CN102332162A - Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network - Google Patents
Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network Download PDFInfo
- Publication number
- CN102332162A CN102332162A CN201110276173A CN201110276173A CN102332162A CN 102332162 A CN102332162 A CN 102332162A CN 201110276173 A CN201110276173 A CN 201110276173A CN 201110276173 A CN201110276173 A CN 201110276173A CN 102332162 A CN102332162 A CN 102332162A
- Authority
- CN
- China
- Prior art keywords
- interest
- region
- image
- artificial neural
- compression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention relates to a method for automatic recognition and stage compression of medical image regions of interest based on an artificial neural network. Medical image files in a digital diagnostic system are generally larger, and due to the limitation by factors of bandwidth and the like, the transmission speed is low, the effect is not good, and the diagnosis quality is influenced. By the method, medical digital images are subjected to noise elimination, the tissue outline of a human body is recognized, tissue images are subjected to multiple times of overlay operation, image features of the regions of interest are strengthened, feature values are extracted, classification is performed by using an artificial neural network method, the regions of interest and corresponding levels are determined, and tagged image file format (TIFF) images are generated in different compression modes according to different levels of the regions of interest and non-regions of interest. By the method, the medical image files are greatly lessened, the transmission speed is increased, and effective necessary information used for diagnosis and treatment in the images is kept, so the method facilitates the reading of doctors, and can be applied to the digital diagnostic system and a remote medical system, and improve the diagnosis and treatment efficiency and effect.
Description
Technical field
The present invention relates to the disposal route of medical image, be specifically related to a kind of medical image region of interest and discern automatically and scalable compression method based on artificial neural network.
Background technology
The digital health diagnostic system is the important component part that realizes medical institutions' digitizing, networking.Existing distance medical diagnosis system often depends on specific hardware system, and patient information such as case history, medical picture are carried out remote transmission through file or video mode; Often need the doctor to hold a consultation in real time, the doctor who diagnoses is required height through network, and the medical image of patient information as transmitting; Because the medical image file is generally bigger, receives effects limit such as bandwidth, transmission speed is slow; Effect is bad, has also influenced quality of diagnosis.
Present Medical Image Compression technology; Though being arranged, proposition adopt the different compression method to improve transfer efficiency to region of interest (ROI) and non-region-of-interest; But main problem is; Need the doctor to confirm region of interest, receive restrictions such as doctors experience, unit human resources condition and subjective factor, lack the accuracy that fixing standard is weighed region of interest through the man-machine interactively interface.
Summary of the invention
The purpose of this invention is to provide a kind of can to medical image effectively compress and improve the medical image file transfer speed medical image region of interest automatically identification and scalable compression method; Come identification region of interest and classification automatically through image processing method based on artificial neural network; Thereby realize region of interest (ROI) and non-region-of-interest different stage are adopted the different compression method, reach the diagnosis and treatment efficient of raising tele-medicine and the dual purpose of diagnosis and treatment quality.
The technical scheme that the present invention adopted is:
Medical image region of interest based on artificial neural network is discerned and scalable compression method automatically, it is characterized in that:
Realize by following steps:
Step 1: to the medical digital image is that the DICOM image carries out pre-service; Comprise that eliminating ground unrest, window/horizontal adjustment and histogram equalization handles; And utilization is based on the edge detection algorithm of grid search and hough transform; Discern the tissue profile, confirm the boundary of inside of human body tissue and epidermis muscle;
Step 2: through using BPF. and wavelet filter algorithm, characteristics of image is strengthened in the tissue image that step 1 the is split computing that repeatedly superposes, and discerns region of interest; Extract the region of interest eigenwert then, and according to characteristic value information, the using artificial neural networks method is classified, confirm region of interest and corresponding rank;
Step 3: to the Network Transmission effect of requirement, region of interest and non-region of interest that step 2 is confirmed adopt the different compression mode according to different stage according to current network bandwidth condition and doctor.
Pre-service described in the step 1 is meant the image information of the normal society of removing the image border, artificial filling, irrelevant noise data and tissue profile air background in addition.
Eigenwert described in the step 2 is density, form, texture, position and the range information of region of interest image.
In the described step 3; Based on the current network bandwidth condition comprise the broadband, wireless, the dialing network access and current network speed; The doctor comprises that to the requirement of network laser propagation effect number and compress mode that every width of cloth figure comprises region of interest are Fast Compression, suitably compression or precise compression, adopts the different compression mode to generate tiff image by the different stage of region of interest and non-region of interest;
Region of interest to step 3 is confirmed is used LZW coding lossless compress;
The whole figure that comprises region of interest and non-area-of-interest is adopted the JPEG lossy compression method of corresponding different compression factor.
The present invention has the following advantages:
(1) the present invention carries out noise reduction process to the medical image file; And the characteristic of the region of interest image at focus place strengthened, image file is significantly diminished, improved transfer rate; Also kept the effective necessary information that is used for diagnosis and treatment in the image, be convenient to the doctor and read.
(2) patient can be at the community medicine center, MEC or medical image center; Utilize the local facilities image data; And through wired network or wireless network transmissions data to the teleprocessing center, obtain the specialty analysis result, need not that hospital registers, lines up to the down town again.
(3) medical institutions need not buy expensive tele-medicine specialized equipment; Can utilize general calculation machine hardware and network; Can only need a notebook or a TD internet-of-things terminal that has wireless network card in outlying rural area; Just can utilize advanced network technology, obtain the diagnostic result of specialty.And the doctor who implements remote diagnosis need not to drop into the extra time as talk in real time with patient, many people repeat to read the sheet and the consultation of doctors; Through the automatic analysis of intelligent algorithm and the scheduling of network; Can alleviate doctor's workload greatly; Effectively reduce medical operating cost, and promoted the medical skill service level.
