TWI399193B - Measuring total and local changes of knee cartilage volume - Google Patents
Measuring total and local changes of knee cartilage volume Download PDFInfo
- Publication number
- TWI399193B TWI399193B TW98143087A TW98143087A TWI399193B TW I399193 B TWI399193 B TW I399193B TW 98143087 A TW98143087 A TW 98143087A TW 98143087 A TW98143087 A TW 98143087A TW I399193 B TWI399193 B TW I399193B
- Authority
- TW
- Taiwan
- Prior art keywords
- trajectory
- treatment
- humeral surface
- gvf
- cartilage
- Prior art date
Links
Landscapes
- Magnetic Resonance Imaging Apparatus (AREA)
Description
本發明係關於一種膝蓋軟骨量變化之量測系統及方法,尤指一種利用核磁共振影像(MRI)分析膝蓋軟骨量變化的系統及方法。The present invention relates to a measurement system and method for the change of the amount of knee cartilage, and more particularly to a system and method for analyzing changes in the amount of knee cartilage using nuclear magnetic resonance imaging (MRI).
隨著醫學的進步,人類平均壽命的延長,因此原本被設計使用年限為40~60年的人類股關節及膝關節,到老都發生退化關節炎。尤其最近現代人常常忽略關節的保養且過度使用,使得退化性關節炎提早來臨,而其中很多膝蓋骨關節炎治療方法被發展出來,又一般咸認,退化性關節炎與關節軟骨的磨損退化有很大的關聯,因此利用軟骨量變化量測來驗證膝蓋骨關節炎療程的療效,就顯得特別重要。With the advancement of medicine, the average life expectancy of human beings has prolonged. Therefore, human joints and knee joints, which were originally designed to be used for 40 to 60 years, have degenerative arthritis. Especially in recent times, modern people often neglect the maintenance and overuse of joints, which makes degenerative arthritis come early, and many of the treatment methods for knee osteoarthritis have been developed, and it is generally recognized that degenerative arthritis and articular cartilage wear and tear are very degraded. Large associations, so it is particularly important to use the amount of cartilage measurement to verify the efficacy of knee osteoarthritis.
而量測膝蓋軟骨量的文獻大致分成兩大類:第一類是從MRI取像後的影像序列中圈選出膝蓋股骨軟骨,進而找出膝蓋股骨軟骨總量變化。The literature for measuring the amount of knee cartilage is roughly divided into two categories: the first is to select the knee femoral cartilage from the image sequence after MRI imaging, and then to find the total amount of cartilage of the knee femur.
第二類是經過脛骨定位後,只計算受損的局部膝蓋脛骨軟骨量變化(因為脛骨面地形起伏大,所以不採取骰骨軟骨來作量測)。The second type is that after the patella is positioned, only the damaged local knee tibia cartilage volume changes (because the humeral surface is undulating, so the iliac cartilage is not taken for measurement).
Habib et al.提出一個粗略的架構來量測受損的局部膝蓋軟骨量變化,其中第一組是治療前MRI影像序列作輸入和處理,而治療中每段療程的MRI影像序列資料也作輸入和處理,就能做定位處理,進而計算出治療中每段療程改 善程度。由前述的架構中可以看出:已知的軟骨選取是採用手動選取,定義脛骨面是用手動選取或Canny選取,並進行脛骨面三維匹配,而所述的軟骨選取、定義脛骨面及脛骨面三維匹配等三方面都有改善的空間:如既有的軟骨選取及定義脛骨面多採用人工選取,而人工選取對於操作者造成很大的負擔,且準確度相對較低;另一方面,在脛骨面三維匹配時對於每一張MRI影像都必須定義初始軌跡,因此相對加重操作負擔,並延長執行時間。Habib et al. proposed a rough framework to measure changes in damaged local knee cartilage volume. The first group was the pre-treatment MRI image sequence for input and processing, and the MRI image sequence data for each treatment session was also input. And processing, you can do the positioning process, and then calculate the treatment course Goodness. It can be seen from the foregoing structure that the known cartilage selection is manually selected, and the definition of the tibial surface is manually selected or selected by Canny, and the three-dimensional matching of the tibial surface is performed, and the cartilage is selected to define the tibial surface and the tibial surface. There are room for improvement in three aspects such as three-dimensional matching: if the existing cartilage is selected and the tibial surface is defined by manual selection, the manual selection imposes a great burden on the operator and the accuracy is relatively low; on the other hand, When the humeral surface is three-dimensionally matched, the initial trajectory must be defined for each MRI image, thus increasing the operational burden and extending the execution time.
因此,本發明主要目的在提供一種膝蓋軟骨量變化之量測方法,其可針對已知流程進行改善,以提高其準確度並縮短量測執行時間。Accordingly, it is a primary object of the present invention to provide a method of measuring changes in the amount of knee cartilage that can be improved for known procedures to improve its accuracy and to reduce measurement execution time.
