594593 □國外微生物【格式請依··寄存國名；機構；曰期；號碼順序註記】 2. □熟習該項技術者易於獲得，不須寄存。 玖、棚_ 〇 (發明說明應敘明：發明所屬之技術領域、先前技術、1¾容、實施方式及圖式簡單說^ 【發明所屬之技術領域】 本發明係一種配合樣本包埋凝膠的使用，做正、反掃圖後將兩影像 組做影像套合，增加可取像範圍之景深以達到提高顯微鏡影像解析度之 方法。本發明主要但不限於生物顯微影像之立體影像套合。 【先前技術】 習知的共軛焦顯微鏡是藉由去除非焦面所產生的雜訊，以取得在樣 本中不同深度之高解析度顯微影像。其主要原理可分爲三個步驟來說明 :首先雷射光線可以經由物鏡聚焦成單一的光點，來照射樣品單點的特 定深度;另外，由焦點反射或發散的光線還可經由物鏡聚焦成單一的光 束，完全通過影像偵測器前的針孔（pinhole aperture);最後，其他 在焦點之上及之下非焦點所產生的雜訊光子則在此針孔周圍被阻擋下來 ，因而確保偵測器獲取焦面訊號的準確性，進而達到所欲得到之不同深 度高解析度的顯微影像（如圖一所示）。因此，針孔越小，便能去掉越 多的雜訊，而得到專一且淸晰的顯微影像了。 比較共軛焦顯微鏡與傳統式的顯微鏡，前者明顯地具有較多的優點 :在傳統式的螢光顯微鏡中，若要觀察厚片生物組織在Z軸方向的影像 ，僅能侷限地觀察到所使用物鏡景深之內的有限焦面範圍，超出了這個 範圍，非焦面的光線將會嚴重影響焦面的光線，致使對比消弱與影像模 糊，並且，若觀察的樣本同時會散發多種螢光，則欲得到的各種單一螢 光的影像都將混雜有其他光譜螢光的雜訊；而共軛焦顯微鏡則是專對這 594593 些厚片螢光樣本（如生物細胞組織）所設計的，它光學截面的功能可以 大幅減低傳統顯微鏡所無法去除的非焦面性雜訊，在多重螢光的樣本上 ，也能確切地分開不同光譜範圍的螢光訊息，而得到多重螢光厚片組織 之不同深度的淸晰顯微影像。 目前爲增加顯微鏡掃圖的景深，所使用之方法有二種，一爲利用二 台顯微鏡架設於樣本的正面及反面的位置，計算好相對位置後，分別從 正面與反面做共軛取像，所得的影像厚度大約是單台顯微鏡所能得到的 影像組厚度的兩倍，但此做法卻大大地增加所需的硬體成本。其二爲利 用多光子顯微鏡之技術來增加影像組之厚度，其硬體成本亦會較高，且 若將本發明配合多光子顯微鏡使用，可得到的景深約爲單獨使用多光子 顯微鏡景深的2倍。 【發明內容】 本發明爲了得到厚度更深的立體影像，係將樣本做正面掃圖與反面 掃圖，由於樣本在三維空間是被樣本包埋凝膠固定住的，所以正、反掃 圖翻轉前與翻轉後所取得的影像只會發生在XY平面上的平移，與以Z 軸爲軸心旋轉的差異，不會發生在三維空間中Φ角的轉動；找到Z軸上 的重疊位置後，取翻轉前與翻轉後重疊部份中的各一片影像，計算翻轉 前與翻轉後所取得的影像在XY平面上的平移量，與以z軸爲軸心的旋 轉量，再以正面掃圖的影像組當做參考，調整反面掃圖影像的z軸座標 、XY平面的平移量，與以Z軸爲軸心的旋轉量，即可將兩組連續影像 重建成爲一組完整的立體影像。其確定正、反面掃圖影像組在z軸上的 重疊位置，是利用快速傅利葉轉換方法縮小所要尋找Z軸上的重疊位置 的範圍，使用Sobel邊緣檢測的槪念，找出影像中邊緣變化最大的區域 ，再利用此區域使用相關匹配之方法，判斷出反面掃圖內，與正面掃圖 中被選定的影像A最相似的影像（其流程如圖八所示），以判斷正、反 掃圖Z軸上的重疊位置。一旦決定正、反面掃圖影像組在Z軸上的重疊 位置，再使用快速傅利葉轉換方法找出X、Y方向的平移及以Z軸旋轉 的角度，以調整上、下層影像組的位置，使得出的影像爲完整之立體影 第5頁 594593 像0 本發明爲了得到厚度更深的立體影像，乃將樣本做正面掃圖與反面 掃圖；如圖二所示，其中粗線矩形框爲正、反掃圖影像組重疊的區域， 由於樣本被包埋凝膠固定在三維空間中’使得正、反掃圖翻轉前與翻轉 後所取得的影像只會發生在XY平面上的平移’與以z軸爲軸心旋轉的 差異，利用正、反掃圖立體影像重疊部份找出Z軸上的重疊位置，取正 面掃圖與反面掃圖重疊區域內各一片影像’計算正面掃圖與反面掃圖所 取得的影像在XY平面上的平移量，與以Z軸爲軸心的旋轉量，以正面 掃圖的影像組當做參考，調整反面掃圖影像的2軸座標、χγ平面的平 移量，與以Ζ軸爲軸心的旋轉量，如此便可得到一組完整的立體影像。 整個顯微鏡正、反掃圖影像套合系統流程圖如圖三所示。 以下介紹本發明中所使用到的技術與技巧： 1·快速傅利葉轉換應用於影像套合 使用快速傅利葉轉換於影像套合是利用傅利葉轉換的相位關係。若 兩影像僅有平移的差異，即如影像Α是力平移(及，扣)的結果，貝丨J兩影 像關係如式(1)。 式(1)做傅氏轉換後得式⑵，其中兩者之間的位移關係(办，/〇， Λ (叉，>0 = /； 0 — -少〇) a) Ρ2(ξ，η) = β—』2π(ξχ°+ηy。）X ^(ξ，φ (2) 僅出現在相位項一中。 相位項可由式(3)計算得之，其中押爲/的共軛複數。 將式(3)中的相位項做反傅氏轉換後，由反傅式轉換 的計算結果，其脈衝位置即可決定(心，P)。式(4)爲傅氏轉換與反傅 氏轉換對之關係。 F2(€，巧)^ (岑，/7) _ ^]2π{ξ χ0+η y0) (3) S(x-x0,y-y0)<^e~J2^x^^ ⑷ 594593 當兩張影像同時發生旋轉與平移時如式(5)。 /2(^^)=/1(^03¾ +^8111¾ -xsin^o +^008¾ -y) (5) 式(5)取快速傅利葉轉換後，兩影像在頻域的相對關係如式(6) ο F2G，ri) = e-j2—x ⑹ G (f cos % + ；7 sin %，一f sin % + 77 cos ) 式(6)取其振幅麂、邋如式⑺所示，其中必、舶分別爲/l、/2的振 幅。觀察式(7)，#2是必旋轉了仏度的結果’對你、泌做極座標轉換後 得式(8)。 Μ2 (ξ,η) = Μχ(ξ cos + ^sin θ0 -ξsin (90 + ^cos θ0) ⑺ Μ x{p ,θ) ^ Μ 2{ρ,θ - θ0) ⑻ 觀察式(8)，舶與你之間變成平移的關係’其平移量爲仏’利用上述傅利 葉轉換的相位關係就可解得仏。接下來將Ζ旋轉仏後得/”1，广1與Α之 間就只剩平移的關係’再利用傅利葉轉換的相位關係就可得知〇W〇)， 整個執行的過程請參照圖四所示。影像做傅氏轉換後的高頻部份包含原 始圖形的重要特徵’所以在做極座標轉換前加入高通瀘波器’提高角度 評估的準確度，其規格如式⑼’其中仏=0·3。 r" i〇// D(u,v)<D0 {U^V)ZZ\\if D(u9v)>D0 (9) 2. Sobel邊緣檢測 本發明使用Sobel運算子做邊緣檢測，在Sobel運算子對影像/做完 邊緣檢測後，影像/上的每個像素點會得到梯度強度大小Μ，圖五(b)爲 Sabel運算子對圖五(a)做完邊緣檢測後之梯度強度影像，其梯度強度値越 大(影像上越亮），代表該處是越明顯的邊緣’即影像明亮變化大的地方， 本發明利用影像，完成邊緣檢測後的梯度強度影像，尋找影像，內有明顯 邊緣變化的區域。設定一適當大小的區塊，尋找影像/的梯度強度圖中在 第7頁 594593 區塊內|v/|總和最大的位置，此位置爲影像/內有明顯邊緣變化的區域如 圖五(c) ’利用此區域當做樣板，突顯其餘待比對之影像與樣板間的差異 性，以利後續做相關匹配之用。 3·相關匹配方法 利用相關匹配之方法如式(10)所得的相關係數Ks，〖），可在影像/ 中找出與樣板π最相似的區域。Kz，/)樣板(template)在Μ xN的影像/ U，7)內由左而右，由上而下移動使用式(10)得到rCs，ί)相關係數，其 中s-0，1，2，…Μ-1，卜0，1，2，…Ν-1，7爲影像/與樣板π重疊的 區域，即式(2-1)只在重疊區域進行，TUj；)代表影像/與樣板w重疊的 區域亮度平均値，5代表樣板r亮度平均値。 