201025032 - 六、發明說明： . 【發明所屬之技術領域】 本發明是有關於一種分析系統及方法，特別是指一種 子宮頸細胞分析系統及方法。 【先前技術】 - 子宮頸癌曾經是普遍又嚴重的婦女腫瘤。幸運地，由 於子宮頸抹片篩檢計畫的執行，子宮頸癌的發生率與死亡 • 率已被有效地降低。子宮頸抹片篩檢是簡單又便宜的篩檢 方法，可以有效地檢驗出早期癌前病變。然而，人工閱片 及判斷只能作主觀的定性分析，且容易因人為因素而導致 誤判《因此，世界各國都致力於研發儀器來進行客觀的定 量分析，但其中有許多細節需被考慮，以提高正確率。 【發明内容】 因此，本發明之目的即在提供一種正確率高的子宮頸 細胞分析系統。 ❹ 於疋，本發明子呂頸細胞分析系統包含一影像分析裝 置。該影像分析裝置包括一影像處理單元、一參數量測單 * 元及一分類單元。 該影像處理單元用於根捸一待分析細胞影像，產生一 標示其中一待分析細胞之細胞質及細胞核的遮罩影像。 該參數量測單元用於根據該待分析細胞影像、該遮罩 影像及一微測尺影像，計算該待分析細胞的一組參數。該 組參數包括細胞核週長、細胞核面積、細胞核最大長度、 細胞核最大寬度、細胞核細胞質面積比、細胞核軸心到邊 201025032 界最大長度、細胞核轴心到邊界平均長度、細胞核重心到 邊界最大長度、細胞核重心到邊界平均長度、細胞核共生 矩陣熵、細胞核共生矩陣對比度及細胞核粗链度。 該分類單元用於根據該組參數分類該待分析細胞，以 產生一分析結果。該分析結果指示該待分析細胞是一淺層 鱗狀上皮細胞、一中層鱗狀上皮細胞、一側基底鱗狀上皮 細胞及一鱗狀表皮内病灶細胞中的哪一種。 而本發明之另一目的即在提供一種正確率高的子宮頸 細胞分析方法。 於是’本發明子宮頸細胞分析方法包含以下步驟： 根據一待分析細胞影像，產生一標示其中一待分析細 胞之細胞質及細胞核的遮罩影像； 根據該待分析細胞影像、該遮罩影像及一微測尺影像 ’計算該待分析細胞的一組參數，該組參數包括細胞核週 長、細胞核面積、細胞核最大長度、細胞核最大寬度、細 胞核細胞質面積比、細胞核轴心到邊界最大長度、細胞核 轴心到邊界平均長度、細胞核重心到邊界最大長度、細胞 核重心到邊界平均長度、細胞核共生矩陣熵、細胞核共生 矩陣對比度及細胞核粗链度；及 根據該組參數分類該待分析細胞，以產生一分析結果 ’該分析結果指示該待分析細胞是一淺層鱗狀上皮細胞、 一中層鱗狀上皮細胞、一侧基底鱗狀上皮細胞及一鱗狀表 皮内病灶細胞中的哪一種。 201025032 - 【實施方式】 . 有關本發明之前述及其他技術内容、特點與功效，在 、下配5參考圖式之二個較佳實施例的詳細說明中，將可 清楚地呈現。 參閱圖1 ’本發明子宮頸細胞分析系統之較佳實施例包 含一影像擷取裝置丨及一影像分析裝置2。 - 影像擷取裝置1包括一顯微鏡11及一數位相機12。顯 Φ 微鏡U用於放大一子宮頸抹片。數位相機12用於擷取子宮 頸抹片經顯微鏡11放大後的影像，以產生一待分析細胞影 像（如圊2(a)所示）。在本實施例中，顯微鏡u的放大倍率 疋00倍’子呂頸抹片疋採用液基抹片（HqUid_based smear )° 影像分析裝置2包括一影像處理單元21、一參數量測 單元22及一分類單元23。 影像處理單元21用於對待分析細胞影像進行一影像增 Φ 強處理，以使細胞與背景能被清楚分辨，並根據處理好的 細胞影像，產生一以不同顏色來標示其中一待分析細胞之 細胞質與細胞核的遮罩影像（如圖2(b)所示）。在本實施例 中’影像增強處理包括直方圖正規化（hist〇gram normalization)及直方圖均等化（hist〇gram equalizati〇n); 遮罩影像以藍色來標示待分析細胞的細胞質，以綠色來標 示待分析細胞的細胞核。 參數量測單元22用於根據待分析細胞影像、遮罩影像 及一微測尺影像（如圖2(c)所示），計算待分析細胞的一組 201025032 參數。該組參數包括細胞核週長、細胞核面積、細胞核最 大長度、細胞核最大寬度、細胞核細胞質面積比、細胞核 轴心到邊界最大長度、細胞核軸心到邊界平均長度、細胞 核重心到邊界最大長度、細胞核重心到邊界平均長度、細 胞核共生矩陣嫡（entrophy of co-occurrence matrix)' 細胞 核共生矩陣對比度（contrast of co-occurrence matrix )、細 胞核粗縫度（coarseness)及細胞核對比度，其中，細胞核 週長、細胞核面積、細胞核最大長度及細胞核最大寬度描 述細胞核尺寸，細胞核軸心到邊界最大長度、細胞核軸心 到邊界平均長度、細胞核重心到邊界最大長度及細胞核重 心到邊界平均長度描述細胞核形狀，細胞核共生矩陣燏、 細胞核共生矩陣對比度、細胞核粗糙度及細胞核對比度描 述細胞核紋理（texture )。 微測尺影像是預先藉由影像棟取裝置1對一微測尺玻 片進行放大並擷取影像，且二值化擷取到的影像而獲得的 ’用於辅助衡量待分析細胞的實際尺寸。 參閱圖3與圖4，在本實施例中，參數量測單元12以 迴旋種子區域增長法（circular seeded regi〇n gr〇wing method)對遮罩影像進行邊緣偵測，以獲得待分析細胞的 細胞質及細胞核的輪廓，用於輔助計算該組參數。參數量 測單元12所使用的迴旋種子區域增長法包括以下步驟： 步驟51是在遮罩影像3卜由左到右且由上到下尋找 符合細胞質或細胞核.特徵的第一個點作為一起始種子點Η 201025032 =驟52疋將起始種子點31設定為一基準種子點u。 ^驟53疋在圍繞基準種子點32的八個點中（從左上 開始，依川員時針方向，分別位於索引 位置I〜VIII )’以順 &針方向環繞並挑選符合細胞質或細胞核特徵的—個點作 為一成長候選點33。201025032 - VI. Description of the Invention: [Technical Field of the Invention] The present invention relates to an analysis system and method, and more particularly to a cervical cell analysis system and method. [Prior Art] - Cervical cancer was once a common and serious tumor in women. Fortunately, the incidence of cervical cancer and mortality rates have been effectively reduced due to the implementation of the Pap smear screening program. Pap smear screening is a simple and inexpensive screening method that can effectively detect early precancerous lesions. However, manual reading and judgment can only be subjective qualitative analysis, and it is easy to cause misjudgment due to human factors. Therefore, all countries in the world are committed to developing instruments for objective quantitative analysis, but many of them need to be considered to Improve the accuracy rate. SUMMARY OF THE INVENTION Accordingly, it is an object of the present invention to provide a cervical cell analysis system having a high accuracy. In the present invention, the