為達成前述目的採取的主要技術手段係針對治療前後的MRI影像序列執行以下工作流程,包括:軟骨選取程序,係在載入MRI影像後,先大略框出軟骨所在,並在框內進行邊緣偵測,利用偵測出的邊緣作為選取軟骨的參考;脛骨面選取程序,在載入數張MRI影像後,先定義出一初始軌跡,並修正該初始軌跡以去除雜訊及平滑化,再利用脛骨面軌跡快速GVF擴展演算法找出脛骨面結果軌跡,並作為下一張影像的軌跡;脛骨面3D疊合程序,利用找到的脛骨面結果軌跡對軟骨進行定位,其包括整數座標內插、分別定義治療前後的最高特徵點、範圍擴展最佳對應點、快速3D縮小; 由於驗證膝蓋骨關節炎的療效,必須量測局部膝蓋軟骨量的變化,利用前述方法可以在治療前及治療後的MRI影像序列中對軟骨模型作定位,並且找出其中局部變化量;且利用前述方法可將傳統量測方法中大部分採取人工選取的部分進行自動化處理,除可減少操作者的負擔、提高準確度外,並可縮短量測執行時間。The main technical means for achieving the above objectives is to perform the following workflow for the MRI image sequence before and after treatment, including: cartilage selection procedure, after loading the MRI image, the cartilage is roughly framed, and the edge detection is performed in the frame. Measure, use the detected edge as a reference for selecting cartilage; the humeral surface selection program, after loading several MRI images, first define an initial trajectory, and correct the initial trajectory to remove noise and smoothing, and then reuse The patella trajectory rapid GVF expansion algorithm finds the patella surface trajectory and serves as the trajectory of the next image; the humeral surface 3D superimposition procedure uses the found humeral surface trajectory to locate the cartilage, including integer coordinate interpolation, Define the highest feature points before and after treatment, the best corresponding points for range expansion, and fast 3D reduction; In order to verify the curative effect of knee osteoarthritis, it is necessary to measure the change of local knee cartilage volume. The above method can be used to locate the cartilage model in the MRI image sequence before and after treatment, and find out the local variation; The method can automatically process most of the traditional measurement methods in the traditional measurement method, in addition to reducing the burden on the operator, improving the accuracy, and shortening the measurement execution time.
關於本發明之一較佳實施例,首先請參閱第一圖所示,其揭示本發明之整體架構,包括一輸入部分(10)、一運算部分(20)及一輸出部分(30);其中:該輸入部分(10)係載入治療前及治療後的核磁共振(MRI,以下簡稱為MRI)影像序列,並針對該MRI影像序列進行座標定義、體素(Voxel)定義、整體與局部膝蓋MRI影像序列定義等;運算部分(20)則包括一軟骨選取程序(21)、一脛骨面選取程序(22)及一3D疊合程序(23);以下即分別針對前述輸入部分(10)、運算部分(20)及輸出部分(30)的內容及細節進一步說明后:輸入部分(10)執行內容包括:MRI影像序列座標定義、體素(Voxel)定義、整體與局部膝蓋MRI影像序列定義;其中:MRI影像序列座標定義With regard to a preferred embodiment of the present invention, first referring to the first figure, the overall architecture of the present invention is disclosed, including an input portion (10), an arithmetic portion (20), and an output portion (30); The input portion (10) is a pre- and post-treatment nuclear magnetic resonance (MRI, hereinafter referred to as MRI) image sequence, and coordinates definition, voxel definition, overall and local knees for the MRI image sequence. The MRI image sequence definition and the like; the operation part (20) includes a cartilage selection program (21), a humeral surface selection program (22), and a 3D superimposition program (23); the following is respectively for the input portion (10), The contents and details of the operation part (20) and the output part (30) are further described: the input part (10) execution content includes: MRI image sequence coordinate definition, Voxel definition, and overall knee local MRI image sequence definition; Where: MRI image sequence coordinates definition
如前揭所述,本發明主要係在治療前及治療後的MRI影像序列中對軟骨模型作定位,並且找出其中局部變化量,因此必須分別提供並輸入治療前及治療後的MRI影像序 列,又因快速3D疊合之局部膝蓋軟骨量變化量測必須針對三維座標資料做處理,故輸入部分(10)將在MRI影像序列中作座標定義,其方法請參閱第二圖所示,主要係先定義出三維原點,接著定義出MRI影像的長高,再把編號0到N-1的各張MRI影像進行堆疊,藉此即可定義出三維座標軸與三維座標。As described above, the present invention mainly locates the cartilage model in the MRI image sequence before and after treatment, and finds the local variation, so it is necessary to separately provide and input the MRI image sequence before and after treatment. Columns, because of the rapid 3D overlap of the local knee cartilage volume measurement must be processed for the three-dimensional coordinate data, so the input part (10) will be defined in the MRI image sequence, the method shown in the second figure, The main system first defines the three-dimensional origin, then defines the length of the MRI image, and then stacks the MRI images numbered 0 to N-1, thereby defining the three-dimensional coordinate axis and the three-dimensional coordinates.
體素(Voxel)定義:在以下實施例中,係以體素作為運算單位,若想轉換成實際體積,則公式如下
其中、和分別為MRI實際影像寬(mm)、高(mm)和總厚度(mm),X、Y和Z分別是虛擬MRI影像寬(pixel)、高(pixel)和總張數(pixel),a則是轉換係數。among them , with The MRI actual image width (mm), height (mm) and total thickness (mm), respectively, X, Y and Z are the virtual MRI image width (pixel), height (pixel) and total number of pixels (pixel), a Is the conversion factor.
整體與局部膝蓋MRI影像序列定義:Whole and partial knee MRI image sequence definitions:
(一)整體膝蓋MRI影像序列定義(MRI_1和MRI_2)(1) Definition of overall knee MRI image sequence (MRI_1 and MRI_2)
整體膝蓋MRI影像資料係分別輸入運算部分(20)的軟骨選取程序(21)及脛骨面選取程序(22);其中:The overall knee MRI image data is input into the cartilage selection program (21) of the operation part (20) and the tibial surface selection program (22);
(1)輸入到軟骨選取程序(21)的整體膝蓋MRI影像序列定義:只要含有軟骨的MRI影像就做選取,因此必須定義出MRI_1(治療前)和MRI_2(治療後)是含有軟骨成分膝蓋MRI影像的集合。(1) Whole knee MRI image sequence definition input to the cartilage selection program (21): As long as the MRI image containing cartilage is selected, it is necessary to define MRI_1 (before treatment) and MRI_2 (after treatment) to contain cartilage component knee MRI A collection of images.
