TW201000074A - Pathologic detecting system by using the detection of perspective image of breast and the method thereof - Google Patents

Pathologic detecting system by using the detection of perspective image of breast and the method thereof Download PDF

Info

Publication number
TW201000074A
TW201000074A TW97122633A TW97122633A TW201000074A TW 201000074 A TW201000074 A TW 201000074A TW 97122633 A TW97122633 A TW 97122633A TW 97122633 A TW97122633 A TW 97122633A TW 201000074 A TW201000074 A TW 201000074A
Authority
TW
Taiwan
Prior art keywords
image
breast
pathological
module
block
Prior art date
Application number
TW97122633A
Other languages
Chinese (zh)
Inventor
Qi-Wen Xie
Cui-Mei Diao
zi-qiang Liu
Tai-Lang Zhong
Original Assignee
Qi-Wen Xie
Cui-Mei Diao
zi-qiang Liu
Tai-Lang Zhong
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Qi-Wen Xie, Cui-Mei Diao, zi-qiang Liu, Tai-Lang Zhong filed Critical Qi-Wen Xie
Priority to TW97122633A priority Critical patent/TW201000074A/en
Publication of TW201000074A publication Critical patent/TW201000074A/en

Links

Landscapes

  • Image Analysis (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

A pathologic detecting system and the method by using the detection of perspective image of breast are disclosed. The purpose is to sift a pathologic image from the perspective image of breast comprising: acquiring first the perspective image of breast; then dividing the perspective image of breast into multiple area images; producing multiple characteristic parameters for each area image from the area images; acquiring the characteristic parameters, using self organized characteristic to map in the neural network; making the corresponding classification for each area image according to the predetermined multiple grades; acquiring the area images of all possible grades by fuzzy theory method to be the better manner; and finally composing the acquiring area images to find out the real pathologic images. Therefore, the present invention of the pathologic detecting system and the method can detect the possibility of suffering from the breast cancer accurately and automatically, especially for the oriental people having more percentage of fat. Comparing to the prior art, the present method can handle more accurately.

Description

201000074 九、發明說明: 【發明所屬之技術領域】 本發明係關於一種病理檢測系統以及方法,尤指利用於 檢測乳房透視影像之病理檢測系統以及方法。 【先前技術】201000074 IX. Description of the Invention: [Technical Field] The present invention relates to a pathological detection system and method, and more particularly to a pathological detection system and method for detecting a breast fluoroscopic image. [Prior Art]

乳癌是女性十大死因之一,從臨床經驗上得知,若能及 早發現初期乳癌,治癒率則令人滿意。一般而言,早期乳癌 的判定主要可由幾種方式獲得,如:觸診、外觀結構探視、χ 光檢查等等。其中又以X光檢查為最重要的方式,透過χ光 攝影術,可從乳房透視影像可以目視出這些由鈣化斑點或腫 塊組織所組成的病理影像。 除目視之外,現代醫學企望能以電腦科技輔助醫生檢 查’除了^*減少醫生因疲勞產生的人為疏失也可峨為醫生 養成教育的學習平台。如中華民國第〇〇492858號專利,係意 圖以博爾濾波器篩除長形構造物等假陽性鎮群檢波再由類 神,,星網路將簇群區分為可疑及非可疑竊群,以提昇判斷的精 確度以及效率。 ^技術與儀器係自西方引進,大多係自乳房透視影像 中以义階值大小來觸病理影像,然而,東、西方女性的乳 大ΐ差異’東方女辩肪較多,於χ光透射之 像巾’脂轉像的灰階值會躺化賴影像的灰 不論是以目視或是儀關讀,皆料產生誤差。 本發明的主要目的树提供__翻帛以檢測乳房 .201000074 透視影像之病理檢測系統以及方法,以改善上述問題。 【發明内容】 〇Breast cancer is one of the top ten causes of death in women. It is known from clinical experience that if early breast cancer is detected early, the cure rate is satisfactory. In general, the diagnosis of early breast cancer can be obtained in several ways, such as palpation, appearance structure visit, calender examination and so on. X-ray examination is the most important way. Through the photogrammetry, these pathological images composed of calcified spots or lumps can be visually observed from the breast fluoroscopic image. In addition to visual inspection, modern medicine hopes to be able to use computer technology to assist doctors to check. In addition to reducing the doctor's fatigue caused by fatigue, it can also become a learning platform for doctors to develop education. For example, the Patent No. 492858 of the Republic of China is intended to use a Boer filter to screen out false-positive towns such as long structures and then to detect them by the gods. The star network divides the clusters into suspicious and non-suspicious groups. To improve the accuracy and efficiency of judgment. ^The Department of Technology and Instruments has been introduced from the West. Most of them are from the breast fluoroscopy image to the pathological image with the size of the scale. However, the difference between the milk and the female in the East and West is more than that. The grayscale value of the image of the fat-like image of the towel will lie on the gray of the image, whether it is visual or instrumental reading, it is expected to produce errors. The main purpose tree of the present invention provides a pathological detection system and method for detecting breasts. 201000074 fluoroscopic images to improve the above problems. SUMMARY OF THE INVENTION 〇