Rie Neck Cell Analysis System of the present invention comprises an image analysis device. The image analysis device comprises an image processing unit, a parameter measurement unit, and a classification unit. The image processing unit is configured to image a cell image to be analyzed to generate a mask image indicating a cytoplasm and a nucleus of one of the cells to be analyzed. The parameter measuring unit is configured to calculate a set of parameters of the cell to be analyzed according to the image of the cell to be analyzed, the mask image and a micro-scale image. The parameters of the group include the circumference of the nucleus, the area of the nucleus, the maximum length of the nucleus, the maximum width of the nucleus, the ratio of the cytoplasmic area of the nucleus, the maximum length of the core axis to the edge of the 201025032, the average length of the core axis to the boundary, the maximum center of gravity of the nucleus to the boundary, and the nucleus. The center of gravity to the average length of the boundary, the nuclear co-occurrence matrix entropy, the nuclear symbiotic matrix contrast and the nuclear thick chain. The classification unit is configured to classify the cells to be analyzed according to the set of parameters to generate an analysis result. The result of the analysis indicates which of the superficial squamous epithelial cells, a middle squamous epithelial cell, one basal squamous epithelial cell, and a squamous intraepithelial lesion cell. Yet another object of the present invention is to provide a method for analyzing cervical cells with high accuracy. Thus, the method for analyzing cervical cells of the present invention comprises the steps of: generating a mask image indicating the cytoplasm and nucleus of one of the cells to be analyzed according to a cell image to be analyzed; according to the image of the cell to be analyzed, the mask image and a The micro-scale image 'calculates a set of parameters of the cell to be analyzed, including the circumference of the nucleus, the area of the nucleus, the maximum length of the nucleus, the maximum width of the nucleus, the ratio of the cytoplasmic area of the nucleus, the maximum length of the nucleus to the boundary, and the axis of the nucleus. The average length to the boundary, the center of gravity of the cell to the maximum length of the boundary, the center of gravity to the average length of the boundary, the nuclear co-occurrence matrix entropy, the nuclear symbiotic matrix contrast and the nuclear thick chain degree; and classify the cells to be analyzed according to the set of parameters to generate an analysis result The result of the analysis indicates which of the superficial squamous epithelial cells, a middle squamous epithelial cell, one basal squamous epithelial cell, and a squamous intraepithelial lesion cell. 201025032 - [Embodiment] The foregoing and other technical contents, features and advantages of the present invention will be apparent from the detailed description of the preferred embodiments of the accompanying drawings. Referring to Figure 1, a preferred embodiment of the cervical cell analysis system of the present invention includes an image capture device and an image analysis device 2. - The image capturing device 1 includes a microscope 11 and a digital camera 12. Φ Microscope U is used to magnify a Pap smear. The digital camera 12 is used to capture an image of the cervix smear magnified by the