(2)輸入到脛骨面選取程序(22)的整體膝蓋MRI影像序列定義:在含有腓骨特徵的MRI影像中作排除後才輸入。(2) The definition of the overall knee MRI image sequence input to the tibial surface selection program (22): input after exclusion in the MRI image containing the tibial feature.
(二)局部膝蓋MRI影像序列定義(MRI_3和MRI_4)(2) Local knee MRI image sequence definition (MRI_3 and MRI_4)
(1)輸入到軟骨選取程序(21)的局部膝蓋MRI影像序列 :由於受損軟骨主要集中在內側軟骨,所以在輸入MRI影像序列中,只要選取內側軟骨,其他部分則反之,進而減少軟骨選取程序(21)的張數。(1) Local knee MRI image sequence input to the cartilage selection program (21) Since the damaged cartilage is mainly concentrated in the medial cartilage, in the input MRI image sequence, as long as the medial cartilage is selected, the other parts are reversed, thereby reducing the number of sheets of the cartilage selection program (21).
(2)輸入到脛骨面選取程序(22)的局部膝蓋MRI影像序列:與輸入的整體膝蓋MRI影像序列一樣。(2) Local knee MRI image sequence input to the tibial surface selection program (22): same as the input overall knee MRI image sequence.
仍請參閱第一圖所示,本發明的運算部分(20)包括軟骨選取程序(21)、脛骨面選取程序(22)及3D疊合程序(23);其中:該軟骨選取程序(21)係先大略框出軟骨所在(如附件一所示),並框內做canny邊緣偵測,利用偵測出的邊緣做為選取軟骨時的參考。Still referring to the first figure, the operation part (20) of the present invention includes a cartilage selection program (21), a tibial surface selection program (22), and a 3D superimposition program (23); wherein: the cartilage selection program (21) The carcass is roughly framed (as shown in Appendix 1), and canny edge detection is performed in the frame, and the detected edge is used as a reference for selecting cartilage.
請參閱第三圖所示,該脛骨面選取程序(22)包括:一定義GVF初始圓形軌跡的步驟(401):先在MRI影像序列的第一張MRI影像上點選兩點,第一點為圓心,第二點決定半徑;一修正前述GVF初始軌跡之步驟(402),包括去雜訊及平滑化;一對已經定義過初始圓形軌跡的MRI影像序列進行快速GVF擴展演算法之步驟(403)。Referring to the third figure, the humeral surface selection program (22) includes: a step (401) of defining an initial circular trajectory of the GVF: first selecting two points on the first MRI image of the MRI image sequence, first The point is the center of the circle, the second point determines the radius; a step (402) of correcting the initial GVF trajectory, including denoising and smoothing; and a pair of MRI image sequences that have defined the initial circular trajectory for fast GVF expansion algorithm Step (403).
前述步驟(402)進一步的技術內容係如第四圖所示之流程,其包括:令軌跡a範圍內pixel平均值趨近於微小值α,以減少脛骨面的雜訊(501):係先利用otsu’s thresholding找出自動闕值T,之後利用自動闕值T做二值化在跟原圖做相乘並計算GVF初始軌跡內平均pixel值,找出其pixel值 後判斷是否小於微小值α,若不是的話,再做一次,最後會得到脛骨面內雜訊幾乎消除的MRI影像結果;接著把T值以1為單位微調,在不大幅影響脛骨面真實度下讓雜訊盡量消失;脛骨面影像之平滑化處理(502):目的是透過平滑化處理以消除影像中脛骨面往內凸的特徵,該平滑化處理之方法如下:(A)以2:1之比例進行次取樣:藉此使影像的pixel值變小,進而加快平滑脛面影像處理技術執行速度(因降低取樣點),如附件二A、B所示,係揭示次取樣前後脛骨面內pixel值的差異比較。因而出現的脛骨面GVF結果軌跡誤差則在MRI影像序列之脛骨面軌跡快速GVF擴展演算法做解決;(B)中值濾波:用以減少類似胡椒鹽雜訊的出現,並可將雜訊模糊;(C)形態學:用以消除影像中脛骨面上所出現往內凸的特徵,而將其平滑化,且使雜訊消失;(D)增強脛骨面:將影像中非零的pixel值提高(例如設為255),其中一種方式如下:I”(x,y )=I’(x,y) * I(x,y) ÷255The further technical content of the foregoing step (402) is a flow as shown in the fourth figure, which comprises: making the average value of the pixel in the range of the trajectory a close to a small value α to reduce the noise of the humeral surface (501): Use the otsu's thresholding to find the automatic 阙 value T, then use the automatic 阙 value T to do the binarization and multiply with the original image and calculate the average pixel value in the initial trajectory of the GVF, find out the pixel value and judge whether it is less than the small value α, If not, do it again, and finally get the MRI image results that almost eliminate the intra-surface noise of the humerus; then fine-tune the T value by 1 to make the noise disappear as much as possible without significantly affecting the trueness of the humeral surface; Image smoothing (502): The purpose is to eliminate the inward convex feature of the humerus surface in the image by smoothing. The smoothing process is as follows: (A) subsampling at a ratio of 2:1: The pixel value of the image is made smaller, thereby speeding up the smoothing of the image processing technique (due to lowering the sampling point), as shown in Annexes A and B, revealing the difference in pixel values of the humerus surface before and after the sub-sampling. Therefore, the trajectory error of the humeral surface GVF results is solved in the fast GVF expansion algorithm of the humeral surface trajectory of the MRI image sequence; (B) median filtering: to reduce the appearance of pepper-like noise and blur the noise (C) Morphology: used to eliminate the inward convex features appearing on the humeral surface of the image, smoothing it and making the noise disappear; (D) enhancing the humeral surface: the non-zero pixel value in the image Raise (for example, set to 255), one of which is as follows: I" (x, y ) = I' (x, y) * I (x, y) ÷ 255
前述I代表一張MRI的圖像,I(x,y) 代表經過中值濾波處理過MRI圖像內座標(x,y)的pixel值,I’(x,y) 代表一張做過形態學MRI圖像內座標(x,y)的pixel值,I”(x,y) 仍代表一張MRI圖像內座標(x,y)的pixel值,且該張MRI圖像係做過形態學並經過增強脛骨面與降低雜訊處理的。The above I represents an MRI image, I (x, y) represents the pixel value of the inner coordinate (x, y) of the MRI image after median filtering, and I' (x, y) represents a form. Learn the pixel value of the inner coordinate (x, y) of the MRI image, I" (x, y) still represents the pixel value of the coordinates (x, y) of an MRI image, and the MRI image has been shaped Learn and enhance the humeral surface and reduce noise processing.