(4) existing health check-up has often limited the effect of its prophylactic treatment owing to lack the expert of accurate otherwise effective technique and association area.Present technique will help promoting the reduction of medical examination cost, popularize conventional gonosome inspection.Along with the raising of medical science remote Diagnosis Technology, annual routine physical examination will become the important means of prophylactic treatment, eliminate for a lot of major diseases to provide safeguard in early days, realize the advanced theory of prophylactic treatment, improve the masses' healthy living level.
Description of drawings
Fig. 1 is a flow chart of steps of the present invention.
Fig. 2 is 4 BPF.s with Gaussian convolution operator formation of various criterion difference.
Fig. 3 is the algorithm design figure of artificial neural network backpropagation algorithm.
Fig. 4 is pending image original.
Fig. 5 is the histogram curve of pending image original.
Fig. 6 is the striograph after the optimization process.
Fig. 7 is the histogram curve of the striograph after the optimization process.
Fig. 8 is the image after the processed compressed, and A is whole figure lossy compression method, and region of interest identifies with square frame; B, C are lossless compress, and B is doubtful lump, and C is a micro-calcification clusters.
Embodiment
Below in conjunction with embodiment the present invention is carried out detailed explanation.
As shown in Figure 1, the compression method of image file in the numerical dialing system medical image transmission course involved in the present invention, realize by following steps:
Step 1: pre-service:
Gather the medical digital image in client, promptly the DICOM image carries out pre-service to this image file, comprises the background elimination, window/level correction and histogram equalization and removal noise and irrelevant information.
Image background comprises because collimating apparatus is blocked the normal society at the X-ray film edge cause, artificial annotate zone outside information and the patient contour.Eliminating background and will help to strengthen visual quality for images, is that follow-up processing reduces workload simultaneously.
In order to ensure not wiping the relevant image error of diagnosis, adopt the dynamic thresholding method of analyzing based on image histogram, promptly on histogram, confirm two peak values, then the lowest part of frequency is made as between the two
T 1
Window with a fixed size partly reaches scanning on every side in picture centre again, and the window that picture centre partly is higher than mean value should calculate mean value and variance in the exit window in tissue, and the frequency values of choosing one-sided 0.95 degree of belief then is made as
T 2
The minimum value of choosing in the two is come separating background as threshold value:
Min{T 1 ,
T 2 .
This method is more safer than traditional image histogram analysis method; (region of interest comprises the tissue and the fat deposit of bulk in worst case; But its edge is very fuzzy) under, because the light intensity of fat deposit generally is lower than its hetero-organization, may also separate the adipose tissue that is lower than 10%; But consider that fat deposit often is not that pathology produces the position, so be safe.
Window/level correction is used to regulate picture contrast and display quality.Window is meant the distribution range of gray-scale value, and the window value reduces then to increase the contrast of image.Level value is window scope center at interval.Automatically find out histogrammic maximal value of entire image and minimum value through computing machine, suppress irrelevant gray-scale value, generate the default look-up table that is used for the image demonstration, thereby improve the display quality of image with window and level.
Histogram equalization is the gradation of image information distribution to be got even as far as possible, through analysis image intensity and light step, from the histogram calculation normalization cumulative histogram of image, thereby generates the image that image detail is obviously strengthened.
Hough transform (Hough Transform) algorithm is to utilize image overall Characteristics Detection objective contour, and edge pixel is connected to form a kind of common method of closed boundary, and its advantage is that the influence that be interrupted by noise and curve is less.In order to check the boundary of human body interior tissue and epidermis muscle, two kinds of methods can be arranged: grid search and curve vector method.The grid method marks boundary information sometimes in some cases by error, and is big but the shortcoming of curve vector method is a calculated amount.In order to satisfy system to response time and performance demands, adopted the search of comprehensive grid and based on the edge detection algorithm of hough transform, realize confirming rapidly and accurately the border of tissue and epidermis, generate and organize mask, thereby isolate the tissue image.
Step 2: identification region of interest:
The tissue image that step 1 is split carries out graphical analysis, judges region of interest (ROI).
BPF. (band-pass filters) algorithm carries out computing through the pixel grey scale with a field around the pending pixel with corresponding BPF. element and improves signal to noise ratio (S/N ratio) and strengthen image.What Fig. 2 showed is the BPF. that has Gaussian convolution operator (Gaussian kernels) formation of various criterion poor (standard deviation) by 4, is used to strengthen central tissue's signal.
The image border is the point of discontinuity of gray scale on the image or the place of gray scale acute variation, and in real image, because the existence of noise, rim detection becomes a difficult problem.Classical edge detection method such as Roberts operator, the Sobel operator, the Prewitt operator, Log operator and Canny operator all have weak point, under some concrete condition, can not detect the best edge of object.Through research, adopt the Edge-Detection Algorithm of using two-dimensional discrete wavelet conversion, at first to discrete wavelet transform coefficients
W
m
1
t(n
1
?+?s,?n
2
?+?s)
And
W m 2 T (n 1 + s, n 2 + s)
Do the enhancing conversion,, can do the nonlinear contrast degree to image and strengthen, when suppressing noise, higher edge precision is provided then to adjusted coefficient reconstruction.
With the result of the above two kinds of algorithm process computing that superposes, thereby region of interest (ROI, Region of Interest) characteristic is strengthened in identification; Again to the image of region of interest, extract density, form, texture, position, image intensity average and standard deviation that eigenwert comprises image, with information such as the number of epidermis distance, spicule and size.
