1324750 九、發明說明： 【發明所屬之技術領域】 本發明係提供一種微流體細胞之顯微影像分析與判 讀系統，特別是指一種可將待測微流體細胞經由顯微影像 分析與判讀後可容易辨識細胞的位置及其顆粒數目，其主 要係經由顯微影像擷取程序、影像分析與判讀程序、顯示 程序等步驟，即可將待測微流體細胞顯微影像輸入並依據 影像的特性進行不同的影像前處理，以辨識每張影像中細 φ 胞的位置及顆數，並對序列影像自動計數流動細胞的數目。 【先前技術】 細胞，是人體組織中最基本的單位，在自然和病理現 象上，常經由觀察細胞的現象做為臨床上的研究，比如說 利用細胞計數的功能計算尿液中尿蛋白的個數，以求出濃 度來診斷疾病是否為陽性（比如：腎病症後群、腎絲球炎、 懷孕等等；計算血液中白血球的濃度，檢查是否有發炎的 • 情沉。目前臨床上觀察細胞行為與計數的方式，大約分為 利用人工及儀器檢測兩種，人工方式進行計數，不僅耗費 時間而且判定的準確性及客觀性亦受到賀疑，儀器檢測常 使用流式細胞儀，由於造價昂貴，並且操作繁覆，無法達 到普及的效果。再如現階段於科學期刊上所揭露之技術， 得知目前用於濃度計算的微流體晶片，方法有光學檢測及 電阻檢測法兩種，光學檢測法有製作能力上的限制，比如 說光纖通道大小大約40um，無法準確檢測2um的生物微 5 1324750 粒，電阻法則有不易測量的缺點，由於量測到的電阻值極 小，容易受雜訊的干擾而產生誤判，所以測量時必需有良 好的屏蔽裝置來進行測量，這也是不易實現的地方。復如 •台灣新型M290504號專利，其提出一種偵測微流體晶片之 流式細胞儀，係利用光源投射在偵測區段，使流經該區段 的細胞被激發出一定波長的螢光，然後拍攝出細胞檢體的 影像訊號。另如台灣公告第593677號發明專利’其係利用 待測物與螢光染料結合，分析流體的狀態與分選待測物。 φ前揭已知方法主要用在偵測微流體細胞的偵測和取像、分 析待測物的生物參數、控制樣品流動的方向與分選。因此’ 若能加入影像分析技術及計數方法，提供臨床或研究人員 更多的分析資訊。 另由相關領域中之學者Chen的光學影像自動細胞計 數方法中可歸納出細胞顯微影像的幾個特性：雜質存在的 問題、細胞形狀的多形性、細胞重疊的問題、細胞像素有 較低的灰階值特徵、細胞常黏在一起等。其解決方法包括 ⑩（1)利用空間濾波器來做濾除雜訊的功能，（2)使用一階微分 的遮罩偵測像素點較低位置（即細胞邊緣），（3)依據影像 中較低灰階值位置的周圍，訂出一個特定臨界值，來描述 出細胞的位置以及形狀，（4)再利用形心運算及面積大小對 細胞做篩選，找出正端的細胞。然而Chen的研究主要應用 於靜態.影像的細胞計數，無法直接應用於微流體細胞的計 數。台灣新型M274600號專利另揭示出一種影像分析計數 裝置，包括一取像模組、一取樣表、一顯示裝置，此方法 必須預先建立物體的特徵值並儲存於記憶裝置中，再利用 6 1324750 模糊比對原理進行預設特徵值與物體輪廓特徵值比對，以 顯示器顯示比對結果，並未提到連續序列影像的細胞計數 方式。 統觀習知方法有者僅能對靜態影像的細胞計數， 直接對微流體細胞進行計數，有者無法提供連續序列影像 的=胞°十數，其存在較大之計算誤差。本發明人即是有鑑 = 有其使用上之缺失，期以-種可提供臨床人 體細胞能進行顯微影像分析 化計數之效電腦輔助分析可達成細胞定位及自動 >皿，俾讓這些自動化的處理可 擔，降低主顴划此-T* X罕丄入马的負 .,,aa Λ 斷可能造成的誤判，提高臨床研究的效益。 經由發明人精心研發，終乃發展出本發明。 【發明内容】 緣是： 本發明之主^ 影像分析與判讀糸祐'，/、係提供一種微流體細胞之顯微 微影像擷取、与德，、、’該系統係將待測微流體細胞經由顯 胞的位置及顆=析與判讀，即可精切辨識出微流體細 本發明之另— 顯微影像分析與判^要目的’其係提供一種微流體細胞之 該程序可H由㈣^統’线含括有1像擷取程序， 影像；一影攝影機拾取待測微流體細胞之 影像識別處理及白^項程序，可對待測微流體細胞進行 至很低，有效=計數細胞顆粒’能將系統的誤差率降 欢地進行細胞計數。 7 I324750 有關本發明之實質内容及其功效，可佐以圖示詳加說 明如后。 【實施方式】 凊先參考第一及二圖，本發明微流體細胞之顯微影像 分析與判讀系統，該系統係可將待測微流體細胞經由顯微 影像擷取程序、影像分析與判讀程序、顯示程序等步驟完 成。顯微影像擷取程序係利用包括光學顯微鏡、CCD攝影 鲁機作為影像擷取裝置。亦即待測微流體細胞經顯微影像放 大後，以CCD攝影機拍攝，微流體細胞顯微影像經類比轉 數位影像裝置轉成數位影像後，可儲存到個人電腦中。 顯微影像分析與判讀程序，前揭被拾取之待測微流體 ' 細胞之顯微影像依其特性，可結合不同影像處理與分析的 步驟包括細胞識別影像處理分析和自動計數演算方法，以 進行細胞顯微影像的判讀和細胞計數，本發明系統同時設 計一使用者介面以呈現序列影像與計數的結果。請參考第 φ二及三圖，待測微流體細胞經影像讀取後可依第三圖所示 微流體細胞識別影像處理分析程序之流程圖進行處理。其 中： 1·色版轉換：原始微流體影像為RGB彩色影像，由於 RGB影像包含了亮度(Luminance)的資訊，因此容易受 到光線強弱的影響，而造成影像亮度不均的現象，影 響對感興趣之顏色的判斷’若未經過影像前處理，會 影響到分析微粒位置和計數的正確率，在RGB影像上 若要改善背景亮度不均的問題，必須分別針對三個色 8 版做分析’是相當麻填且不易實現的。因此為了處理 這個問題，將RGB之彩色影像轉換至對光線變化較 不敏感的色彩空間’例如將RGB色版轉換至HSL(HUe，1324750 IX. Description of the Invention: [Technical Field] The present invention provides a microscopic image analysis and interpretation system for microfluidic cells, in particular, a microfluidic cell can be analyzed and interpreted by microscopic images. It is easy to identify the location of the cells and the number of particles. The microscopic image of the microfluidic cells to be tested can be input and processed according to the characteristics of the image, mainly through microscopic image acquisition procedures, image analysis and interpretation procedures, and display procedures. Different image pre-processing to identify the position and number of fine φ cells in each image, and automatically count the number of flowing cells in the sequence image. [Prior Art] Cells, which are the most basic units in human tissues, are often studied clinically by observing the phenomenon of cells in natural and pathological phenomena. For example, the function of counting cells is used to calculate the urine protein in urine. Count, to determine the concentration to diagnose whether the disease is positive (such as: kidney disease group, kidney glomerulitis, pregnancy, etc.; calculate the concentration of white blood cells in the blood, check for inflammation • affection. Currently