如圖六，其啦)/U，y)的原點位於左上角，功）y)的原點位於它的 中心，位於/U，y)內的任何位置，ί)都使用式(10)得到一相關係數r Σ Σ [你，w -你，少)]卜(x - J^ - - w] r(")= 為 一_=_ (1〇) ’ {Σ Σ [你，>"卜 /(x，洲2 Σ Σ y - w]2)x594593 □ Foreign microorganisms [Please follow the format of the country name of the deposit; organization; date; note of number sequence] 2. □ Those who are familiar with the technology are easy to obtain and do not need to deposit.玖 、 舍 _ 〇 (The description of the invention should state: the technical field to which the invention belongs, the prior art, the features, the embodiments, and the drawings briefly. [Technical field to which the invention belongs] The present invention is a kind of The method is used to fit the two image groups after the forward and reverse scans, and increase the depth of field of the available image range to achieve a method for improving the resolution of the microscope image. The present invention is mainly but not limited to the stereo image registration of biological microscopic images. [Prior technology] The conventional conjugate focus microscope removes the noise generated by the non-focus surface to obtain high-resolution microscopic images at different depths in the sample. The main principle can be divided into three steps to explain : First, the laser light can be focused into a single light spot through the objective lens to illuminate a specific depth of a single point of the sample; In addition, the light reflected or diverged by the focus can also be focused into a single beam through the objective lens, completely passing through the front of the image detector Pinhole aperture; finally, other noise photons generated above and below the focus are blocked around this pinhole, thus ensuring detection The detector acquires the accuracy of the focal plane signal, and then achieves the desired high-resolution microscopic images of different depths (as shown in Figure 1). Therefore, the smaller the pinhole, the more noise can be removed, and Obtain a specific and clear microscopic image. Comparing the conjugate focus microscope with the traditional microscope, the former obviously has more advantages: In the traditional fluorescence microscope, if you want to observe the thick biological tissue in the Z axis Images in the directional direction can only observe the limited focal plane range within the depth of field of the objective lens used. Beyond this range, non-focus plane light will seriously affect the focal plane light, resulting in weak contrast and blurred images. If the observed sample emits multiple kinds of fluorescent light at the same time, the images of various single fluorescent lights that are to be obtained will be mixed with the noise of other spectral fluorescent lights; while the conjugate focus microscope is specifically designed for these 594593 thick slices of fluorescent light. Specimen (such as biological cell tissue) is designed, its optical cross-section function can greatly reduce the non-focal surface noise that can not be removed by traditional microscope, and it can also be exact on multi-fluorescence samples. Open the fluorescence information of different spectral ranges to obtain clear microscopic images of different depths of multiple fluorescent slabs. At present, to increase the depth of field of the microscope scan, there are two methods used. One is to use two microscopes. Set up the positions of the front and back of the sample. After calculating the relative positions, take conjugate images from the front and back respectively. The thickness of the obtained image is about twice the thickness of the image group that can be obtained by a