Characteristic value information with the above region of interest is input to artificial neural network, confirms region of interest.Artificial neural network is exactly that interactional dynamic process of while between the neuron is passed through in anthropomorphic dummy's thinking, is a nonlinear dynamic system, and its characteristic is the distributed storage and the concurrent collaborative processing of information.The most ripe at present, use the most widely artificial neural network " backpropagation algorithm (Back Propagation; BP) "; Artificial neural network backpropagation algorithm (Back Propagation; BP) Algorithm design mainly comprises the several aspects of transition function between input layer, latent layer, output layer and each layer, sees Fig. 3._
1), the network number of plies.The BP network can comprise different latent layers, and is verified in theory: have latent layer of network that adds a linear output layer of deviation and at least one S type, can approach any rational function.Latent experience of counting layer by layer is chosen as: generally can adopt perceptron or adaptive network to solve for linear problem, and not adopt nonlinear network, because individual layer can not have been given play to the speciality of non-linear activation function; Nonlinear problem; Two-layer or the two-layer above latent layer of general employing; But the raising of error precision in fact also can obtain through the neuron number that increases in the latent layer; Its training effect is also observed more easily and is adjusted than increasing the number of plies, so generally speaking, should pay the utmost attention to the neuron number that increases in the latent layer.2), the node number of the node number of input layer and output layer.Input layer plays memory buffer, and it receives outside input data, so its node number depends on the dimension of input vector
m , corresponding input layer also has the neuron of different numbers.The node number of output layer depends on two aspects, output data type and the required size of data of expression the type.Because the output sample of prediction is the output vector of 2 dimensions is region of interest type (like lump or micro-calcification clusters) and (between 0 and 1,0 expression is not a pathology to rank, and 1 representes to make a definite diagnosis pathology; Rank is high more, and the possibility of pathology is big more), therefore, output layer has 2 neurons._
3), the node number of hidden neuron.The node number of hidden neuron is confirmed through different neuron numbers being trained contrast, some surpluses of suitable then increase.Rule of thumb, design with reference to following formula:
In the formula:
n Be the number of hidden nodes;
n i Be input number of nodes;
n 0 Be the output node number;
a It is the constant between 1～10.4), transition function.The BP network structure all is to adopt S type activation function at latent layer generally speaking, and output layer adopts linear activation function._ 5), training method chooses.Adopting the additional momentum method to make backpropagation reduce network is absorbed in low ebb on the error surface possibility helps to reduce the training time.Too big learning rate causes the instability learnt, too little value to cause the extremely long training time again.Adaptive learning speed has reached rational two-forty through under the prerequisite that guarantees stable training, can reduce the training time.6), the experience of learning rate is selected.Tend to choose less learning rate generally speaking to guarantee the stability of system, the scope of choosing of learning rate is recommended between 0.01～0.8.
7), anticipation error chooses.If the target output value of sample does
T j , error function then
E For:
E=∑(O
j-T
j)
2
Connecting weight is modified to by following formula:
ΔW
ij(n+1)=ηδ
jO
j+αΔW
ij(n)
Wherein: Δ W
Ij(n)=∑ Δ pW
IjBe sample total
Generally speaking, as a comparison, can train the network of two different expected error value simultaneously, confirm one of them network through the consideration of composite factor at last.In sum: the BP network topology structure does
m *
n * 2 structure.The transport function of hidden neuron is a S type tan
Tansig Because the output rank is normalized in the interval ［ 0,1 ］, the method for normalizing of employing is:
The neuronic transport function of output layer adopts
Purelin Function.The training function of network adopts the weights and the threshold values function of the variable momentum BP algorithm correction neural network of learning rate
Traingdx _
Use the backpropagation algorithm neural network, need not the mathematical model of the system that sets up, as long as there are enough training samples (being obtained by experimental data or emulated data) can confirm region of interest and corresponding rank, rank is high more, possibly be lesion region more.This method is not to judge according to the experience to problem, thereby has adaptation function, has overcome the deficiency of regression methods analysis small sample data.
Through overtesting, the Application of BP neural network has obtained effect preferably in the region of interest identification of breast X-ray image.Collected 84 mammary gland cases through the internet from the database of southern good fortune Flo-Rida-Low university; 50 examples have been used training network; 34 examples are as test set; The result shows that actual interest district (suspicious focus) and prediction region of interest difference do not have statistical significance, and the related coefficient of predicted value and actual value is 0.9162.
Step 3: generate images:
Comprise that based on the current network bandwidth condition network speed and network access are broadband, wireless, dialing; With the doctor to the requirement of network laser propagation effect be Fast Compression, the region of interest number that suitably comprises at most among compression or precise compression and the every width of cloth figure, adopt the different compression mode to generate tiff image by the different stage of region of interest and non-region of interest: the region of interest rank based on step 2 is confirmed selects the high region of interest of rank to use LZW (Lempel-Ziv & Welch) coding lossless compress; Whole figure to comprising region of interest and non-area-of-interest adopts the JPEG lossy compression method.The different stage of pressing region of interest and non-region of interest like this adopts different compression mode and compression factor to generate tiff image, reaches the requirement on both side to diagnostic imaging effect and remote transmission efficient.
TIFF (full name Tagged Image File Format) image, file extent TIF by name or TIFF are to use industry standard bitmap file form the most widely.GIFf format can be stored single raster image by any color depth, also becomes bit depth (like 32 bit images) de facto standards picture format, is widely used in image processing program.GIFf format is supported multiple compress modes such as RAW, RLE, LZW, JPEG, CCITT3 group and 4 groups.
Wherein, Compression method (LZW) is non-loss property (data of image does not reduce, and promptly information can not lost in processing procedure), with the abbreviation (Lempel of three inventor's names; Ziv; Welch) name, it is right that its principle is that the value with each byte all will be made into a character with the value of next byte, and for each character to setting a code.When a same character when occurring once again, just replace this character right, and then match with this code name and next character with code name.A key character of LZW coding principle is that code not only can replace a string data with value, also can replace the data of a string different value.In view data,, also can find a code name to replace these serial datas if there are the data of some different value often to repeat to occur.Compression is reversible with dispose procedure, and all information all keep.LZW lossless compress mode can produce 2:1 or better ratio of compression, thereby source document is subdued to over half.