clinical observation of cells The way of behavior and counting is roughly divided into two types, manual and instrumental detection. Manual counting is not only time-consuming but also the accuracy and objectivity of the judgment. The instrumentation often uses flow cytometry because of the high cost. And the operation is complicated, and the effect of popularization cannot be achieved. As the technology disclosed in the scientific journal at this stage, the microfluidic wafer currently used for concentration calculation is known as optical detection and resistance detection, and optical detection The method has limitations in production capabilities. For example, the size of the Fibre Channel is about 40um, and it is impossible to accurately detect 2um of biological micro- 5 1324750 particles, the resistance law has the disadvantage of being difficult to measure. Because the measured resistance value is extremely small, it is easy to be misjudged by the interference of noise, so it is necessary to have a good shielding device to measure when measuring, which is not easy to achieve. Furu • Taiwan's new M290504 patent, which proposes a flow cytometer for detecting microfluidic wafers, which uses a light source to project on the detection section, so that cells flowing through the section are excited to emit fluorescence of a certain wavelength. Then, the image signal of the cell sample is taken. Another example is the invention patent of Taiwan Publication No. 593677, which uses the test object to combine with the fluorescent dye to analyze the state of the fluid and sort the sample to be tested. It is mainly used to detect the detection and imaging of microfluidic cells, analyze the biological parameters of the analyte, and control the direction and sorting of the sample flow. Therefore, if you can add image analysis technology and counting methods, provide clinical or research personnel. A lot of analysis information. In addition, Chen's optical image automatic cell counting method in the related field can summarize several characteristics of cell microscopic images: miscellaneous Problems, cell shape pleomorphism, cell overlap problems, cell pixels have lower gray-scale values, cells are often stuck together, etc. The solution includes 10(1) using spatial filters for filtering The function of the noise, (2) using a first-order differential mask to detect the lower position of the pixel (ie, the cell edge), and (3) setting a specific threshold based on the position of the lower gray-scale value in the image, To describe the location and shape of the cells, (4) to use the centroid calculation and area size to screen the cells to find the positive cells. However, Chen's research is mainly applied to the static cell count of images, which cannot be directly applied to micro. The counting of fluid cells. Taiwan's new M274600 patent further discloses an image analysis and counting device, which comprises an image capturing module, a sampling table and a display device. The method must pre-establish the feature values of the object and store them in the memory device. Then use the 6 1324750 fuzzy comparison principle to compare the preset eigenvalue with the contour eigenvalue of the object, and display the comparison result on the display. No continuous sequence image is mentioned. Cell counting. The conventional method can only count the cells of the static image and directly count the microfluidic cells. Some of them cannot provide the continuous sequence image = the number of cells, which has a large calculation error. The present inventors have the knowledge that there is a lack of use, and