(E)利用canny偵測,找出邊緣。(E) Use canny detection to find the edge.
在完成前述步驟(A)~(E)後,接著執行GVF演算法(503),並將GVF結果軌跡還原為原來大小;如附件三A、B分別揭示執行GVF演算法前後的脛骨面軌跡之影像圖。After completing the foregoing steps (A) to (E), the GVF algorithm (503) is executed, and the GVF result track is restored to the original size; as shown in Annexes A and B, respectively, the humeral surface trajectory before and after the execution of the GVF algorithm is revealed. Image map.
接著將修改過脛骨面GVF初始軌跡以定義初始軌跡的圓心為基準作微幅縮小(如0.95倍)(504);執行此一步驟具有兩個作用與優點:Then, the initial trajectory of the humeral surface GVF is modified to define a micro-scale reduction (eg, 0.95 times) based on the center of the initial trajectory (504); performing this step has two functions and advantages:
1.不用針對每一張MRI影像去定義GVF初始圓形軌跡,因為可以參考前一張MRI影像的GVF結果軌跡。1. It is not necessary to define the initial circular path of GVF for each MRI image, because it can refer to the GVF result track of the previous MRI image.
2.因為下一張MRI影像會參考前一張MRI影像的結果軌跡,且GVF初始軌跡很靠近脛骨面,可大幅縮短下一張MRI影像作GVF擴展的時間。2. Because the next MRI image will refer to the result track of the previous MRI image, and the initial GVF trajectory is very close to the humeral surface, the time of the next MRI image for GVF expansion can be greatly shortened.
再請參閱第三圖所示,在完成初始軌跡的修正後,即進一步對前述MRI影像序列之脛骨面軌跡進行快速GVF擴展演算法(403),如前揭所述,其主要是針對已經定義過初始圓形軌跡的MRI影像序列實施,供每張MRI影像自動搜尋脛骨面軌跡,其實施內容部分與前述脛骨面選取程序相同,如第五圖所示,其包括:以2:1之比例進行次取樣(801);中值濾波:用以減少類似胡椒鹽雜訊的出現,並可將雜訊模糊(802);形態學(803):用以消除影像中脛骨面上所出現往內凸的特徵,而將其平滑化,且令脛骨面內的雜訊些微模糊化; 增強脛骨面及進一步降低雜訊(804);此一步驟在增強脛骨面上可能有所不足,則有賴下一步驟(805)來解決;應用在縮小版影像處理與GVF技術之增強脛骨面(805):主要係利用找出自動二值化中的闕值(作用係將骨頭成分和非骨頭成分予以分開),之後利用對比度擴展以增強脛骨面;而此舉有可能出現雜訊也跟著增強的疑慮,如第六圖所示,但由於之前的步驟已經把含有脛骨面內雜訊成分(A)都清除完畢,所以再利用對比度擴展增強脛骨面,並不會使雜訊增強。Referring to the third figure, after the initial trajectory correction is completed, a fast GVF expansion algorithm (403) is further performed on the humeral surface trajectory of the aforementioned MRI image sequence. As described above, it is mainly for the defined The MRI image sequence of the initial circular trajectory is implemented for each MRI image to automatically search for the humeral surface trajectory, and the implementation content is the same as the aforementioned humeral surface selection procedure, as shown in the fifth figure, which includes: a ratio of 2:1 Sub-sampling (801); median filtering: to reduce the appearance of pepper-like noise, and to blur the noise (802); morphology (803): to eliminate the appearance of the humeral surface in the image Convex features, smoothing them, and slightly blurring the noise in the cheekbones; Enhance the humeral surface and further reduce the noise (804); this step may be insufficient in the enhanced humeral surface, which is solved by the next step (805); applied to the enhanced humeral surface of the reduced image processing and GVF technology ( 805): Mainly to find out the enthalpy in the automatic binarization (the function is to separate the bone component and the non-bone component), and then use the contrast extension to enhance the humeral surface; and this may cause noise to be enhanced. The doubts, as shown in the sixth figure, but because the previous steps have been cleared of the intra-aortic noise component (A), the use of contrast expansion to enhance the humeral surface does not enhance the noise.