The JPEG lossy compression method has been utilized the characteristic of people's visual angle system, uses to quantize and lossless compression-encoding the combine redundant information of removing the visual angle and the redundant information of data itself, can significantly reduce image file size in the maintenance high-quality while of image.Use this lossy compression method method, be generally the ratio of compression of 20:1,, then can reach 60:1 or higher ratio of compression for the 8bit grayscale image for coloured image.
The advantage of tiff format is to support the high resolving power color simultaneously, and it is divided into bulk to the different piece of piece image, or perhaps data block.For each bulk portion, all preserved a sign, wherein provide bulk to look like which type of information.Block advantage is to support the software package of tiff format only need preserve the current that part of image that is presented on the screen.And also be not kept on the hard disk ability graftabl when needing by the time in images displayed part on the screen.When very large high-definition picture of editor, this characteristic is just very important.It is pointed out that in recent years the continuous development along with the Image Compression standard, tiff format is also supported the new JPEG2000 standard of releasing, can be in a width of cloth figure compatible lossless compress and lossy compression method, and support encoding region of interest.Like this, just can use JPEG2000 region of interest is adopted lossless compression-encoding, the lossy compression method coding is adopted in other zones, can realize to exceed the compressibility about 30% than JPEG.
Owing to removed DICOM markup information, noise, background and other irrelevant informations; And the Applied Digital image processing techniques is to the computing that repeatedly superposes of the tissue image curve in the specific region; Change numerical information into visual information, naked eyes can't see at all before making information and minute lesion can clearly manifest comprehensively.In addition, regulate window/level automatically to the best visual effect, optimized peripheral image,, want the image transmitted data volume, improved the readability of image thereby significantly reduced like skin and muscle layer image based on the characteristic of the region of interest of discerning.
The original DICOM image (seeing accompanying drawing two) of one width of cloth 26.6M; The tiff image size that generates after treatment is merely 523K; The compressed image file size is merely 1/50th of source document, and image space is saved and reached 98.4%, and picture quality is read than being more suitable for the doctor before handling.Even the file of size in the backcountry that does not have optical fiber, also can carry out remote transmission through mobile phone or telephone modem easily like this, for carrying out Network Transmission and remote diagnosis very advantageous conditions is provided.
With mammary X-ray digitizing remote diagnosis system is example, and Telemedicine System is made up of several parts such as community clinic or health check-up service centre, cloud processing enter, hospital diagnosis workstations.After the patient of community film making; After the pre-service of X line image process, ROI reinforcement and the processed compressed; With other information of patient; Can upload or wired cloud processing enter that is networked to via cdma wireless together like electronic health record etc., patient information can pass through ordering, encrypt and stack suspicious lesions information, and the consultation of doctors of informing network hospital expert.The expert uses the hospital diagnosis workstation to insert the cloud processing enter, accomplishes that diagnosis is identified and report generates, and passes Community Service Center back, if desired, also can provide advice, services such as audio frequency and video conversation.
Be the image comparison before and after handling like Fig. 4-8.
Specific operation process comprises:
1. community medicine outpatient service/MEC: the patient carries out the film making of breast molybdenum target X-ray through reservation, and the digitized picture of generation uploads to the cloud computing processing enter automatically through after pre-service, optimizing reinforcement, compression and encryption.
2, cloud computing processing enter: after the image behind the compress-encrypt passes to the cloud processing enter; Can be according to the analysis in like manner of anatomical knowledge and different views; Use multiple classifying and analyzing method comprise the sorting technique of artificial neural network algorithm, rule-based (rule), based on the sorting technique of discrimination threshold (threshold); Use expert system to remove wrong report false positive data (False Positive), supply decision analysis with the testing result that optimization is provided.According to demand to the image of uploading sort, after the decryption processing, application of expert system is screened suspicious focus, and the diagnosis of informing network hospital expert
*Simultaneously for each hospital database provides communication, inquiry service, and the monitoring management of responsible workflow.
3. hospital backstage service management system: use medical diagnosis workstation and case image data base etc. by hospital expert, the testing result of comparison expert system provides the last diagnostic suggestion.
4. cloud computing processing enter: the last diagnostic suggestion according to the result and the hospital of expert system uploads generates diagnosis report, and passes community medicine outpatient service/MEC back.
5. community medicine outpatient service/MEC: the diagnosis report that can print return; And on the image of diagnosis report mark suspicious lesions position and relevant information; Notify patient to come to obtain the result with modes such as SMS or Emails, or directly post the address that patient stays.
If do not find suspicious focus through second step, can be without the 3rd step, and directly return to client.
Claims (4)
1. discern automatically and scalable compression method based on the medical image region of interest of artificial neural network, it is characterized in that:
Realize by following steps:
Step 1: to the medical digital image is that the DICOM image carries out pre-service; Comprise that eliminating ground unrest, window/horizontal adjustment and histogram equalization handles; And utilization is based on the edge detection algorithm of grid search and hough transform; Discern the tissue profile, confirm the boundary of inside of human body tissue and epidermis muscle;
Step 2: through using BPF. and wavelet filter algorithm, characteristics of image is strengthened in the tissue image that step 1 the is split computing that repeatedly superposes, and discerns region of interest; Extract the region of interest eigenwert then, and according to characteristic value information, the using artificial neural networks method is classified, confirm region of interest and corresponding rank;
Step 3: to the Network Transmission effect of requirement, region of interest and non-region of interest that step 2 is confirmed adopt the different compression mode according to different stage according to current network bandwidth condition and doctor.