that the kind of clinical human body cells can provide microscopic image analysis and the effect of computer-aided analysis can achieve cell localization and automatic > Automated processing can be used to reduce the negative impact of the main 颧 此 - T T T , , , , , , , , , , , , , , , , , , , , , , , , , , , The invention was developed through careful research and development by the inventors. SUMMARY OF THE INVENTION The edge is: The main image analysis and interpretation of the present invention, Yuyou', /, provides a micro-microscopic image of microfluidic cells, and the German,,,,, Through the position and particle analysis and interpretation of the cells, it is possible to accurately identify the microfluidic fines of the present invention - microscopic image analysis and determination of the purpose of the system to provide a microfluidic cell can be H (4) The system's line includes 1 image capture program, image; a video camera picks up the image recognition process of the microfluidic cells to be tested and the white program, which can be used to measure microfluidic cells to very low, effective = count cell pellets 'Can count the error rate of the system to count the cells. 7 I324750 The substance of the present invention and its effects can be further illustrated by the following figures. [Embodiment] Referring to the first and second figures, the microscopic image analysis and interpretation system of the microfluidic cells of the present invention can perform microscopic image acquisition, image analysis and interpretation procedures on the microfluidic cells to be tested. , the display program and other steps are completed. The microscopic image capture program utilizes an optical microscope and a CCD camera as an image capture device. That is to say, the microfluidic cells to be tested are enlarged by microscopic images, and then photographed by a CCD camera, and the microfluidic cell microscopic images are converted into digital images by an analog-to-digital image device, and then stored in a personal computer. Microscopic image analysis and interpretation procedures, which reveal the microscopic images of the microfluids to be tested. The microscopic images of the cells can be combined with different image processing and analysis steps including cell recognition image processing analysis and automatic counting calculation methods. The interpretation of cell microscopic images and cell counting, the system of the present invention simultaneously designs a user interface to present sequence images and counts. Please refer to the second and third graphs. The microfluidic cells to be tested can be processed according to the flow chart of the microfluidic cell recognition image processing analysis program shown in the third figure. Among them: 1·Color plate conversion: The original microfluid image is RGB color image. Since the RGB image contains Luminance information, it is easy to be affected by the intensity of light, which causes the image brightness to be uneven and affects the interest. The judgment of the color 'If it is not processed by the image, it will affect the correct rate of the particle position and counting. If you want to improve the unevenness of the background brightness on the RGB image, you must analyze the three color 8 versions separately. It is quite numb and not easy to achieve. So to deal with this problem, convert RGB color images to a color space that is less sensitive to light changes. For example, convert RGB color plates to HSL (HUe,