microscope 11 to produce an image of the cell to be analyzed (as shown by 圊 2(a)). In the present embodiment, the magnification of the microscope u is 疋00 times, and the liquid smears are used. The image analysis device 2 includes an image processing unit 21, a parameter measuring unit 22, and a Classification unit 23. The image processing unit 21 is configured to perform an image enhancement processing on the cell image to be analyzed, so that the cells and the background can be clearly distinguished, and according to the processed cell image, a different color is used to indicate the cytoplasm of one of the cells to be analyzed. A mask image with the nucleus (as shown in Figure 2(b)). In the present embodiment, the image enhancement processing includes histogram normalization and histogram equalization (hist〇gram equalizati〇n); the mask image indicates the cytoplasm of the cells to be analyzed in blue, in green. To identify the nuclei of the cells to be analyzed. The parameter measuring unit 22 is configured to calculate a set of 201025032 parameters of the cells to be analyzed according to the cell image to be analyzed, the mask image, and a micro-scale image (as shown in FIG. 2(c)). The parameters of the group include the circumference of the nucleus, the area of the nucleus, the maximum length of the nucleus, the maximum width of the nucleus, the ratio of the cytoplasmic area of the nucleus, the maximum length of the nucleus to the boundary, the average length of the nucleus from the center of the nucleus to the boundary, the maximum length of the center of the nucleus to the boundary, and the center of gravity of the nucleus. The average length of the boundary, the entrophy of co-occurrence matrix', the contrast of co-occurrence matrix, the coarseness of the nucleus, and the contrast of the nucleus, wherein the nucleus circumference, the nuclear area, The maximum length of the nucleus and the maximum width of the nucleus describe the size of the nucleus, the maximum length from the axis of the nucleus to the boundary, the average length from the core axis to the boundary, the maximum length from the center of gravity of the nucleus to the boundary, and the average length of the center of the nucleus. The shape of the nucleus, the nucleus of the nucleus, the nucleus Symbiotic matrix contrast, nuclear roughness, and nuclear contrast describe the nuclear texture. The micro-scale image is obtained by amplifying and capturing an image of a micro-scale slide by the image-taking device 1 and binarizing the captured image to assist in measuring the actual size of the cells to be analyzed. . Referring to FIG. 3 and FIG. 4, in the embodiment, the parameter measuring unit 12 performs edge detection on the mask image by using a circular seeded regi〇n gr〇wing method to obtain cells to be analyzed. The outline of the cytoplasm and nucleus is used to aid in the calculation of this set of parameters. The cyclotron seed region growth method used by the parameter measuring unit 12 includes the following steps: Step 51: as a starting point in the mask image 3 from left to right and top to bottom looking for the first point corresponding to the cytoplasm or nucleus. The seed point Η 201025032 = step 52 疋 sets the starting seed point 31 as a reference seed point u. ^ 疋 疋 疋 八个 疋 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 围绕 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在 在A point is used as a growth candidate point 33.