判斷MRI影像脛骨面軌跡內的平均Pixel值是否大於某值(如1.5)(806),若未大於該值,即進行canny邊緣偵測(808);若大於該值,則執行下一步驟(807);在MRI影像脛骨面軌跡內進行roifill處理(807);前述步驟(806)(807)主要是在MRI影像脛骨面內雜訊過重,且步驟(802)(803)(804)仍不能完全清除時所採用的步驟;利用canny進行邊緣偵測(808);進行GVF擴展演算法(809),用以找尋前述步驟(807)處理結果中的脛骨面軌跡;將前述步驟(809)找到的脛骨面軌跡放大數倍(如4倍)(810)。Determine whether the average Pixel value in the patella track of the MRI image is greater than a certain value (such as 1.5) (806). If it is not greater than the value, perform canny edge detection (808); if it is greater than the value, perform the next step ( 807); performing a roifill treatment (807) in the patella trajectory of the MRI image; the aforementioned step (806) (807) is mainly that the intramural noise of the tibia in the MRI image is too heavy, and the steps (802) (803) (804) still cannot The steps used to completely clear; use canny for edge detection (808); perform GVF expansion algorithm (809) to find the humeral surface trajectory in the processing result of the foregoing step (807); find the aforementioned step (809) The humeral surface track is magnified several times (eg, 4 times) (810).
前述步驟(801)~(810)係進行一縮小版影像處理與GVF演算,主要係為找出一粗略的脛骨面軌跡,接著將進行一還原影像處理與GVF技術,以便對前述脛骨面軌跡進行細部微調,該還原之影像處理與GVF技術請參閱第七圖所示,包括有以下步驟: 將脛骨面軌跡還原(放大4倍)(101);中值濾波(102);canny邊緣偵測(103);進行GVF擴展演算法(104),以找出經過微調處理的脛骨面軌跡;將前述步驟(104)找出的脛骨面軌跡略微往內縮小(如0.95倍),供作為下一張MRI影像的軌跡(105)。The foregoing steps (801)-(810) perform a reduced version of image processing and GVF calculation, mainly to find a rough humeral surface trajectory, and then perform a reduced image processing and GVF technique to perform the aforementioned humeral surface trajectory. Detailed fine-tuning, the reduced image processing and GVF technology, please refer to the seventh figure, including the following steps: Restore the humeral surface track (magnification 4 times) (101); median filter (102); canny edge detection (103); perform GVF expansion algorithm (104) to find the tibial surface track after fine tuning; The trajectory of the humerus surface found in the foregoing step (104) is slightly reduced inward (e.g., 0.95 times) for the trajectory (105) of the next MRI image.
而MRI影像序列經過前列步驟一一完成快速GVF擴展演算法之後可找出一脛骨面軌跡,且每一張MRI影像可作為下一張MRI影像的軌跡,以方便自動找尋脛骨面軌跡。儘管前述步驟會自動找出脛骨面軌跡,但這些軌跡並不一定絕對正確,故可適度加入手動修正,使其更具準確性。The MRI image sequence can find a humeral surface track after completing the fast GVF expansion algorithm, and each MRI image can be used as the trajectory of the next MRI image to facilitate automatic search for the humeral surface track. Although the aforementioned steps will automatically find the humeral surface trajectory, these trajectories are not necessarily correct, so manual correction can be added to make it more accurate.
仍請參閱第一圖所示,在完成脛骨面選取程序(22)之後,隨即進入3D疊合程序(23),以透過脛骨面3D疊合技術對軟骨進行定位,其流程請參閱第八圖所示,包括有:一整數座標內插步驟(1101),由於輸入的MRI影像序列其軌跡取樣點具有稀疏性,故執行此一步驟使其具有一定的密集性(座標以1為單位),主要係利用在MRI影像序列中每張MRI影像的脛骨面軌跡,構成一三維地形圖(如附件四A所示),又將該三維地形圖轉換成影像格式儲存(如附件四B所示),其中三維地形圖與影像檔是採用相同的長寬,又三維地形圖的高度為影像檔的pixel值,如此即可使用影像處理概念進行後續的處理;一治療前後個別最高特徵點搜尋步驟(1102),由於前 一步驟已經將脛骨面轉換成影像格式,故在此步驟中採用一個mask以判斷特徵點,再利用高度找出其最高特徵點,其搜尋方式係如第九圖所示,在分別載入治療前與治療後標記的脛骨面數據後,即利用一3*3的區塊(block)來判斷特徵點(1102A),該block即為前述的mask;若3*3區塊中心的pixel值為T,其周圍的pixel值都比T值小,該點就判斷為特徵點;接著利用高度找出最高特徵點(1102B),由於利用高度找出特徵點,但最高點並不一定只有一點,例如同時找到高度相同的3點,則從三點取較佳的一點,令其成為一較佳的最高特徵點(1102C)。Still referring to the first figure, after completing the humeral surface selection procedure (22), the 3D superimposition procedure (23) is followed to locate the cartilage through the humeral surface 3D superposition technique. The method includes: an integer coordinate interpolation step (1101). Since the input MRI image sequence has a smeared sampling point having sparsity, performing this step has a certain density (coordinates are in units of 1). The main purpose is to use a tibial surface trajectory of each MRI image in the MRI image sequence to form a three-dimensional topographic map (as shown in Annex IV A), and convert the three-dimensional topographic map into image format storage (as shown in Annex IV B). The three-dimensional topographic map and the image file adopt the same length and width, and the height of the three-dimensional topographic map is the pixel value of the image file, so that the image processing concept can be used for subsequent processing; and the highest highest feature point searching step before and after treatment ( 1102), due to the former In one step, the humeral surface has been converted into an image format, so a mask is used in this step to judge the feature points, and then the height is used to find the highest feature point. The search method is as shown in the ninth figure, and the treatment is separately loaded. After the pre- and post-treatment marked humeral surface data, a 3*3 block is used to determine the feature point (1102A), which is the aforementioned mask; if the pixel value of the center of the 3*3 block is T, the pixel value around it is smaller than the T value, the point is judged as the feature point; then the height is used to find the highest feature point (1102B), because the feature point is found by using the height, but the highest point is not necessarily only a little, For example, if three points with the same height are found at the same time, a better point is taken from the three points to make it a better highest feature point (1102C).