Lumi_ce)色版，並可將亮度(Luminance， L)資訊單獨獨立出來。 2. 影像強化處理程序：針對影像亮度科的情形採取 對冗度色版做均值化的動作，所謂均值化的意思就 是，先求出每張影像L色版的亮度平均值，再將N張影 像L色版的党度平均值相加後，除以N，即為總序列影 像亮度平均值。然後再讀取這N張影像，分別將總序 列影像亮度平均值與每張影像之亮度平均值相減，求 ^每張影像的亮度差值，將此亮度差值加到影像上的 每個像素亮度值上。即如第四圖所示影像強化處理程 序之示意圖。 3. 影像負片轉換處理程序：當原始細胞影像的灰階值落 在分佈曲線的左尾，可藉由影像負片轉換的方法，改 變細胞影像灰階值使其落在分佈曲線的右尾 ，反之亦 然，即如第五圖所示負片轉換處理程序之示意圖。 4·適應法取臨界值：為了找出細胞的位置和顆數，利用 影像二值化的方法分割背景影像和細胞影像，擷取細 胞。在臨界值的選擇，本發明經過實驗測試和分析每 張影像的細胞的位置，利用細胞影像灰階值之分佈特 性如高斯分配、卜瓦松分配、Gama分配、卡方分配等 機率函數及其分配參數，根據統計推論原理，找出最 小的錯誤機率，設計一適應法取臨界值的決策程序， 來決定每張影像的臨界值，因此不同的影像具有不同 的臨界值’即如第六圖所示適應法取臨界值的示意 圖。 5.形態學處理：形態學處理之原理是藉由外形比對，再 經由空間上點與點之間的關係，定義出欲尋找的外 开）°針對=值化後的圖形，仍存在許多非乳膠微粒的 細小雜點’此時利用形態學的方法進行影像雜點濾 除’包括：侵蝕、膨脹、斷開及粒子分析，剩餘的影 鲁 像為最後我們所要觀察的細胞影像，為了將粒子影像 中之不完整的-hole"填補，並將其包覆為一個更完整 的區塊’可以採用形態學中的斷開（Open)、粒子分析 (Particle analysis)、凸形封包（Convex hull)運算來完 成。 有關本發明之影像分析判讀程序之自動細胞計數演 算程序，詳為陳述如后。 本發明係根據微流體細胞移動距離和流動速度，設計 #微流體細胞影像自動細胞計數演算法。首先在細胞流過的 路控上選取AOI (Area of Interesting)區域，二種情況下自動 演算法會對微流體序列影像進行計數，分別陳述如下： 1.進行細胞位置的分析：將AOI分成三個子區域，每一 個子區域令為一個子狀態（True or False)，當有細胞出 現時，則子狀態為致能(True)，以1來表示；反之，當 沒有細胞出現，則子狀態為非致能（False)，以〇來表 示，總共有3個子狀態，有8種排列的組合，第七圖為 AOI區域内的八種狀態示意圖。例如有某張影像其子 1324750 區域1為False，子區域2為True，子區域^仏以，則 其屬於狀態2 ’表示出現一顆細胞在此Αοι内；若子區 域1為True，子區域2為True，子區域3為False ,狀態即 為6，表示出現二顆細胞在此A〇I内，以此類推。 2.自動細胞計數判斷法則：兩種情況下必須進行計數， 第八圖即為微流體序列影像自動細胞計數演算程序 的示意圖。 a.情況一：比較影像序列中前後兩張影像的狀態後再 進行計數，規則如下：先判斷第一張影像(狀態一）是 否有被致能的子狀態，當沒有被致能的子狀態時，依 照第二張影像的狀態（狀態二）中之顆粒數進行計數， 例如說狀態一為〇〇〇，代表第一張影像沒有細胞通 過’所以第二張影像若有偵測到的細胞，為新流進的 細胞’例如：若狀態二為〇1〇、1〇〇或〇〇1，進行加丄動 作，若狀態二為011、11〇、1〇1進行加2的動作。當狀 態一有子狀態被致能時，判斷狀態二是否有領先狀態 一情形，成立的話就不進行計數，領先的意思是比較 狀態一與狀態二中的子狀態，座標位置愈大的代表等 級愈高’座標由右到左分別為高（子區域3)、中（子區 域2)、低(子區域1)三個狀態，當狀態二有比狀態一等 級還要高的狀態時，視為同一顆粒子移動，不進行計 數’若是狀態一與狀態二有相同等級的子狀態，則進 行計數，例如：狀態一為1〇〇，若狀態二為010，視為 同一顆細胞在前後張影像中移動，不進行計數，但狀 態二若為100時，則視為不同的細胞移動，進行計數。 11 1324750 b.情況二：針對若有二顆細胞同時存在一個子區域内 時，這種狀態只會計數1，則會產生誤判的情形，例 如子區域1内同時出現二顆細胞、子區域2無細胞及子 區域3有一顆細胞時，八01區域内的狀態會紀錄為 101，根據第七圖所示，狀態細胞數目只會紀錄2顆， 但實際上AOI區域内有3顆細胞存在。因此，本發明進 一步對ΛΟΙ區域内的細胞影像進行總數計數的程序， 將每一張影像中的實際細胞總數減去其狀態細胞數 # 目，若差值大於零，則進行計數，例如：某一張影像 AOI區域内的狀態為1〇1，其狀態細胞數目為2，但實 際上AOI區域内有3顆細胞，代表同時有二顆細胞出現 在一個子區域内，差值為1，所以對總數進行加i的動 作。 本發明人以本發明系統進行實驗，即本發明使用兩種 不同的微流體影像，分別為酵母菌和乳膠微粒序列影像， 進行系統的實驗測試與驗證。欲觀察的細胞經顯微鏡放大 癱後，以CCD攝影機拍攝，經影像擷取卡轉成數位影像後， 存到個人電腦（Pentium IV)，再利用本發明所設計的影像分 析判讀程序進行分析。本實驗利用National Instrument公司 所發展的Labview7.0及Vision Assistant 7.0等軟體實現影像 分析判讀程序及使用者圖形介面。實驗結果發現乳膠微粒 微流體序列影像經過影像強化處理程序後，能將原始影像 中不明顯的細胞突顯出來，比較影像強化處理前與處理後 256張每張影像上的細胞數目與位置，使用符號檢定結果顯 示屬於正號數有29組，負號數有122組，等值號有105組，p 12 1324750 值小於0.05，代表影像強化前後有統計上的顯著差異。根 據符號檢定分析方法’負號數的122組代表影像強化處理後 所觀察到的細胞數比處理前還要多，正號數29組代表經過 影像強化處理後，因為背景亮度變強，反而無法找到細胞。 經過影像強化處理程序後，實驗結果顯示，本系統的（1)靈 敏度(sensitivity)為90.07%，比人工計數原始影像之靈敏度 71.59% 還向，（2)特異性（Specificity)為 96.47% ; (3)偽陽性 (false-positive)為 〇.8% ; (4)偽陰性（false_negative)為 _ 33.33% ; (5)正確率（accuracy)為9116%，比人工計數原始影 像的正確率76.1%還高。利用統計上的卡方檢定，分析本系 統之正確率是否高於傳統人工的方式，檢定結果顯示卡方 值為41.262 (p < 0.05)具顯著差異，代表本系統自動分析每 張影像的細胞數目及位置的正確率在統計上的確高於傳統 人工的方式。 利用本發明所設計的自動細胞計數演算程序進行微 流體乳膠微粒計數，實驗結果顯示自動計數的誤差率 # 2.25%，遠低於傳統人工計數的誤差率32.58%。而酵母菌微 流體細胞計數，結果顯示本系統的誤差率為4.17%« 統觀前論，依本發明之微流體細胞之顯微影像分析與 判讀系統，不僅可克服習知方法或裝置在使用中所呈現之 缺失，本發明重要性及實用性係在於應用影像處理和分析 的技術，設計一細胞識別影像處理程序和一自動細胞計數 演算程序，對微流體影像自動計數細胞，能將系統的誤差 率降至报低，有效地進行細胞計數。 13 1324750 Λ . ’. * /-【圖式簡單說明】 第一圖：係本發明處理待測微流體細胞之流程圖。 第二圖：係本發明顯微影像分析與判讀系統示意圖。 '第三圖：係本發明微流體細胞識別影像處理分析程序示 意圖。 第四圖：係本發明影像強化處理程序之示意圖。 第五圖：係本發明負片轉換處理程序之示意圖。 第六圖：係本發明適應法取臨界值的決策程序示意圖。 φ 第七圖：係本發明AOI區域内的八種狀態示意圖。 第八圖：係本發明微流體序列影像自動細胞計數演算程 序的不意圖。 【主要元件符號說明】.Lumi_ce) color version, and can separate the brightness (Luminance, L) information. 