步驟54是判斷成長候選 如果否（如圖3(a)〜⑷所示） 圖3(d)所示），則結束。 步驟55是將成長候選點 跳到步驟53。 點33是否是起始種子點31， ，則跳到步驟55，如果是（如 33設定為基準種子點32，並 因此，所有種子點即標示出細胞質或細胞核的輪廓。 ® 4〇)所不’在圍繞起始種子點的八個點中，是 二Γ位置1的點開始’以順時針方向環繞並挑選符合Step 54 is to judge the growth candidate. If no (as shown in Fig. 3 (a) to (4)), as shown in Fig. 3 (d), the process ends. Step 55 is to jump to the growth candidate point to step 53. If point 33 is the starting seed point 31, then jump to step 55, if yes (as 33 is set as the reference seed point 32, and therefore, all seed points indicate the contour of the cytoplasm or nucleus. ® 4〇) 'In the eight points around the starting seed point, the point where the two positions are at the beginning of the point' is surrounded by a clockwise direction and selected to match
是在索的第、—個點作為成長候選點33,因此 '、 处找到成長候選點33。如圖4(b)〜(d)所- 子祕基㈣子點32 H點中，是從前任基準種 :^下-點開始，以順時針方向環繞並挑選符合 或細胞㈣徵的第-個點作為成長候選點33,如此可2 免挑選核基準種子㈣作^長候選點& 例’在圍繞基準種子點32的峨 （）為 =置―此是™位置I:::: 二並她丨位置以處找到成錢選點33。㈣=找 圍繞基準種子點32的八個財，前任基準種子點： 索引位置1V’因此是一丨位置V的點開二:位： 201025032 在索引位置I處找到成長候選點33。以圖4⑷為例，在圍 繞基準種子點32的八個點中，前任基準種子點是位於索引 位置VIII，因此是從位於索引位置j的點開始尋找，並在索 引位置III處找到成長候選點33，且由於成長候選點Μ是 起始種子點31，因此成長過程停止。 由於本領域中具有通常知識者熟知該組參數的計算方 式，故此處不再多加說明。 參閱圖1，分類單元23用於根據該組參數分類該待分 析細胞，以產生一分析結果。分析結果指示待分析細胞是 一淺層鱗狀上皮細胞（superficial squani〇us celi )、一中層 鱗狀上皮細胞（intermediate squam〇us celi)、一側基底鱗狀 上皮細胞（parabasal squamous cell)及一鱗狀表皮内病灶 田胞（squamous intraepithelial lesion cell )中的哪一種，盆 中，淺層鱗狀上皮細胞、中層鱗狀上皮細胞及側基底鱗狀 上皮細胞是正常細胞，鱗狀表皮内病灶細胞是異常細胞。 在本實施例中，分類單元23包括一正規化模組231及 支援向量機232’其中，正規化模組231用於正規化該組 參數’以產生一組正規化參數；支援向量機232預先以複 數細胞樣本的正規化參數及專家判定結果進行訓練，以使 其根據正規化模組23 1所產生的該組正規化參數分類待分 析細胞’並產生分析結果。 由於各個參數的值域可能差距很大，藉由使支援向量 機232以正規化參數（例如：值域皆為0〜1 )進行訓練，可 以避免其誤將較大權重放在值域較突出的參數上。在訓練 201025032 - _’首先’將搜集到的細胞樣本分成—訓練組及-測試组 . ;接著’使支援向量機攻以訓練組的正規化參數及專家 判定結果進行訓練’並根據測試組的正規化參數進行分類 ’以產生分析結果；然後，比較測試組的分析結果及專家 判定結果’如果正確率不如預期，則調整支援向量機232 • 的成本參數，並使其重覆上述訓練及分類過程，否則，結 束訓練。 φ 當使支援向量機232對具有表1所示特性的503個細 胞樣本（包括139個淺層鱗狀上皮細胞、178個中層鱗狀上 皮細胞、128個側基底鱗狀上皮細胞及58個鱗狀表皮内病 灶細胞）進行分類時，其正確率如表2所示。由表2可知 ，分析結果的正確率相當高。 表1 參數 淺層鱗狀 上皮細胞 中層鱗狀 上皮細胞 側基底鱗 狀上皮細 胞 鱗狀表皮内 病灶細胞 T測試的 Ρ值 細胞核週 15.8±2.7 26.7±3.1 24.6±3.8 40.2±8.1 <0.001 長 __( μηι) (μιη) (μπι) (μ ιη ) 細胞核面 29.57±1〇.9 82·8±2〇.〇 71.8±21.4 192.6±80.5 <0.001 積 (μηι2) (μιη2 ) (μηι2) (μιη2 ) 細胞核最 6.2±1.3 10.3 士 1.1 9.5±1.5 15.6士3_1 <0.001 大長度 (μηι) (μιη ) (μηι) (μπι) 細胞核最 5.0±0.8 8.4±1.2 7.8±1.3 12·6士 2.7 <0.001 大寬度 (μηι) (μηι ) (μιη) (μιη ) 9 201025032 細胞核細 胞質面積 比 1.1±0.0 (%) 4.0±1.5 (%) 16·6±8·1 (%) 48·4±75·1 (%) <0.001 細胞核轴 心到邊界 最大長度 42.7±8.2 (μπι) 70.8±9.4 (μιη) 66_0±11·6 (μιη) 107.8±24_0 (μιη) <0.001 細胞核軸 心到邊界 平均長度 33.7±5.7 (μιη ) 57·0±6·6 (μιη ) 52.8±8.1 (μιη ) 86.0±17.4 (μπι ) <0.001 細胞核重 心到邊界 最大長度 39.2±8.0 (μιη ) 65.1±7.3 (μιη ) 60·1±9.5 (μιη) 99·1±19.8 (μιη ) <0.001 細胞核重 心到邊界 平均長度 (μιη ) 33.4±5.7 (μιη ) 56.5 土 6.5 (μιη ) 52.4±8.1 (μιη ) 85.4±17.2 (μπι) <0.001 細胞核共 生矩陣熵 -5.3±3.1 (1〇3) -18±7·5 (1〇3) -19±7·5 (1〇3) -65±35 (ΐ〇3) <0.001 