single microscope. However, it greatly increases the required hardware cost. The second is to increase the thickness of the imaging group by using the technology of multi-photon microscopy. The hardware cost will also be higher, and if the present invention is used with a multi-photon microscope, it can be obtained. Depth of field is about twice the depth of field of a multi-photon microscope alone. [Abstract] In order to obtain a deeper stereo image, the present invention scans the front and back of the sample. Because the sample is embedded and condensed in the three-dimensional space The glue is fixed, so the images obtained before and after the flip and reverse scans will only occur in the translation on the XY plane, with the Z axis as the axis The difference in rotation does not occur in the rotation of the Φ angle in the three-dimensional space; after finding the overlapping position on the Z axis, take one of the images in the overlapping part before and after the flip, and calculate the images obtained before and after the flip. The amount of translation in the XY plane and the amount of rotation with the z axis as the axis, and then use the image group of the front scan as a reference to adjust the z-axis coordinates of the back scan image, the amount of translation in the XY plane, and the Z axis. The amount of rotation of the axis can reconstruct two sets of continuous images into a set of complete stereo images. It determines the overlapping position of the positive and negative scanning image groups on the z-axis. It uses the fast Fourier transform method to narrow the range of the overlapping position on the z-axis that is to be found. It uses Sobel edge detection to find the largest edge change in the image. Area, and then use this area to use the relevant matching method to determine the image in the back scan that is most similar to the selected image A in the front scan (the process is shown in Figure 8) to determine the positive and negative scans. Overlapping position on the Z axis of the graph. Once the overlapping position of the positive and negative scan image groups on the Z axis is determined, the fast Fourier transform method is used to find the translation in the X and Y directions and the angle of rotation in the Z axis to adjust the position of the upper and lower image groups so that The output image is a complete three-dimensional image. Page 5594593 Image 0 In order to obtain a deeper three-dimensional image, the sample is scanned from the front and the back; as shown in Figure 2, the thick rectangular frame is The area of the overlapping scan image group, because the sample is fixed in the three-dimensional space by the embedding gel, 'the images obtained before and after the forward and reverse scans will only be translated on the XY plane' and z The axis is the difference of the axis rotation. The overlapping position of the three-dimensional image of the positive and negative scans is used to find the overlapping position on the Z axis. Take one image in the overlapping area of the front scan and the back scan to calculate the front scan and the back scan. The amount of translation of the image obtained in the image on the XY plane, and the amount of rotation with the Z axis as the axis, and the image group of the front scan as a reference, adjust the 2 axis coordinates of the back scan image, and the amount of χγ plane translation. In Ζ axis rotation amount as the axis, can