再請參閱第八圖所示,利用脛骨面3D疊合技術對軟骨進行定位,進一步包括:一範圍擴展最佳應點之步驟(1103),由於治療前後找出的較佳最高特徵點並不一定相對應,若強制讓治療前後所找的點相位移,將造成軟骨定位的準確率下降,故執行範圍擴展最佳對應點可提高軟骨定位的準確性,其方法請配合參閱第十圖,假設治療前的脛骨面軌跡上具有最高特徵點A、B和C,治療後的脛骨面軌跡上具有最高特徵點D,在治療前的脛骨面軌跡上要找到對應D的特徵點D’,因此將利用特徵點A、B和C中的其中一點作為中心,去找出D’來,如第十一圖所示,若取出一點較佳最高特徵點A,則採用P*Q為範圍去搜尋D’點,找到之後,D(治療後)和D’(治療前)分別都是三維位移旋轉中心;一快速3D縮小步驟(1104),由於前一步驟的範圍擴 展最佳對應點技術可能大幅增加執行時間,因此利用此步驟在不降低軟骨定位準確性情況下,縮短演算法執行時間,主要係將已知三維脛骨面座標轉換成影像(圖片)格式,再利用x倍的bilinear縮小,以減少三維脛骨面取樣點;一3D旋轉疊合步驟(1105),用以定位脛骨面及軟骨,其流程係如下列:載入先前步驟中所找出的特徵點D與D’;以特徵點D和D’之間作為治療前後的位移依據,之後分別做治療前後三維脛骨面座標相位移;進行三維旋轉,從每個角度和向量找出一脛骨面相似度平均差(Erroravg )的集合,並從中找出最小值,該值即為治療前後脛骨面最相似時,從而可得到治療前後脛骨面最相似時的角度、向量...等參數。Referring to Figure 8, the positioning of the cartilage using the 3D superimposition technique of the humerus surface further includes: a step of expanding the optimal point (1103), because the best feature points found before and after treatment are not Correspondingly, if the position of the point found before and after treatment is forced to be displaced, the accuracy of cartilage positioning will be reduced. Therefore, the best corresponding point of the expansion range can improve the accuracy of cartilage positioning. For the method, please refer to the tenth figure. It is assumed that the patella trajectory before treatment has the highest characteristic points A, B and C, and the patella trajectory after treatment has the highest characteristic point D, and the feature point D' corresponding to D is found on the patella surface trajectory before treatment. Using one of the feature points A, B, and C as the center, to find D', as shown in the eleventh figure, if a better highest feature point A is taken, P*Q is used as the range to search. D' point, after finding, D (after treatment) and D' (before treatment) are respectively three-dimensional displacement rotation centers; a fast 3D reduction step (1104), due to the range expansion of the previous step, the best corresponding point technology may be significantly Increase execution time, so Use this step to shorten the execution time of the algorithm without reducing the accuracy of cartilage positioning. The main method is to convert the known three-dimensional humeral surface coordinates into an image (picture) format, and then use x-fold bilinear reduction to reduce the three-dimensional tibial surface sampling. a 3D rotation lamination step (1105) for locating the humeral surface and cartilage, the flow of which is as follows: loading the feature points D and D' found in the previous step; with feature points D and D' As the basis of displacement before and after treatment, the three-dimensional humeral surface coordinate displacement before and after treatment were performed separately; three-dimensional rotation was performed, and a set of the average abnormity difference (Error avg ) was found from each angle and vector, and the set was found. The minimum value, which is the most similar when the humeral surface before and after treatment, so that the angle, vector, etc. of the humeral surface before and after treatment can be obtained.
仍請參閱第一圖所示,在運算部分(20)對MRI影像序列進行軟骨選取程序(21)、脛骨面選取程序(22)及3D疊合程序(23)後,即由輸出部分(30)分別顯示整體膝蓋軟骨量的變化及局部膝蓋軟骨量的變化,以供判斷確認關節炎進行治療後的療效。Still referring to the first figure, after the operation part (20) performs the cartilage selection process (21), the tibial surface selection program (22), and the 3D superimposition program (23) on the MRI image sequence, the output portion (30) ) The change in the amount of total knee cartilage and the change in the amount of local knee cartilage are separately shown for the purpose of judging the efficacy of arthritis after treatment.
由上述說明可瞭解本發明量測軟骨變化量的具體技術內容,以下進一步針對實驗數據進行分析:From the above description, the specific technical content of measuring the amount of cartilage change of the present invention can be understood, and the following is further analyzed for experimental data:
(一)在脛骨面選取方面(1) in the selection of the humeral surface
請參閱第第十二、十三圖所示,分別為治療前及治療後採取人工脛骨面修正及利用本發明進行自動脛骨面分析之次取樣倍數百分比曲線圖,由圖中可以明顯看出,以2倍進行次取樣,所得數據比其他倍數理想,而治療前後使 用次取樣2倍會比全手動降低操作者負擔降低約94%(4.6%是平均人工脛骨面修改與自動脛骨面所占比例)。Please refer to the twelfth and thirteenth figures, respectively, for the artificial humeral surface correction before and after treatment and the secondary sampling percentage curve of the automatic humeral surface analysis using the present invention, as is apparent from the figure, Sub-sampling at 2 times, the data obtained is better than other multiples, and before and after treatment A two-times sampling will reduce the operator's burden by approximately 94% (4.6% is the ratio of the average artificial humeral surface modification to the automatic humeral surface).