2. Image enhancement processing program: For the case of the image brightness section, the operation of averaging the redundancy color plate is adopted. The so-called mean value means that the average value of the brightness of each image L color version is first obtained, and then N sheets are obtained. After adding the party averages of the image L color version, dividing by N is the average value of the total sequence image brightness. Then, the N images are read, and the average brightness of the total sequence image is subtracted from the average value of the brightness of each image, and the brightness difference of each image is obtained, and the brightness difference is added to each of the images. The pixel brightness value. This is a schematic diagram of the image enhancement processing procedure shown in the fourth figure. 3. Image negative film conversion processing program: When the gray scale value of the original cell image falls on the left end of the distribution curve, the image image gray scale value can be changed to fall to the right end of the distribution curve by the method of image negative film conversion, and vice versa. Also, a schematic diagram of the negative film conversion processing program as shown in the fifth figure. 4. Adaptation method takes the critical value: In order to find out the position and number of cells, the image binarization method is used to segment the background image and the cell image to capture the cells. In the selection of the critical value, the present invention experimentally tests and analyzes the position of the cells of each image, and utilizes the distribution characteristics of the gray scale values of the cell image such as Gaussian distribution, Buisson distribution, Gama distribution, chi-square distribution, and the like. Assign parameters, according to the statistical inference principle, find the minimum probability of error, design a decision-making procedure that takes the critical value to determine the critical value of each image, so different images have different critical values', ie as the sixth figure A schematic diagram of the adaptation method taken to take a critical value. 5. Morphological processing: The principle of morphological processing is to define the external opening to be searched by the relationship between the shape and the point through the space.) For the graph after the value, there are still many The fine spots of non-latex particles 'At this time, the morphological method is used for image noise filtering' including: erosion, expansion, disconnection and particle analysis, and the remaining image is the last cell image we want to observe, in order to The incomplete -hole" fill in the particle image and wrap it into a more complete block' can use morphological open, particle analysis, convex packets (Convex hull) ) The operation is done. The automatic cytometric calculation procedure for the image analysis interpretation program of the present invention is described in detail below. The invention designs an automatic cell counting algorithm for microfluidic cell images according to the moving distance and flow velocity of the microfluidic cells. First, the AOI (Area of Interesting) region is selected on the road through which the cells flow. In both cases, the automatic algorithm will count the microfluidic sequence images, which are respectively stated as follows: 1. Perform cell position analysis: divide the AOI into three Sub-area, each sub-area is a sub-state (True or False), when there is a cell, the sub-state is enabled (True), represented by 1; conversely, when no cell appears, the sub-state is non-induced (False), represented by 〇, there are 3 sub-states in total, there are 8 kinds of permutation combinations, and the seventh picture is a schematic diagram of eight states in the AOI area. For example, if there is an image whose sub 1324750 area 1 is False, sub area 2 is True, and the sub area is ^, then it belongs to state 2 ' indicates that a cell appears in this Αοι; if sub-area 1 is True, sub-area 2 True, sub-region 3 is False, and the state is 6, indicating that two cells are present in this A〇I, and so on. 2. Automatic cell counting rule: In both cases, counting must be performed. The eighth figure is a schematic diagram of the automatic cell counting calculation program for microfluidic sequence images. a. Case 1: Compare the status of the two images in the image sequence and then count them. The rules are as follows: first determine whether the first image (state 1) has an enabled substate, and when it is not enabled. When counting, according to the number of particles in the state of the second image (state 2), for example, the state one is 〇〇〇, which means that the first image has no cells passing through, so if the second image has detected cells, For cells that are newly flown, for example, if the state 2 is 〇1〇, 1〇〇 or 〇〇1, the twisting operation is performed, and if the state 2 is 011, 11〇, 1〇1, the action of adding 2 is performed. When the state has a substate enabled, it is judged whether the state 2 has a leading state. If it is established, it will not count. The leading meaning is to compare the substates in state one and state two, and the representative level of the coordinate position is larger. The higher the 'coordinates' are from high to right (sub-region 3), medium (sub-region 2), and low (sub-region 1) from right to left, and when state 2 has a state higher than the state level, For the same particle to move, do not count 'If the state one has the same level as the state 2 sub-state, then count, for example: state one is 1〇〇, if state two is 010, it is considered that the same cell is before and after When the image moves, it does not count, but if the state 2 is 100, it is regarded as different cell movement and counting. 11 1324750 b. Case 2: If there are two cells in a sub-area at the same time, this state will only count 1, which will lead to misjudgment. For example, two cells and sub-regions appear in sub-region 1 at the same time. When there is one cell in cell-free and sub-region 3, the state in the area of 08 and 01 will be recorded as 101. According to the seventh figure, the number of state cells will only be recorded as 2, but in fact, there are 3 cells in the AOI region. Therefore, the present invention further calculates a total number of cell images in the temporal region, and subtracts the total number of cells in each image from the number of state cells. If the difference is greater than zero, the counting is performed, for example: The state of an image AOI is 1〇1, and its state cell number is 2, but in fact there are 3 cells in the AOI region, indicating that two cells appear in one sub-region at the same time, the difference is 1, so Add i to the total number of actions. The inventors conducted experiments using the system of the present invention, i.e., the present invention uses two different microfluidic images, respectively, for yeast and latex particle sequence images, for systematic experimental testing and verification. The cells to be observed are magnified by a microscope, photographed by a CCD camera, converted into digital images by an image capture card, stored in a personal computer (Pentium IV), and analyzed by the image analysis interpretation program designed by the present invention. This experiment uses software such as Labview7.0 and Vision Assistant 7.0 developed by National Instrument to realize image analysis and interpretation program and user graphical interface. The experimental results show that after the image enhancement process of the latex microfluidic