細胞核共 生矩陣對 比度 2.7±3.6 (1〇5) 9.3±5.5 (1〇5) 4.5±2.4 (ίο5) 12±7.2 (1〇5) <0.001 細胞核粗 糙度 240·4±41.2 157.2±35.8 87.1±18.7 107.5±29·8 <0.001The first point of the cable is the growth candidate point 33, so the growth candidate point 33 is found at '. As shown in Fig. 4(b) to (d) - sub-base (4) sub-point 32 H point, starting from the previous reference species: ^ down-point, rounding in a clockwise direction and selecting the first or the cell (four) sign - The point is used as the growth candidate point 33, so that the selection of the nuclear reference seed (4) is made as the candidate point & the example 'in the 种子() around the reference seed point 32 is set to = this is the TM position I:::: And she found a place to find a money selection point 33. (4) = Find the eight financial assets around the reference seed point 32. The predecessor reference seed point: The index position 1V' is therefore the point of the position V. The second position: 201025032 The growth candidate point 33 is found at the index position I. Taking FIG. 4(4) as an example, among the eight points surrounding the reference seed point 32, the former reference seed point is located at the index position VIII, so the search is started from the point located at the index position j, and the growth candidate point is found at the index position III. 33, and since the growth candidate point Μ is the starting seed point 31, the growth process stops. Since those skilled in the art are familiar with the calculation of the set of parameters, they will not be explained here. Referring to Figure 1, classification unit 23 is operative to classify the cells to be analyzed based on the set of parameters to produce an analysis result. The results of the analysis indicated that the cells to be analyzed were a superficial squamous epithelial cell (superficial squani〇us celi), a middle squamous epithelial cell (intermediate squam〇us celi), one basal squamous cell (parabasal squamous cell) and one Which of the squamous intraepithelial lesion cells, the superficial squamous epithelial cells, the middle squamous epithelial cells, and the lateral basal squamous epithelial cells are normal cells, squamous intraepithelial lesion cells It is an abnormal cell. In this embodiment, the classification unit 23 includes a normalization module 231 and a support vector machine 232', wherein the normalization module 231 is used to normalize the set of parameters 'to generate a set of normalization parameters; the support vector machine 232 advances The normalized parameters of the plurality of cell samples and the results of the expert determination are trained to classify the cells to be analyzed according to the set of normalized parameters generated by the normalization module 23 1 and generate an analysis result. Since the range of values of each parameter may be