thus obtain a complete stereoscopic image. The flow chart of the whole microscope positive and negative scanning image fitting system is shown in Figure 3. The following introduces the techniques and techniques used in the present invention: 1. Fast Fourier transform applied to image fitting The use of fast Fourier transform for image fitting uses the phase relationship of Fourier transform. If there is only a difference in translation between the two images, that is, if image A is the result of force translation (and, deduction), the relationship between the two images is as shown in equation (1). Equation (1) is obtained after Fourier transformation, where the displacement relationship between the two is ((0, 〇, Λ (fork, > 0 = /; 0--less 〇) a) P2 (ξ, η ) = β—′2π (ξχ ° + ηy.) X ^ (ξ, φ (2) only appears in the phase term 1. The phase term can be calculated from equation (3), where the complex conjugate number is /. After inverse Fourier transform of the phase term in equation (3), the pulse position can be determined from the calculation result of inverse Fourier transform (heart, P). Equation (4) is Fourier transform and inverse Fourier transform The relationship between them. F2 (€, clever) ^ (cen, / 7) _ ^] 2π {ξ χ0 + η y0) (3) S (x-x0, y-y0) < ^ e ~ J2 ^ x ^ ^ ⑷ 594593 When the two images are rotated and translated at the same time, the equation (5) is used. / 2 (^^) = / 1 (^ 03¾ + ^ 8111¾ -xsin ^ o + ^ 008¾ -y) (5) Equation (5) After taking a fast Fourier transform, the relative relationship between the two images in the frequency domain is as shown in equation (6 ) ο F2G, ri) = e-j2—x ⑹ G (f cos% +; 7 sin%, f sin% + 77 cos) Eq. (6) Take the amplitude as shown in Eq. , And the amplitude of / l and / 2, respectively. Observing equation (7), # 2 is the result that the degree must be rotated ’After performing polar coordinate transformations on you and Meng, we get equation (8). Μ2 (ξ, η) = Μχ (ξ cos + ^ sin θ0-ξsin (90 + ^ cos θ0) ⑺ Μ x (p, θ) ^ Μ 2 {ρ, θ-θ0) ⑻ Observation formula (8), The relationship of translation with you 'its translation is 仏' can be solved by using the phase relationship of the Fourier transform described above. Next, rotate the Z to get / "1, and there is only translational relationship between Guang1 and A ', and then use the phase relationship of the Fourier transform to know 〇W〇). Please refer to Figure 4 for the entire execution process. The high-frequency portion of the image after Fourier transform contains important features of the original graphic 'so adding a high-pass chirp before the polar coordinate conversion' improves the accuracy of the angle evaluation, and its specifications are as follows: 'where 仏 = 0 · 3. r " i〇 // D (u, v) < D0 {U ^ V) ZZ \\ if D (u9v) > D0 (9) 2. Sobel edge detection The present invention uses Sobel operator for edge detection After the Sobel operator performs edge detection on the image /, each pixel on the image / will get the gradient intensity M. Figure 5 (b) is the result of the Sabel operator after edge detection on Figure 5 (a). Gradient intensity images, the larger the gradient intensity (the brighter the image), the more obvious the edge is, that is, the place where the image changes brightly. The present invention uses the image to complete the gradient intensity image after edge detection to find the image. There are areas with obvious edge changes. Set an appropriate size block to find the image / The gradient intensity map is the position where the sum of | v / | is the largest in the 545993 block on page 7. This position is the area with obvious edge changes in the image /. Figure 5 (c) 'Use this area as a template to highlight the rest of the comparison. The difference between the image and the template is used to facilitate subsequent correlation matching. 