2. the medical image region of interest based on artificial neural network according to claim 1 is discerned and scalable compression method automatically, it is characterized in that:
Pre-service described in the step 1 is meant the image information of the normal society of removing the image border, artificial filling, irrelevant noise data and tissue profile air background in addition.
3. the medical image region of interest based on artificial neural network according to claim 1 and 2 is discerned and scalable compression method automatically, it is characterized in that:
Eigenwert described in the step 2 is density, form, texture, position and the range information of region of interest image.
4. the medical image region of interest based on artificial neural network according to claim 3 is discerned and scalable compression method automatically, it is characterized in that:
In the described step 3; Based on the current network bandwidth condition comprise the broadband, wireless, the dialing network access and current network speed; The doctor comprises that to the requirement of network laser propagation effect number and compress mode that every width of cloth figure comprises region of interest are Fast Compression, suitably compression or precise compression, adopts the different compression mode to generate tiff image by the different stage of region of interest and non-region of interest;
Region of interest to step 3 is confirmed is used LZW coding lossless compress;
The whole figure that comprises region of interest and non-area-of-interest is adopted the JPEG lossy compression method of corresponding different compression factor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110276173A CN102332162A (en) | 2011-09-19 | 2011-09-19 | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110276173A CN102332162A (en) | 2011-09-19 | 2011-09-19 | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102332162A true CN102332162A (en) | 2012-01-25 |
Family
ID=45483925
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110276173A Pending CN102332162A (en) | 2011-09-19 | 2011-09-19 | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102332162A (en) |
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324981A (en) * | 2013-05-20 | 2013-09-25 | 浙江大学 | Chemoradiotherapy standardized quality control quantitative method based on neural network |
| CN103489003A (en) * | 2013-09-29 | 2014-01-01 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| CN103500439A (en) * | 2013-09-03 | 2014-01-08 | 西安理工大学 | Image printing method based on image processing technology |
| CN103870688A (en) * | 2014-03-10 | 2014-06-18 | 青岛大学医学院附属医院 | Head and neck superficial disease preliminary screening image similarity comparison method under mobile internet environment |
| WO2015100910A1 (en) * | 2013-12-31 | 2015-07-09 | 深圳迈瑞生物医疗电子股份有限公司 | Method, system and medical device for compressing physiological signal |
| CN104952044A (en) * | 2014-03-28 | 2015-09-30 | 上海联影医疗科技有限公司 | Mammary X-ray image enhancement method |
| CN105709302A (en) * | 2016-04-21 | 2016-06-29 | 青岛大学 | Medical-care system with infusion monitor |
| CN105930223A (en) * | 2016-04-24 | 2016-09-07 | 湖南大学 | Method for reducing size of check point file |
| CN105975740A (en) * | 2016-04-21 | 2016-09-28 | 寇玮蔚 | Medical system with intelligent diagnosis function |
| CN105975741A (en) * | 2016-04-21 | 2016-09-28 | 张明飞 | Medical system with patient state grading function |
| CN105975742A (en) * | 2016-04-21 | 2016-09-28 | 张明飞 | Comprehensive medical system with family care function |
| CN106228185A (en) * | 2016-07-20 | 2016-12-14 | 武汉盈力科技有限公司 | A kind of general image classifying and identifying system based on neutral net and method |
| CN106339571A (en) * | 2015-07-10 | 2017-01-18 | 西门子保健有限责任公司 | Artificial Neural Network And A Method For The Classification Of Medical Image Data Records |
| CN106341664A (en) * | 2016-09-29 | 2017-01-18 | 浙江宇视科技有限公司 | Data processing method and apparatus |
| CN106340024A (en) * | 2016-08-30 | 2017-01-18 | 厦门美图之家科技有限公司 | Image segmentation method and application and computing device |
| CN106529116A (en) * | 2015-09-11 | 2017-03-22 | 西门子保健有限责任公司 | Diagnostic system and diagnostic method |
| CN106713929A (en) * | 2017-02-16 | 2017-05-24 | 清华大学深圳研究生院 | Video interframe prediction enhancement method based on deep neural network |
| CN107610773A (en) * | 2017-09-12 | 2018-01-19 | 北京即刻叁维数据科技股份有限公司 | A kind of vascular dissection aided diagnosis method based on sustainer medical image |
| CN108230219A (en) * | 2016-12-09 | 2018-06-29 | 中国科学院深圳先进技术研究院 | Enhance the method and system of encrypted domain jpeg image contrast |
| CN108464840A (en) * | 2017-12-26 | 2018-08-31 | 安徽科大讯飞医疗信息技术有限公司 | A kind of breast lump automatic testing method and system |
| CN109284530A (en) * | 2018-08-02 | 2019-01-29 | 西北工业大学 | Space non-cooperative target appearance rail integration method for parameter estimation based on deep learning |
| CN109509197A (en) * | 2018-09-26 | 2019-03-22 | 沈阳东软医疗系统有限公司 | A kind of method, apparatus, equipment and storage medium for dividing area-of-interest |
| CN110222772A (en) * | 2019-06-10 | 2019-09-10 | 浙江大学 | A kind of medical image mark recommended method based on block rank Active Learning |
| CN110473186A (en) * | 2019-02-14 | 2019-11-19 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN110520909A (en) * | 2017-04-17 | 2019-11-29 | 微软技术许可有限责任公司 | The neural network processor of bandwidth of memory utilization rate is reduced using the compression and decompression of activation data |