sequence, the cells in the original image can be highlighted, and the number and position of the cells on each of the 256 images before and after the image enhancement process are compared. The results of the verification showed that there were 29 groups with positive numbers, 122 groups with negative numbers, 105 groups with equal values, and p 12 1324750 with values less than 0.05, indicating statistically significant differences before and after image enhancement. According to the symbolic analysis method, the number of cells in the negative group of 122 indicates that the number of cells observed after image enhancement treatment is more than that before the treatment. The 29 numbers of positive numbers represent the image enhancement process, because the background brightness becomes stronger, but it cannot Find the cells. After the image enhancement processing program, the experimental results show that the sensitivity of the system is (1) sensitivity is 90.07%, which is 71.59% higher than the sensitivity of manual counting original image, and (2) specificity is 96.47%; 3) false-positive is 〇.8%; (4) false negative (false_negative) is _ 33.33%; (5) accuracy is 9116%, which is 76.1% correct than manual counting of original images. Still high. Using the statistical chi-square test to analyze whether the correct rate of the system is higher than the traditional manual method, the verification result shows that the chi-square value is 41.262 (p < 0.05) with significant difference, which means that the system automatically analyzes the cells of each image. The correct rate of number and location is statistically higher than the traditional manual method. Using the automatic cell counting algorithm designed by the present invention to perform microfluidic latex particle counting, the experimental results show that the error rate of automatic counting is #2.25%, which is much lower than the traditional manual counting error rate of 32.58%. The yeast microfluidic cell count showed that the error rate of the system was 4.17%. << Foresight, the microscopic image analysis and interpretation system of the microfluidic cells according to the present invention can not only overcome the conventional methods or devices. The significance and utility of the present invention lies in the application of image processing and analysis techniques, designing a cell recognition image processing program and an automatic cell counting algorithm, automatically counting cells for microfluidic images, and enabling the system to The error rate is reduced to a low, and the cell count is effectively performed. 13 1324750 Λ . ' / / [ [Simple Description] The first figure: is a flow chart of the present invention for processing microfluidic cells to be tested. The second figure is a schematic diagram of the microscopic image analysis and interpretation system of the present invention. 'Third Figure: is a schematic diagram of the microfluidic cell recognition image processing analysis program of the present invention. Figure 4 is a schematic diagram of the image enhancement processing program of the present invention. Fifth drawing is a schematic diagram of a negative film conversion processing program of the present invention. Figure 6 is a schematic diagram of the decision-making procedure for taking the critical value of the adaptation method of the present invention. φ Figure 7: Schematic diagram of eight states in the AOI region of the present invention. Figure 8 is a schematic representation of the automated cell counting algorithm for microfluidic sequence images of the present invention. [Main component symbol description].