3. Correlation matching method The correlation coefficient method, such as the correlation coefficient Ks obtained by equation (10), can be found in the image / The area most similar to the template π. The Kz, /) template in the image of M xN / U, 7) moves from left to right, and moves from top to bottom. Use equation (10) to obtain the rCs,) correlation coefficient, Among them, s-0,1,2, ... M-1, Bu0,1,2, ... N-1,7 are the areas of the image / overlapping with the template π, that is, formula (2-1) is performed only in the overlapping area. TUj;) represents the average brightness of the image / area overlapping with the template w, 5 represents the average brightness of the template rr. As shown in Figure 6, the origin of / U, y) is located in the upper left corner, and the origin of y) It is located at the center of it, anywhere in / U, y), and ί) uses equation (10) to get a correlation coefficient r Σ Σ [you, w-you, less]] Bu (x-J ^--w ] r (") = Is a _ = _ (1〇) ’{Σ Σ [You, > " Bu / (x, continent 2 Σ Σ y-w) 2) x
xyeT xyeT (r i)，而相似度數値最高的位置即是/U，7)中與Kr 7)最相似的區 域，相關係數，ί)的取値範圍爲-。 4.確定正、反面掃圖影像組在Ζ軸上的重疊位置xyeT xyeT (r i), and the position with the highest degree of similarity 値 is the area most similar to Kr 7) in / U, 7). The correlation coefficient,)), is taken as-. 4. Determine the overlapping position of the positive and negative scan image groups on the Z axis
確定Ζ軸上的重疊位置就是要找到兩影像組中相同的影像，以確定重 疊區域的位置。第一步是利用上述快速傅利葉轉換方法計算兩影像間旋轉 角度的過程中所得到的峰値(peak)來做判斷，因爲當兩張極相似的影像(例 如：影像A與影像B極相似)之間的差異爲旋轉與平移’所得到的峰値會比 將影像B換成影像C時(影像A與影像C不相似)做相同的運算所得的峰値 要來得高許多，所以在反面掃圖影像組中與正面掃圖影像A較相似的幾片 影像位於最突出峰値的附近，如圖七最突出峰値位於粗線矩形框內，所以 與影像A較相似的幾片影像位於反面掃圖的第11片附近（此處定義這幾片 影像爲「影像群K」，以方便後文述敘），利用此觀念找出粗略的定位Z 第8頁 594593 軸上的重疊位置，縮小所要比對的影像片數，以減少後序的計算量。然而 這幾片影像都有可能是反面掃圖中與正面掃圖的影像A最相似的影像。計 算從反面掃圖中所選取的影像群K每一張影像與影像A在X Y平面上的平 移量與以Z軸旋轉的角度，調整其平移量與旋轉的角度，再利用上述所提 之Sobel邊緣檢測的槪念，找出影像中邊緣變化最大的區域，利用此區 域使用相關匹配之方法，判斷出反面掃圖內與正面掃圖中的影像A最相似 的影像，決定正、反面掃圖影像組在Z軸上的重疊位置，其過程如圖八所 示。最後，定位了 Z軸上的重疊位置後，取正面掃圖與反面掃圖重疊區域 內各一片影像，計算正面掃圖與反面掃圖所取得的影像在XY平面上的平 移量與以Z軸的旋轉量，以正面掃圖的影像組當做參考，調整反面掃圖影 像的Z軸座標、XY平面的平移量，與以Z軸爲軸心的旋轉量。 【實施方式】 [第一實施例] 圖九爲生物樣本正、反面掃圖之表面圖，取正面掃圖的最下面一張影 像與反面掃圖的每一張影像，利用快速傅利葉轉換方法計算兩影像間旋轉 角度的過程中所得到的峰値(peak)來做判斷，找出正面掃圖的最下面一張 影像相對於反面掃圖影像組在Z軸上的大致位置（如圖七所示）。再利用 上述所提之Sobel邊緣檢測的槪念，找出影像中邊緣變化較大的區域如 圖五(c)的粗線矩形框即爲影像中邊緣變化較大的區域。利用此區域使用 相關匹配之方法，判斷出反面掃圖內與正面掃圖中的影像A最相似的影像 。決定正、反面掃圖影像組在Z軸上的重疊位置。整個確定Z軸上的重疊 位置流程係如圖八所示。在確定Z軸上的重疊位置後，取正面掃圖與反面 掃圖重疊區域內各一片影像，計算正面掃圖與反面掃圖所取得的影像在X Y平面上的平移量，與以Z軸爲軸心的旋轉量（如圖四所示），再以正面 掃圖的影像組當做參考，調整反面掃圖影像的Z軸座標、XY平面的平移 量，與以Z軸爲軸心的旋轉量，如此便可得到一組完整的立體影像。圖十 爲此生物樣本正、反掃圖做影像套合後之完整立體影像，其厚度約爲正面 掃圖的2倍。 第9頁 594593 [第二實施例] 果蠅的大腦厚度約160#m，將腦神經細胞標示綠色瑩光蛋白，利用4 8 8 n m雷射去激發它可得到完整的3 D大腦影像。但我們可淸楚的發現 雷射掃描所得的影像越到底部就越模糊了。其主要原因爲生物樣本具有吸 光性，激發光的能量或放射光的能量被樣本吸收，而導致所取得的立體影 像在某個深度以下是非常不淸楚的。利用本發明所提之方法，僅需掃描至 稍微超過一半腦的深度得到正面掃圖與反面掃圖之淸晰立體影像組後再做 影像套合即可得到完整而淸晰的3 D大腦影像。 [第三實施例Π 一般作共軛焦顯微影像掃描是將生物組織薄片包埋於甘油中去進行顯 微影像掃描及記錄。配合本發明之方法，可描掃之生物組織薄片厚度可提 高至接近原先生物組織厚度的二倍，即先將生物組織用樣本包埋凝膠固定 於三維空間後，利用共軛焦顯微鏡掃描至稍微超過一半腦的深度得到正面 掃圖與反面掃圖之淸晰立體影像組後再做影像套合即可得到完整而淸晰的 3 Ό生物組織影像，其立體影像厚度約爲一般包埋於甘油中進行顯微影像 掃描所得影像厚度的二倍。 [第四實施例Π 一般作共軛焦顯微影像掃描是將生物組織薄片包埋於甘油中去進行顯 微影像掃描及記錄。若改使用FocusClear澄淸組織再包埋於MountClear™ 則可大大加深可描掃之景深。若再配合本發明之方法，所得之立體影像厚 度可再提高一倍，爲單配合FocusClear澄淸組織再包埋於Mmmtciear™之 方法的二倍。 [第五實施例] 爲取得老鼠全腦之影像必需先將成鼠的大腦做震動切片，傳統的作法 爲將約4 mm厚的成鼠的腦切成1G0-2GG片lG-2G//m的薄片包埋於甘油 第10頁 594593 中去進行顯微影像掃描及記錄。如果使用FocusClear澄淸組織再包埋於 MountClear™，則可掃描約200 //m的厚片。若再利用本發明之方法則可得到 400 //m厚片組織內的淸楚影像，而一個完整成鼠腦貝(J僅需切成約1〇片400 # m厚的組織片，故利用本發明之方法不但可以有效的增加立體影像的景深 深，且在厚度大之生物組織如成鼠的全腦可以有效的減少切片數。 [第六實施例] 配合多光子顯微鏡的使用，利用本發明所提之方法，先利用多光子顯 微鏡做正面掃圖與反面掃圖再做影像套合，即可得到立體影像的厚度（景 深）約爲單使用多光子顯微鏡的兩倍。也就是說，一個完整成鼠腦約4mm 僅需切成5片800 //m的厚片，包埋於MountClear™中，正、反面掃圖後再利 用本發明做影像套合，即可得到完整大腦的顯微影像。 【圖式簡單說明】 圖一係顯示本發明利用雷射光線經由物鏡聚焦成單一的光點，來照射樣品 單點的特定深度。 圖二係顯示本發明將樣本做正面掃圖與反面掃圖之示意圖。 圖三係顯示本發明整個顯微鏡正、反掃圖影像套合系統流程圖。 圖四爲顯示本發明做傅氏轉換整個執行的過程。 圖五（a )爲顯示本發明做邊緣檢測之影像。 圖五（b )爲顯示Sobel運算子對圖五(a)做完邊緣檢測後之梯度強度影像。 圖五（c )爲顯示Sobel運算子做完邊緣檢測後影像/內有明顯邊緣變化 的區域。 圖六爲顯示本發明利用相關匹配之方法所得的相關係數，可在影像/中找 出與樣板π最相似的區域。 圖七爲顯示本發明在反面掃圖影像組中最突出峰値的位置。 圖八係顯示本發明確定正、反面掃圖影像組在Ζ軸上的重疊位置之流程圖。 圖九爲顯示生物樣本正、反面掃圖之表面圖。 圖十爲顯示生物樣本正、反掃圖做影像套合後之完整立體影像。 第11頁Determining the overlapping position on the Z axis is to find the same image in the two image groups to determine the position of the overlapping area. The first step is to use the above-mentioned fast Fourier transform method to calculate the peak value (peak) obtained during the process of calculating the rotation angle between the two images, because when two extremely similar images (for example: image A and image B are very similar) The difference is that the peak value obtained by rotation and translation will be much higher than the peak value obtained when the same operation is performed when image B is replaced by image C (image A and image C are not similar), so scan on the reverse side. In the image group, the several images that are more similar to the front scan image A are located near the most prominent peaks. As shown in Figure 7, the most prominent peaks are located in the thick rectangular frame. Therefore, the several images that are more similar to the image A are on the reverse side. Scan the vicinity of the 11th image (these images are defined here as "image group K" for the convenience of later description), use this concept to find the rough positioning Z Page 8 594593 axis overlap position, zoom out The number of images to be compared to reduce the amount of subsequent calculations. However, these images may be the images most similar to the image A in the front scan. Calculate the amount of translation of each image and image A on the XY plane and the angle rotated by the Z axis in the image group K selected from the back scan, adjust the amount of translation and rotation angle, and then use the Sobel mentioned above The idea of edge detection, find the area with the largest edge change in the image, and use this area to use the relevant matching method to determine the image in the back scan that is most similar to the image A in the front scan, and determine the front and back scans The overlapping position of the image group on the Z axis is shown in Figure 8. Finally, after positioning the overlapping position on the Z axis, take one image in the overlapping area of the front scan and the back scan, and calculate the translation amount of the image obtained by the front scan and the back scan on the XY plane and the Z axis. The amount of rotation is based on the image group of the front scan image as a reference, and the Z-axis coordinate of the back scan image, the translation amount of the XY plane, and the rotation amount around the Z axis are adjusted. [Embodiment] [First Example] FIG. 9 is a surface view of a front and