very different, by supporting the support vector machine 232 with normalized parameters (for example, the value ranges are 0 to 1), it is possible to avoid the mistake of replaying the larger weight in the range. On the parameters. In the training 201025032 - _ 'first 'divided the collected cell samples into - training group and - test group. Then 'make the support vector machine attack the training group's normalization parameters and expert judgment results for training' and according to the test group Normalize the parameters to classify 'to generate the analysis results; then, compare the test results of the test group and the expert judgment results' If the correct rate is not as expected, adjust the cost parameters of the support vector machine 232 • and repeat the above training and classification Process, otherwise, end the training. φ When the support vector machine 232 was paired with 503 cell samples having the characteristics shown in Table 1 (including 139 superficial squamous epithelial cells, 178 intermediate squamous epithelial cells, 128 lateral basal squamous epithelial cells, and 58 scales). The classification accuracy of the lesions in the epidermal cells is shown in Table 2. As can be seen from Table 2, the accuracy of the analysis results is quite high. Table 1 Parameters of superficial squamous epithelial cells in squamous epithelial cells, basal squamous cell epithelial cells, squamous epithelial cells, T-test for sputum cells, nuclear week 15.8 ± 2.7 26.7 ± 3.1 24.6 ± 3.8 40.2 ± 8.1 < 0.001 long _ _( μηι) (μιη) (μπι) (μ ιη ) Cellular surface 29.57±1〇.9 82·8±2〇.〇71.8±21.4 192.6±80.5 <0.001 product (μηι2) (μιη2 ) (μηι2) ( Ιιη2 ) Cell nucleus 6.2 ± 1.3 10.3 ± 1.1 9.5 ± 1.5 15.6 ± 3_1 < 0.001 Large length (μηι) (μιη ) (μηι) (μπι) The most nuclear nuclei 5.0±0.8 8.4±1.2 7.8±1.3 12·6 2.7 lt ;0.001 Width (μηι) (μηι ) (μιη) (μιη ) 9 201025032 Cellular cytoplasmic area ratio 1.1 ± 0.0 (%) 4.0 ± 1.5 (%) 16 · 6 ± 8. 1 (%) 48 · 4 ± 75 · 1 (%) <0.001 The maximum length of the core from the core to the boundary is 42.7 ± 8.2 (μπι) 70.8 ± 9.4 (μιη) 66_0 ± 11 · 6 (μιη) 107.8 ± 24_0 (μιη) < 0.001 cell core axis to the average length of the boundary 33.7±5.7 (μιη) 57·0±6·6 (μιη) 52.8±8.1 (μιη) 86.0±17.4 (μπι) <0.001 cell center of gravity to The maximum length of the boundary is 39.2±8.0 (μιη) 65.1±7.3 (μιη) 60·1±9.5 (μιη) 99·1±19.8 (μιη) <0.001 The center of gravity of the cell to the average length of the boundary (μιη) 33.4±5.7 (μιη) 56.5 Soil 6.5 (μιη) 52.4±8.1 (μιη) 85.4±17.2 (μπι) <0.001 Nuclear symbiotic matrix entropy -5.3±3.1 (1〇3) -18±7·5 (1〇3) -19±7·5 (1〇3) -65±35 (ΐ〇3) <0.001 Cell nuclear symbiosis matrix contrast 2.7±3.6 (1〇5) 9.3±5.5 (1〇5) 4.5±2.4 (ίο5) 12±7.2 (1〇5 <0.001 cell nuclear roughness 240·4±41.2 157.2±35.8 87.1±18.7 107.5±29·8 <0.001
10 20102503210 201025032