(二)範圍擴展最佳對應點技術方面(2) Scope expansion best corresponding point technical aspects
請參閱第十四圖所示,為是否進行範圍擴展最佳對應點所得總體軟骨疊合率之曲線比較圖,由圖中可以明顯看出,採用範圍擴展最佳對應點技術的數據較不使用該技術的數據好(約15%到30%的總體軟骨疊合率)。Please refer to the fourteenth figure, which is a comparison chart of the total cartilage overlap ratio obtained by the best corresponding point of the range expansion. It can be clearly seen from the figure that the data using the range-optimized corresponding point technique is less used. The data for this technique is good (about 15% to 30% of the overall cartilage overlap rate).
(三)快速3D縮小技術方面(3) Rapid 3D reduction technology
由於使用範圍擴展最佳對應點技術的數據會延長演算時間,故使用快速3D縮小技術來縮短時間,而執行3D縮小技術的倍數與執行時間的相對比例係如第十五圖所示(快速3D縮小倍率6倍,降低約31倍的時間)。Since the data of the extended corresponding point technology is used to extend the calculation time, the fast 3D reduction technique is used to shorten the time, and the relative ratio of the multiple of the 3D reduction technique to the execution time is as shown in the fifteenth figure (fast 3D) Reduce the magnification by 6 times and reduce the time by about 31 times).
(10)‧‧‧輸入部分(10)‧‧‧ Input section
(20)‧‧‧運算部分(20) ‧ ‧ arithmetic part
(21)‧‧‧軟骨選取程序(21)‧‧‧Cartilage selection procedure
(22)‧‧‧脛骨面選取程序(22) ‧‧‧Sacral surface selection procedure
(23)‧‧‧3D疊合程序(23) ‧‧3D overlapping procedure
(30)‧‧‧輸出部分(30)‧‧‧ Output section
第一圖:係本發明之系統架構示意圖。The first figure is a schematic diagram of the system architecture of the present invention.
第二圖:係本發明輸入部分針對MRI影像序列進行座標定義之示意圖。The second figure is a schematic diagram of the coordinate definition of the MRI image sequence in the input part of the present invention.
第三圖:係本發明運算部分中軟骨選取程序之流程圖。Fig. 3 is a flow chart showing the cartilage selection procedure in the arithmetic part of the present invention.
第四圖:係本發明修改GVF初始軌跡之流程圖。The fourth figure is a flow chart of modifying the initial trajectory of the GVF by the present invention.
第五圖:係本發明縮小版影像處理與GVF技術的流程圖。Fig. 5 is a flow chart of the reduced image processing and GVF technology of the present invention.
第六圖:係本發明利用對比度增強脛骨面後的訊號成分比示意圖。Fig. 6 is a schematic diagram showing the ratio of signal components after the patella surface is enhanced by contrast in the present invention.
第七圖:係本發明還原影像處理與GVF技術之流程圖 。Figure 7 is a flow chart of the reduced image processing and GVF technology of the present invention. .
第八圖:係本發明之脛骨面3D疊合程序流程圖。Figure 8 is a flow chart of the 3D superimposition procedure of the tibial surface of the present invention.
第九圖:係本發明3D疊合程序中搜尋最高特徵點之流程圖。Fig. 9 is a flow chart for searching for the highest feature point in the 3D superimposing program of the present invention.
第十圖:係本發明3D疊合程序中進行範圍擴展最佳對應點技術之一示意圖。Fig. 10 is a schematic diagram showing one of the techniques for performing range expansion optimal correspondence in the 3D superimposing program of the present invention.
第十一圖:係本發明3D疊合程序中進行範圍擴展最佳對應點技術又一示意圖。Eleventh figure: another schematic diagram of the range-optimized corresponding point technique in the 3D superimposing program of the present invention.
第十二圖:係本發明與人工脛骨面修正之治療前脛骨面所佔比例之曲線比較圖。Twelfth Figure: A comparison of the curves of the proportion of the humeral surface before treatment of the present invention and the artificial humeral surface correction.
第十三圖:係本發明與人工脛骨面修正之治療後脛骨面所佔比例之曲線比較圖。Thirteenth Graph: A comparison chart of the ratio of the proportion of the humerus surface after treatment of the present invention and the artificial humeral surface correction.
第十四圖:係本發明3D疊合程序中採用範圍擴展最佳對應點技術與未使用該技術之總體軟骨疊合率比較圖。Fig. 14 is a comparison diagram of the technique of using the range-extended optimal corresponding point in the 3D superimposing procedure of the present invention and the overall cartilage superposition ratio without using the technique.
第十五圖:係本發明3D疊合程序中實施範圍擴展最佳對應點技術後進一步執行3D縮小技術之執行時間與3D縮小倍率的關係曲線圖。Fifteenth Graph: A graph showing the relationship between the execution time of the 3D reduction technique and the 3D reduction ratio after performing the range expansion optimal corresponding point technique in the 3D superimposing program of the present invention.
附件一:係本發明在排除腓骨特徵後的MRI影像示意圖。Annex I: Schematic diagram of the MRI image of the present invention after excluding the tibial features.
附件二A、B:係本發明進行次取樣前後之脛骨面內pixel值的示意圖。Annex II A, B: is a schematic diagram of the in-plane pixel value of the tibia before and after the sub-sampling of the present invention.