back scan of a biological sample. The bottom image of the front scan and each image of the back scan are taken and calculated using a fast Fourier transform method. The peak obtained during the rotation angle between the two images is used to determine the approximate position of the bottom image of the front scan relative to the image group of the back scan on the Z axis (as shown in Figure 7). Show). Then use the above-mentioned idea of Sobel edge detection to find areas with large edge changes in the image. The thick rectangular frame in Figure 5 (c) is the area with large edge changes in the image. Use this area to use the correlation matching method to determine the image in the back scan that is most similar to the image A in the front scan. Determines the overlap position of the positive and negative scan image groups on the Z axis. The whole process of determining the overlapping position on the Z axis is shown in Figure 8. After determining the overlapping position on the Z axis, take one image in the overlapping area of the front scan and the back scan, calculate the translation amount of the image obtained by the front scan and the back scan on the XY plane, and use the Z axis as the The amount of rotation of the axis (as shown in Figure 4), and then use the image group of the front scan as a reference to adjust the Z-axis coordinates, the translation amount of the XY plane, and the amount of rotation around the Z axis. , So you can get a complete set of stereo images. Figure 10 A complete stereo image of the front and back scans of this biological sample. The thickness is about twice that of the front scan. Page 9 594593 [Second Example] The brain thickness of Drosophila is about 160 # m. The brain nerve cells are marked with green fluorescent protein, and a 4 8 8 n m laser is used to excite it to obtain a complete 3D brain image. But we can clearly find that the image obtained by laser scanning becomes more blurred as it reaches the bottom. The main reason is that the biological sample is absorbent. The energy of the excitation light or the emitted light is absorbed by the sample, and the resulting stereo image is very unsightly below a certain depth. Using the method provided by the present invention, only a scan to more than half the depth of the brain is required to obtain the clear stereo image group of the front scan and the back scan, and then the image is combined to obtain a complete and clear 3D brain image. . [Third Embodiment Π Generally, conjugate focus microscopic image scanning is performed by embedding a biological tissue sheet in glycerin for scanning and recording of microscopic images. Combined with the method of the present invention, the thickness of the scannable biological tissue sheet can be increased to approximately twice the thickness of the original biological tissue, that is, the biological tissue sample embedding gel is first fixed in a three-dimensional space, and then scanned by a conjugate focus microscope to Obtain a clear stereo image group of the front scan and the back scan slightly more than half the depth of the brain, and then perform image fitting to obtain a complete and clear 3Ό biological tissue image. The thickness of the stereo image is approximately embedded in The image obtained by scanning the micrograph in glycerol was twice as thick. [Fourth Embodiment Π Generally, conjugate focus microscopic image scanning is performed by embedding a biological