23.2±22.〇 185.0±541.3 2.2±2.7 11.2±7.0 0.01923.2±22.〇 185.0±541.3 2.2±2.7 11.2±7.0 0.019
細胞核對 比度 表2 細胞類型 數目 ~~" ---〜^ 一分類正 134 ~~------ 正確率 淺層鱗狀上皮細 胞 139 96.40% 中層鱗狀上皮細 胞 178 163 91.57% 側基底鱗狀上皮 細胞 128 120 93.75% 鱗狀表皮内病灶 細胞 58 58 100% 全部 503 475 94.43% 值得注意的是，以T測試評估各個參數對分析結果的 « 影響，所得的P值如表1所示。由表〗可知，細胞核對比 度對分析結果的影響較小。因此，該組參數可以不包括細 胞核對比度，這樣仍能使本實施例具有相當高的正確率。 在本實施例中，影像分析裝置2是以一電腦來實現， 影像處理單元21、參數量測單元22及分類單元23是以軟 體方式實施，並於電腦中執行。 參閱圖5’本發明子宮頸細胞分析方法之較佳實施例包 含以下步驟： 步驟41是放大_ 細胞經放大後的影像， 子宮頸抹片細胞，並摘取子宮頸抹片 以產生一待分析細胞影像。 201025032 步驟42是對待分析細胞影像進行影像增強處理，以使 . 細胞與背景能被清楚分辨，並根據處理好的細胞影像，產 生一以不同顏色來標示其中一待分析細胞之細胞質與細胞 核的遮罩影像。 步驟43是根據待分析細胞影像、遮罩影像及一微測尺 影像’計鼻待分析細胞的一組參數，該組參數包括細胞核 週長、細胞核面積、細胞核最大長度、細胞核最大寬度、 細胞核細胞質面積比、細胞核轴心到邊界最大長度、細胞 〇. 核軸心到邊界平均長度、細胞核重心到邊界最大長度、細 胞核重心到邊界平均長度、細胞核共生矩陣熵、細胞核共 生矩陣對比度、細胞核粗糙度及細胞核對比度。 步驟44是根據該組參數分類該待分析細胞，以產生一 分析結果，分析結果指示待分析細胞是一淺層鱗狀上皮細 胞、一中層鱗狀上皮細胞、一側基底鱗狀上皮細胞及一鱗 狀表皮内病灶細胞中的哪一種。 步驟41〜44的詳細作法可以參考上述子宮頸細胞分析❹ 系統之較佳實施例，此處不再多加說明。 综上所述，藉由根據該組參數分類待分析細胞，上述 實施例可以獲得相當高的正確率，故確實能達成本發明之 目的。 惟以上所述者，僅為本發明之較佳實施例而已，當不 能以此限定本發明實施之範圍，即大凡依本發明申請專利 範圍及發明說明内容所作之簡單的等效變化與修飾，皆仍 屬本發明專利涵蓋之範圍内。 12 201025032 【圊式簡單說明】 圖1是一方塊圖’說明本發明子宮頸細胞分析系統之 較佳實施例； 圖2是一示意圖，說明圖丨較佳實施例所使用的一待 刀析細胞影像、—遮罩影像及—微測尺影像； 圖 疋机程圖，說明圖1較佳實施例所使用的一迴 旋種子區域增長法_ .Cell nuclear contrast Table 2 Number of cell types ~~" ---~^ One classification is positive 134 ~~------ Correct rate of shallow squamous epithelial cells 139 96.40% Middle squamous epithelial cells 178 163 91.57% Lateral basal Squamous epithelial cells 128 120 93.75% Squamous intraepithelial lesion cells 58 58 100% All 503 475 94.43% It is worth noting that the T test evaluates the effect of each parameter on the analysis results, and the resulting P values are shown in Table 1. . It can be seen from the table that the nuclear contrast has little effect on the analysis results. Therefore, the set of parameters may not include cell nuclear contrast, which still enables the present embodiment to have a relatively high accuracy. In the present embodiment, the image analyzing device 2 is implemented by a computer. The image processing unit 21, the parameter measuring unit 22, and the sorting unit 23 are implemented in a software manner and executed in a computer. Referring to Figure 5, a preferred embodiment of the method for analyzing cervical cells of the present invention comprises the following steps: Step 41 is to magnify the enlarged image of the cells, the Pap smear cells, and extract the Pap smear to generate an analysis. Cell image. 