| CN110766644A (en) * | 2018-07-27 | 2020-02-07 | 杭州海康威视数字技术股份有限公司 | Image down-sampling method and device |
| CN110856035A (en) * | 2018-07-24 | 2020-02-28 | 顶级公司 | Processing image data to perform object detection |
| CN110876062A (en) * | 2018-08-31 | 2020-03-10 | 三星电子株式会社 | Electronic device for high-speed compression processing of feature map and control method thereof |
| CN111583292A (en) * | 2020-05-11 | 2020-08-25 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111654660A (en) * | 2020-06-08 | 2020-09-11 | 上海纽菲斯信息科技有限公司 | Video conference system coding transmission method based on image segmentation |
| CN111841017A (en) * | 2020-05-29 | 2020-10-30 | 北京编程猫科技有限公司 | Game AI programming realization method and device |
| CN112401829A (en) * | 2020-11-26 | 2021-02-26 | 温州眼视光国际创新中心 | Remote intelligent control system of slit lamp |
| CN112616054A (en) * | 2020-12-11 | 2021-04-06 | 北京林业大学 | Self-adaptive compression transmission and recovery method and device for wild animal monitoring image |
| CN112616040A (en) * | 2020-12-11 | 2021-04-06 | 北京林业大学 | Wild animal image transmission method and system based on distributed architecture |
| CN113038118A (en) * | 2020-12-08 | 2021-06-25 | 重庆邮电大学 | Video QoE prediction system and method based on big data environment |
| CN113538253A (en) * | 2020-04-21 | 2021-10-22 | 杭州普健医疗科技有限公司 | Medical image processing method, medium and CT device |
| CN113935913A (en) * | 2021-10-08 | 2022-01-14 | 北京计算机技术及应用研究所 | Black box image confrontation sample generation method with visual perception concealment |
| WO2022067656A1 (en) * | 2020-09-30 | 2022-04-07 | 华为技术有限公司 | Image processing method and apparatus |
| WO2023279964A1 (en) * | 2021-07-08 | 2023-01-12 | 华为技术有限公司 | Data compression method and apparatus, and computing device and storage medium |
| CN116436987A (en) * | 2023-06-12 | 2023-07-14 | 深圳舜昌自动化控制技术有限公司 | IO-Link master station data message transmission processing method and system |
| CN117115045A (en) * | 2023-10-24 | 2023-11-24 | 吉林大学 | Method for improving medical image data quality based on Internet generation type artificial intelligence |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1819621A (en) * | 2006-01-25 | 2006-08-16 | 杭州维科软件工程有限责任公司 | Medical image enhancing processing method |
| CN101639936A (en) * | 2009-04-28 | 2010-02-03 | 北京捷科惠康科技有限公司 | X-ray image enhancing method and system thereof |
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| US20100246924A1 (en) * | 2009-03-31 | 2010-09-30 | Fujifilm Corporation | Mammary gland content rate estimating apparatus, method and recording medium |
| CN101908891A (en) * | 2010-08-23 | 2010-12-08 | 南京信息工程大学 | Medical image ROI (Region of Interest) compression method based on lifting wavelet and PCNN (Pulse Coupled Neural Network) |
| CN102156715A (en) * | 2011-03-23 | 2011-08-17 | 中国科学院上海技术物理研究所 | Retrieval system based on multi-lesion region characteristic and oriented to medical image database |
-
2011
- 2011-09-19 CN CN201110276173A patent/CN102332162A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1819621A (en) * | 2006-01-25 | 2006-08-16 | 杭州维科软件工程有限责任公司 | Medical image enhancing processing method |
| US20100246924A1 (en) * | 2009-03-31 | 2010-09-30 | Fujifilm Corporation | Mammary gland content rate estimating apparatus, method and recording medium |
| CN101639936A (en) * | 2009-04-28 | 2010-02-03 | 北京捷科惠康科技有限公司 | X-ray image enhancing method and system thereof |
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| CN101908891A (en) * | 2010-08-23 | 2010-12-08 | 南京信息工程大学 | Medical image ROI (Region of Interest) compression method based on lifting wavelet and PCNN (Pulse Coupled Neural Network) |
| CN102156715A (en) * | 2011-03-23 | 2011-08-17 | 中国科学院上海技术物理研究所 | Retrieval system based on multi-lesion region characteristic and oriented to medical image database |
Non-Patent Citations (1)
| Title |
|---|
| 魏本征: "医学图像分割与压缩算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (70)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324981A (en) * | 2013-05-20 | 2013-09-25 | 浙江大学 | Chemoradiotherapy standardized quality control quantitative method based on neural network |
| CN103324981B (en) * | 2013-05-20 | 2015-12-02 | 浙江大学 | Based on the quantization method that the chemicotherapy normalized quality of neural network controls |
| CN103500439B (en) * | 2013-09-03 | 2016-04-20 | 西安理工大学 | Drawing method is beaten based on image processing techniques |
| CN103500439A (en) * | 2013-09-03 | 2014-01-08 | 西安理工大学 | Image printing method based on image processing technology |
| CN103489003A (en) * | 2013-09-29 | 2014-01-01 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| CN103489003B (en) * | 2013-09-29 | 2017-04-19 | 华南理工大学 | Mobile phone image annotation method based on cloud computing |
| WO2015100910A1 (en) * | 2013-12-31 | 2015-07-09 | 深圳迈瑞生物医疗电子股份有限公司 | Method, system and medical device for compressing physiological signal |
| CN103870688A (en) * | 2014-03-10 | 2014-06-18 | 青岛大学医学院附属医院 | Head and neck superficial disease preliminary screening image similarity comparison method under mobile internet environment |
| CN104952044A (en) * | 2014-03-28 | 2015-09-30 | 上海联影医疗科技有限公司 | Mammary X-ray image enhancement method |
| CN106339571B (en) * | 2015-07-10 | 2020-09-18 | 西门子保健有限责任公司 | Artificial neural network for classifying medical image data sets |