tissue sheet in glycerin for scanning and recording of microscopic images. If you use FocusClear to embed tissue in MountClear ™, the depth of field can be greatly increased. If combined with the method of the present invention, the thickness of the obtained stereo image can be doubled again, which is twice as high as the method with the FocalClear tissue alone and then embedded in Mmmtciear ™. [Fifth embodiment] In order to obtain an image of the whole brain of a rat, it is necessary to slice the brain of an adult rat first. The traditional method is to cut an adult rat brain of about 4 mm thickness into 1G0-2GG pieces 1G-2G // m The thin slices were embedded in glycerol on page 10 594593 for microscopic image scanning and recording. If FocusClear tissue is used for embedding in MountClear ™, a thick slice of about 200 // m can be scanned. If the method of the present invention is re-used, the image of the chrysanthemum in a 400 // m thick piece of tissue can be obtained, and a complete adult rat brain shell (J only needs to be cut into about 10 pieces of 400 #m thick tissue pieces, so use The method of the present invention can not only effectively increase the depth of field of a stereo image, but also effectively reduce the number of slices in a thick biological tissue such as an adult rat's whole brain. [Sixth embodiment] With the use of a multi-photon microscope, the present invention is used. The method mentioned in the invention firstly uses a multi-photon microscope to perform a front scan and a back scan, and then images are combined, and the thickness (depth of field) of the stereo image is about twice that of the single-photon microscope. That is, A complete adult mouse brain is about 4mm. It only needs to be cut into 5 800 // m thick slices and embedded in MountClear ™. After scanning the front and back sides and using the present invention for image fitting, the complete brain display can be obtained. Micro image. [Schematic description] Figure 1 shows that the present invention uses laser light to focus into a single spot through the objective lens to illuminate a specific depth of a single point of the sample. Figure 2 shows that the present invention scans the sample with a frontal scan and Scanning the back Schematic diagram. Figure 3 shows the flow chart of the entire microscope positive and negative scanning image fitting system of the present invention. Figure 4 shows the entire process of performing Fourier transform of the present invention. Figure 5 (a) shows the edge detection of the present invention. Figure 5 (b) shows the gradient intensity image of Sobel operator after edge detection in Figure 5 (a). Figure 5 (c) shows the obvious edge change in image / inside of Sobel operator after edge detection. Figure 6 shows the correlation coefficient obtained by using the correlation matching method of the present invention, and the region most similar to the template π can be found in the image /. Figure 7 shows the most prominent peak in the negative scan image group of the present invention. The position of puppet. Figure 8 shows the flow chart of determining the overlapping position of the front and back scan image groups on the Z axis according to the present invention. Figure 9 is a surface view showing the front and back scans of a biological sample. Figure 10 shows the biological sample. The positive and negative scans are used to complete the stereo image after image fitting. Page 11