附件三A、B:係本發明進行GVF演算法前後之脛骨面示意圖。Annexes A and B are schematic diagrams of the humeral surface before and after the GVF algorithm of the present invention.
附件四A、B:係本發明3D疊合程序中整數內插技術 之流程示意圖。Annex IV A, B: Integer interpolation technique in the 3D superimposition program of the present invention Schematic diagram of the process.
(10)‧‧‧輸入部分(10)‧‧‧ Input section
(20)‧‧‧運算部分(20) ‧ ‧ arithmetic part
(21)‧‧‧軟骨選取程序(21)‧‧‧Cartilage selection procedure
(22)‧‧‧脛骨面選取程序(22) ‧‧‧Sacral surface selection procedure
(23)‧‧‧3D疊合程序(23) ‧‧3D overlapping procedure
(30)‧‧‧輸出部分(30)‧‧‧ Output section
Claims (15)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW98143087A TWI399193B (en) | 2009-12-16 | 2009-12-16 | Measuring total and local changes of knee cartilage volume |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW98143087A TWI399193B (en) | 2009-12-16 | 2009-12-16 | Measuring total and local changes of knee cartilage volume |
Publications (2)
Publication Number | Publication Date |
---|---|
TW201121505A TW201121505A (en) | 2011-07-01 |
TWI399193B true TWI399193B (en) | 2013-06-21 |
Family
ID=45045630
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW98143087A TWI399193B (en) | 2009-12-16 | 2009-12-16 | Measuring total and local changes of knee cartilage volume |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI399193B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI399194B (en) * | 2011-01-25 | 2013-06-21 | Univ Nat Yunlin Sci & Tech | Semi-automatic knee cartilage mri image segmentation based on cellular automata |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050084145A1 (en) * | 1999-11-01 | 2005-04-21 | Pelletier Jean P. | Evaluating disease progression using magnetic resonance imaging |
US20080292169A1 (en) * | 2007-05-21 | 2008-11-27 | Cornell University | Method for segmenting objects in images |
US20090285466A1 (en) * | 2001-11-07 | 2009-11-19 | Medical Metrics, Inc. | Method, Computer Software, And System For Tracking, Stabilizing, And Reporting Motion Between |
-
2009
- 2009-12-16 TW TW98143087A patent/TWI399193B/en not_active IP Right Cessation
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050084145A1 (en) * | 1999-11-01 | 2005-04-21 | Pelletier Jean P. | Evaluating disease progression using magnetic resonance imaging |
US20090285466A1 (en) * | 2001-11-07 | 2009-11-19 | Medical Metrics, Inc. | Method, Computer Software, And System For Tracking, Stabilizing, And Reporting Motion Between |
US20080292169A1 (en) * | 2007-05-21 | 2008-11-27 | Cornell University | Method for segmenting objects in images |
Non-Patent Citations (1)
Title |
---|
張哲輔,基於細胞自動機之半自動膝蓋軟體影像分割技術,2010年06月 * |
Also Published As
Publication number | Publication date |
---|---|
TW201121505A (en) | 2011-07-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ambellan et al. | Automated segmentation of knee bone and cartilage combining statistical shape knowledge and convolutional neural networks: Data from the Osteoarthritis Initiative | |
CN102737395B (en) | Image processing method and device in a kind of medical X-ray system | |
Xia et al. | Automatic hip cartilage segmentation from 3D MR images using arc-weighted graph searching | |
Marai et al. | Super-resolution registration using tissue-classified distance fields | |
CN104809740A (en) | Automatic knee cartilage image partitioning method based on SVM (support vector machine) and elastic region growth | |
EP3122425A1 (en) | Suppression of vascular structures in images | |
Paulano-Godino et al. | Identification of fracture zones and its application in automatic bone fracture reduction | |
CN105741293A (en) | Method for positioning organs in medical image | |
Jin et al. | Object recognition in medical images via anatomy-guided deep learning | |
JP2004222864A (en) | Diagnosis supporting system, diagnosis supporting method and diagnosis supporting program | |
Felfeliyan et al. | Improved-Mask R-CNN: Towards an accurate generic MSK MRI instance segmentation platform (data from the Osteoarthritis Initiative) | |
Forsberg et al. | Model-based registration for assessment of spinal deformities in idiopathic scoliosis | |
Nikolopoulos et al. | Personalized knee geometry modeling based on multi-atlas segmentation and mesh refinement | |
Chu et al. | Endoscopic image feature matching via motion consensus and global bilateral regression | |
Goutham et al. | Automatic localization of landmarks in cephalometric images via modified U-Net | |
TWI399193B (en) | Measuring total and local changes of knee cartilage volume | |
TW201231017A (en) | Semi-automatic knee cartilage MRI image segmentation based on cellular automata | |
Pang et al. | On the use of coupled shape priors for segmentation of magnetic resonance images of the knee | |
Huang et al. | DA-GAN: learning structured noise removal in ultrasound volume projection imaging for enhanced spine segmentation | |
Mayya et al. | Scapula statistical shape model construction based on watershed segmentation and elastic registration | |
da Silva Tavares | Image processing and analysis: applications and trends | |
Zhang et al. | Blood vessel segmentation based on digital subtraction angiography sequence | |
Pan et al. | Highlight removal for endoscopic images based on accelerated adaptive non-convex RPCA decomposition | |
Mahmoudi et al. | A new approach for cervical vertebrae segmentation | |
CN103854288B (en) | Cruciate ligament segmentation method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
MM4A | Annulment or lapse of patent due to non-payment of fees |