| CN106339571A (en) * | 2015-07-10 | 2017-01-18 | 西门子保健有限责任公司 | Artificial Neural Network And A Method For The Classification Of Medical Image Data Records |
| CN106529116B (en) * | 2015-09-11 | 2020-02-11 | 西门子保健有限责任公司 | Diagnostic system |
| US10176580B2 (en) | 2015-09-11 | 2019-01-08 | Siemens Healthcare Gmbh | Diagnostic system and diagnostic method |
| CN106529116A (en) * | 2015-09-11 | 2017-03-22 | 西门子保健有限责任公司 | Diagnostic system and diagnostic method |
| CN105709302A (en) * | 2016-04-21 | 2016-06-29 | 青岛大学 | Medical-care system with infusion monitor |
| CN105975740A (en) * | 2016-04-21 | 2016-09-28 | 寇玮蔚 | Medical system with intelligent diagnosis function |
| CN105709302B (en) * | 2016-04-21 | 2018-11-09 | 王芳 | A kind of medical system with monitoring infusion |
| CN105975740B (en) * | 2016-04-21 | 2018-09-11 | 寇玮蔚 | A kind of medical system with intelligent diagnostics |
| CN105975742A (en) * | 2016-04-21 | 2016-09-28 | 张明飞 | Comprehensive medical system with family care function |
| CN105975741A (en) * | 2016-04-21 | 2016-09-28 | 张明飞 | Medical system with patient state grading function |
| CN105975741B (en) * | 2016-04-21 | 2018-07-24 | 张明飞 | A kind of medical system with patient condition classification |
| CN105975742B (en) * | 2016-04-21 | 2018-07-24 | 张明飞 | A kind of general practice system with residential care |
| CN105930223A (en) * | 2016-04-24 | 2016-09-07 | 湖南大学 | Method for reducing size of check point file |
| CN106228185A (en) * | 2016-07-20 | 2016-12-14 | 武汉盈力科技有限公司 | A kind of general image classifying and identifying system based on neutral net and method |
| CN106228185B (en) * | 2016-07-20 | 2019-10-15 | 武汉盈力科技有限公司 | A kind of general image classifying and identifying system neural network based and method |
| CN106340024A (en) * | 2016-08-30 | 2017-01-18 | 厦门美图之家科技有限公司 | Image segmentation method and application and computing device |
| CN106340024B (en) * | 2016-08-30 | 2019-07-02 | 厦门美图之家科技有限公司 | A kind of image partition method, application and calculate equipment |
| CN106341664A (en) * | 2016-09-29 | 2017-01-18 | 浙江宇视科技有限公司 | Data processing method and apparatus |
| CN106341664B (en) * | 2016-09-29 | 2019-12-13 | 浙江宇视科技有限公司 | data processing method and device |
| CN108230219A (en) * | 2016-12-09 | 2018-06-29 | 中国科学院深圳先进技术研究院 | Enhance the method and system of encrypted domain jpeg image contrast |
| CN108230219B (en) * | 2016-12-09 | 2021-10-29 | 中国科学院深圳先进技术研究院 | Method and system for enhancing JPEG image contrast of encrypted domain |
| CN106713929A (en) * | 2017-02-16 | 2017-05-24 | 清华大学深圳研究生院 | Video interframe prediction enhancement method based on deep neural network |
| CN106713929B (en) * | 2017-02-16 | 2019-06-28 | 清华大学深圳研究生院 | A kind of video inter-prediction Enhancement Method based on deep neural network |
| CN110520909A (en) * | 2017-04-17 | 2019-11-29 | 微软技术许可有限责任公司 | The neural network processor of bandwidth of memory utilization rate is reduced using the compression and decompression of activation data |
| US11528033B2 (en) | 2017-04-17 | 2022-12-13 | Microsoft Technology Licensing, Llc | Neural network processor using compression and decompression of activation data to reduce memory bandwidth utilization |
| US11182667B2 (en) | 2017-04-17 | 2021-11-23 | Microsoft Technology Licensing, Llc | Minimizing memory reads and increasing performance by leveraging aligned blob data in a processing unit of a neural network environment |
| CN107610773A (en) * | 2017-09-12 | 2018-01-19 | 北京即刻叁维数据科技股份有限公司 | A kind of vascular dissection aided diagnosis method based on sustainer medical image |
| CN108464840B (en) * | 2017-12-26 | 2021-10-19 | 安徽科大讯飞医疗信息技术有限公司 | Automatic detection method and system for breast lumps |
| CN108464840A (en) * | 2017-12-26 | 2018-08-31 | 安徽科大讯飞医疗信息技术有限公司 | A kind of breast lump automatic testing method and system |
| CN110856035A (en) * | 2018-07-24 | 2020-02-28 | 顶级公司 | Processing image data to perform object detection |
| CN110856035B (en) * | 2018-07-24 | 2024-01-26 | Arm有限公司 | Processing image data to perform object detection |
| CN110766644B (en) * | 2018-07-27 | 2023-03-24 | 杭州海康威视数字技术股份有限公司 | Image down-sampling method and device |
| CN110766644A (en) * | 2018-07-27 | 2020-02-07 | 杭州海康威视数字技术股份有限公司 | Image down-sampling method and device |
| CN109284530A (en) * | 2018-08-02 | 2019-01-29 | 西北工业大学 | Space non-cooperative target appearance rail integration method for parameter estimation based on deep learning |
| CN109284530B (en) * | 2018-08-02 | 2021-05-18 | 西北工业大学 | Deep learning-based attitude and orbit integrated parameter estimation method for space non-cooperative target |
| CN110876062A (en) * | 2018-08-31 | 2020-03-10 | 三星电子株式会社 | Electronic device for high-speed compression processing of feature map and control method thereof |
| CN109509197B (en) * | 2018-09-26 | 2021-11-09 | 东软医疗系统股份有限公司 | Method, device and equipment for dividing region of interest and storage medium |
| CN109509197A (en) * | 2018-09-26 | 2019-03-22 | 沈阳东软医疗系统有限公司 | A kind of method, apparatus, equipment and storage medium for dividing area-of-interest |
| CN110473186B (en) * | 2019-02-14 | 2021-02-09 | 腾讯科技（深圳）有限公司 | Detection method based on medical image, model training method and device |
| CN110473186A (en) * | 2019-02-14 | 2019-11-19 | 腾讯科技（深圳）有限公司 | A kind of detection method based on medical image, the method and device of model training |
