is not read, the operation is performed, and the operation record item is added to The data storage unit 20, that is, the XML database in the embodiment of the present invention, at the same time, generates the record item of this operation through the record generator 22, which will be queried for the next time the record queryer 21 inquires.

Fig. 6 is a mass image intermediate data format generated by the file header information generation unit of the present invention. Since a large amount of memory and hard disk cache operations are required, the present invention uses run-length coding to save the result of image segmentation to reduce the demand for hard disk buffering. The file format includes two parts, header file information and run-length encoding. In order to better record patient information, header file information includes patient ID, scan ID, file name, creation date, author, program, module, image size, and image space information etc.

The run-length encoding is used to save the result of image segmentation through the segmented result hard disk cache 6, which includes a sequentially connected file header information generation unit, run-length encoding compression unit, hard disk buffer, and run-length encoding decompression unit. Since the results after segmentation are all binary images, that is, only foreground and background colors are included. It is represented by 0 and 1 here, therefore, only 1 can be run-length coded. And improve the original run-length encoding, add relevant DICOM header information, so as to save and reference the relevant DICOM header information in subsequent data processing.

The specific steps are divided into compression process and decompression process. After the image is divided, the header information of the image is first generated by the header file header information generation unit, including patient ID, scan ID, file name, creation date, author, program, module , image size, and image space information, etc. Then the run-length encoding and compression unit traverses the image pixel values, and when it encounters 1, it records the current image subscript (x, y, z), and steps until it encounters a value of 0, and the number of 1s that appear during the recording process is used as the run length, which is recorded as r. and write (x, y, z, r) into the hard disk buffer.

When decompressing, the run-length coding decompression unit first reads the header information of the compressed file in the hard disk buffer to generate the corresponding image size, then reads in the previously saved (x, y, z, r) values in sequence, and subscripts the image as ( x, y, z) and the values of the following r pixels are set to 1.

Claims (8)

1. a large amount of medical image three-dimensional visualization processing system, comprising picture import module (1), database, image three-dimensional visualization processing module (24) and three-dimensional model display module (9) connected successively; described image three-dimensional visualization processing module ( 24) comprise client (25), service end (26) and hard disk cache (27), described client (25) is connected with hard disk cache (27) by server (26), described hard disk cache ( 27) Connect with the client and the three-dimensional model display module (9). 2. The massive medical image three-dimensional visualization processing system according to claim 1, characterized in that: the server (26) includes an image smoothing module (3), an image segmentation module (5), and a three-dimensional reconstruction module (7); The hard disk cache (27) includes a smoothing result hard disk cache (4), a segmentation result hard disk cache (6) and a three-dimensional reconstruction result hard disk cache (8); the image smoothing module (3), the smoothing result hard disk cache (4) , the image segmentation module (5), the segmentation result hard disk cache (6), the three-dimensional reconstruction module (7), and the three-dimensional reconstruction result hard disk cache (8) are sequentially connected. 3. The three-dimensional visualization processing system for massive medical images according to claim 2, characterized in that: the segmentation result hard disk cache (6) includes a sequentially connected file header information generation unit, a run-length encoding compression unit, a hard disk buffer and a run-length Code decompression unit. 4. The three-dimensional visualization processing system for massive medical images according to claim 3, characterized in that: the picture import module (1) includes a sequentially connected file header information reading unit (14) and picture sorting unit (15). 5. The three-dimensional visualization processing system for massive medical images according to claim 4, characterized in that: the image import module (1) includes an image size scaling selection unit (16). 6. The three-dimensional visualization processing system for massive medical images according to claim 5, characterized in that: the image size scaling selection unit (16) includes a scaling command executor (17) and a file information modifier (18) connected in sequence . 7. The three-dimensional visualization processing system for massive medical images according to claim 6, characterized in that: said database includes a data management unit (19) and a data storage unit (20) connected in sequence. 8. The massive medical image three-dimensional visualization processing system according to claim 7, characterized in that: the data management unit (19) includes a record queryer (21), a record generator (22) connected to the record queryer (21) , the record generator (22) is connected to the data storage unit (20).

Images