201025032 Step 42 is to perform image enhancement processing on the analyzed cell image so that the cells and the background can be clearly distinguished, and according to the processed cell image, a different color is used to indicate the cytoplasm and nucleus of one of the cells to be analyzed. Cover image. Step 43 is a set of parameters of the cells to be analyzed according to the cell image to be analyzed, the mask image and the micro-scale image. The parameters include the circumference of the cell, the area of the nucleus, the maximum length of the nucleus, the maximum width of the nucleus, and the cytoplasm of the nucleus. Area ratio, the maximum length of the core axis to the boundary, cell 〇. The average length from the core axis to the boundary, the maximum length of the center of the cell to the boundary, the average length of the center of the cell to the boundary, the entropy of the nuclear co-occurrence matrix, the contrast of the nuclear co-occurrence matrix, and the nuclear roughness Nuclear contrast. Step 44 is to classify the cells to be analyzed according to the set of parameters to generate an analysis result indicating that the cells to be analyzed are a shallow squamous epithelial cell, a middle squamous epithelial cell, a basal squamous epithelial cell, and a Which of the lesion cells in the squamous epidermis. For a detailed description of the steps 41 to 44, reference may be made to the preferred embodiment of the cervical cell analysis system described above, and will not be further described herein. In summary, the above embodiments can achieve a relatively high accuracy rate by classifying the cells to be analyzed according to the set of parameters, so that the object of the present invention can be achieved. The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent. 12 201025032 [Brief Description] FIG. 1 is a block diagram showing a preferred embodiment of the cervical cell analysis system of the present invention; FIG. 2 is a schematic view showing a cell to be analyzed used in the preferred embodiment of the invention Image, mask image and micro-scale image; diagram of the turret seed region growth method used in the preferred embodiment of Figure 1.
圖 4 是* 疋不意圖 種子區域增長法；及 圖5是一流_ 較佳實施例。 ’說明圖1較佳實施例所使用的迴旋 ’說明本發明子宮頸細胞分析方法之Figure 4 is a * not intended seed area growth method; and Figure 5 is a first class - preferred embodiment. Illustrating the convolution used in the preferred embodiment of Fig. 1 illustrates the method of analyzing cervical cells of the present invention
13 201025032 【主要元件符號說明】 1....... …影像擷取裝置 232…… •-支援向量機 11...... …顯微鏡 3 ........ -遮罩影像 1 2 -，κ ·' ♦ …數位相機 "起始種子點 2… …影像分析裝置 32....... -·基準種子點 1 …影像處理單元 3 3 * * * * * * # • *成長候選點 22…… …參數量測單元 41〜44.- ·-步驟 23...... …分類單元 51 〜55，· ••步驟 23 1 κ … …正規化模組 1413 201025032 [Explanation of main component symbols] 1.......Image capture device 232... •-Support vector machine 11...Microscope 3 ........ -Mask Image 1 2 -, κ · ' ♦ ... digital camera " starting seed point 2... image analysis device 32....... - reference seed point 1 ... image processing unit 3 3 * * * * * * #• * Growth candidate point 22...... Parameter measurement unit 41~44.---Step 23... Classification unit 51~55, ·••Step 23 1 κ Normalization module 14