| CN110222772A (en) * | 2019-06-10 | 2019-09-10 | 浙江大学 | A kind of medical image mark recommended method based on block rank Active Learning |
| CN113538253B (en) * | 2020-04-21 | 2023-09-15 | 杭州普健医疗科技有限公司 | Medical image processing method, medium and CT device |
| CN113538253A (en) * | 2020-04-21 | 2021-10-22 | 杭州普健医疗科技有限公司 | Medical image processing method, medium and CT device |
| CN111583292B (en) * | 2020-05-11 | 2023-07-07 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111583292A (en) * | 2020-05-11 | 2020-08-25 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111841017A (en) * | 2020-05-29 | 2020-10-30 | 北京编程猫科技有限公司 | Game AI programming realization method and device |
| CN111654660A (en) * | 2020-06-08 | 2020-09-11 | 上海纽菲斯信息科技有限公司 | Video conference system coding transmission method based on image segmentation |
| CN111654660B (en) * | 2020-06-08 | 2021-03-23 | 上海纽菲斯信息科技有限公司 | Video conference system coding transmission method based on image segmentation |
| WO2022067656A1 (en) * | 2020-09-30 | 2022-04-07 | 华为技术有限公司 | Image processing method and apparatus |
| CN112401829A (en) * | 2020-11-26 | 2021-02-26 | 温州眼视光国际创新中心 | Remote intelligent control system of slit lamp |
| CN113038118A (en) * | 2020-12-08 | 2021-06-25 | 重庆邮电大学 | Video QoE prediction system and method based on big data environment |
| CN112616054A (en) * | 2020-12-11 | 2021-04-06 | 北京林业大学 | Self-adaptive compression transmission and recovery method and device for wild animal monitoring image |
| CN112616054B (en) * | 2020-12-11 | 2023-03-03 | 北京林业大学 | Self-adaptive compression transmission and recovery method and device for wild animal monitoring image |
| CN112616040A (en) * | 2020-12-11 | 2021-04-06 | 北京林业大学 | Wild animal image transmission method and system based on distributed architecture |
| WO2023279964A1 (en) * | 2021-07-08 | 2023-01-12 | 华为技术有限公司 | Data compression method and apparatus, and computing device and storage medium |
| CN113935913A (en) * | 2021-10-08 | 2022-01-14 | 北京计算机技术及应用研究所 | Black box image confrontation sample generation method with visual perception concealment |
| CN116436987A (en) * | 2023-06-12 | 2023-07-14 | 深圳舜昌自动化控制技术有限公司 | IO-Link master station data message transmission processing method and system |
| CN116436987B (en) * | 2023-06-12 | 2023-08-22 | 深圳舜昌自动化控制技术有限公司 | IO-Link master station data message transmission processing method and system |
| CN117115045A (en) * | 2023-10-24 | 2023-11-24 | 吉林大学 | Method for improving medical image data quality based on Internet generation type artificial intelligence |
| CN117115045B (en) * | 2023-10-24 | 2024-01-09 | 吉林大学 | Method for improving medical image data quality based on Internet generation type artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102332162A (en) | Method for automatic recognition and stage compression of medical image regions of interest based on artificial neural network | |
| Kaymak et al. | Breast cancer image classification using artificial neural networks | |
| CN109272048B (en) | Pattern recognition method based on deep convolutional neural network | |
| CN109844808A (en) | Multi-modality medical image processing | |
| WO2014036473A1 (en) | Supervised machine learning technique for reduction of radiation dose in computed tomography imaging | |
| CN111798440A (en) | Medical image artifact automatic identification method, system and storage medium | |
| CN111369574B (en) | Thoracic organ segmentation method and device | |
| Shen et al. | Lossless compression of curated erythrocyte images using deep autoencoders for malaria infection diagnosis | |
| WO2020066257A1 (en) | Classification device, classification method, program, and information recording medium | |
| US20240112329A1 (en) | Distinguishing a Disease State from a Non-Disease State in an Image | |
| CN114566282B (en) | Treatment decision system based on echocardiogram detection report | |
| CN108062779B (en) | Lossless compression method for breast tomography radiographic image | |
| Fergus et al. | Pressure ulcer categorisation using deep learning: A clinical trial to evaluate model performance | |
| CN115130543B (en) | Image recognition method and device, storage medium and electronic equipment | |
| CN114049315B (en) | Joint recognition method, electronic device, storage medium, and computer program product | |
| Rai et al. | An enhanced compression method for medical images using SPIHT encoder for fog computing | |
| Paul et al. | A health care image compression scheme using discrete wavelet transform and convolution neural network | |
| Ab Aziz et al. | The comprehensive review of neural network: an intelligent medical image compression for data sharing | |
| Amri et al. | Protocol for transmission and archiving of medical images (wavREPro. JPEG) | |
| CN108961216A (en) | A kind of Combined bone tumor micro-wound open biopsy device and control method | |
| Van der Sluijs et al. | Diagnostically Lossless Compression of Medical Images | |
| CN116389761B (en) | Clinical simulation teaching data management system of nursing | |
| CN113487587B (en) | CT image segmentation and bone density t value prediction method, system and computer equipment | |
| Yue et al. | Preoperative prediction of prosthetic size in total knee arthroplasty based on multimodal data and deep learning | |
| CN116957524B (en) | Talent information intelligent management method and system in technology transfer process |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120125 |