本發明之目的在提供一種利用以檢測乳房透視影像之病 理檢測系統以及方法,除能快速、準確的判別出病理影像之 外,可藉由調變參數來適應不同的資料庫,如自體脂肪較多 的東方體質中,挑出#5化點叢集,消除用西方標準資料庫現 行作法的缺失,以做出更準確的判定。 本發明係關於-種·以檢測乳錢視影像之病理檢測 系統’用以自-乳房透視影像中篩選出—病理影像。該病理 檢測系統係進-步包含-透視影像擷取裝置、—特徵值搁取 模組、一自我組織特徵映射模組、以及一樣本擷取模組。 該透視影像擷取裝置係用以擷取該乳房透視影像。 /該特徵值練池先㈣乳錢郷·分為複數個區 塊影像,賴,雜徵錢賴_姻 -個區塊影像,產生複數種概參數。 祕之母 該自我組織特徵映射模組先係預賴數種等級,後續該 爾_,以自我組 ,特妓映射類神經網路(如卜〇聊响Map Neural 像,細取至少—種等級所對應之區塊影 為所述之病理影像。所述該樣本齡模組操取至少」種等成 201000074 ,係可進一步採用模糊理論法擷取所有相關等級所對應之 區塊影像,如此對病理影像的擷取可更無遺漏之憾。〜 因此,藉由本發明利用以檢測乳房透視影像之病理檢測 系,以及方法,利用自我組織特徵映射類神經網路進行演算 判讀,除能快速、準確的判別出病理影像之外,更可藉由調 變參數來適應不同的資料庫,如自體脂肪較多的東方體質 中’挑出㉟化喊集,消除用西方鮮㈣庫現行作法的缺 失,以做出更準確的判定。 、 關於本發明之優點與精神可以藉由以下的發明詳述及所 附圖式得到進一步的瞭解。 【實施方式】 請參閱第一圖’第一圖係本發明圖形使用者介面 (Graphical User Interface; Gui)8所顯示乳房透視影像 忉之示意圖。圖左之乳房透視影像1〇為原始影像4,圖右之 乳房透視影像10為處理過的影像結果6,圖形使用者介面8 右塊區域為多種操作指令鍵13,供使用者點選控制之用。乳 房透視影像10係以X光對乳房透視成像所成,圖中可見乳房 12輪廓與輪廓内各種灰階值的影像,灰階值較大的影像14 係有可能為鈣化斑點之病理影像26,但體脂肪的灰階值也頗 大,對判斷會造成干擾。 請參閱第二圖’第二圖係本發明病理檢測系統30之示意 圖^本發明係關於一種利用以檢測乳房透視影像10之病理^ 測系統30,用以自一乳房透視影像1〇中篩選出一病理影像 201000074 2j(病理讀26詳級第六.病理檢嶋期係主要包 ;:;視影侧取裝置32、—特徵值擁取模組34、-自我組 織特徵映咖且36、以及-樣糊取模組38。 =視影像擷取裝置32係用以梅取乳房透視影像1〇,係 及⑤包s X光影像偵測器3202、-前處理模組3204、以 久—影像強化模組3206。 Ο 1〇 彡侧廳謂係自人觀房齡⑽透視影像 則理模組3204係將X光影像偵測器32〇2所掏取之乳 苟遷現影像10調整為適當畫面尺規。 ,合第一圖參閱第二圖,觸動圖形使用者介面8中名稱 處理」之操作指令鍵13係會運作前處理模組3204, :可根據賴自動運傾處雖組_。祕理模組應 =^光影像偵測器遞職取之乳房透視影像1()調整為 為尺規’祕理敝32G4並魏魏娜像10轉換 觸動圖形使用者介面8中名稱為「影像強化」之操作指 13係會運作影像強化敝32{)6,或也報據預設自動 =影像強化模組讓。影像強化模組3施係用以強化乳 辱透硯影像10中灰階值較大之影像。 觸動_使用者介面8中名稱為「特徵值擷取」之操作 ^鏠13係會運作特徵值娜模組34,或也可根據預設自 =作特徵值擷取模組34。特徵值擷取· 34先將透視影 a嫌裳置32所完成之乳房透視影像1〇區分為複數個區塊 像22,後續,特徵值擷取模組34係針對該等區塊影像22 201000074 之每一個區塊影像22,產生複數種特徵參數。 特徵側取模組34係進-步包含_影像切割模組 3402、以及-特徵值產生模組34〇4。在觸動「特徵鋪取 之操作指令鍵13時,圖形使用者介面8會要求使用者在其右 方名為「區塊影像之晝素單錄量」之操作指令鍵13情入 ❹ 一健字,切讎組繼雜據缝字將歸透視影像 1〇區分為複數倾塊雜22。概值產生模組顺係針對 ^區塊影像22之灰階值,產生平均值、熵(Ent贿)、標 準差、以及離散餘弦轉換等四個特徵參數。 =步參考第三圖’舉例而言,第三_本發明切割區 W象22之結果影像6的示意圖。先於「區塊影像之晝素 猶指令鍵13中輸入數字7,代表每個區塊影像22 ^ :畫素單位嶋像組成之最小單位)所形成,因此影 像10會切割為區塊影像22。 然後’特徵值產生模組3404分別對每一個區塊影像烈 中=個畫素單㈣之灰階值來做,在每—個區塊影像 2,產生平均值、職⑽y)、鮮差、錢離散餘弦轉 換^四個特徵參數。其中,平均值、標準差及熵是統計特徵, 而離散餘弦轉換是頻域特徵。 回到第二圖’自我組織特徵映射模組36係進-步包含一 =等級分類模組3602 ;再參閱第一圖,在觸動「自我組織 ^映射運异」之操作指令鍵13時,_使用者介面8會要 用者在其右方名為「等級數量」之操作指令鍵13中輪入 另一個數字’自我組織特徵映射模組36係根據此另一個數字 201000074 將每-倾塊影像22分類為複數種等級 3602係根據所輸人之另-個數字,@〜&於等級刀類模組 數量,因而預定出複數種知為所需要分類等級的 ,進一,參閱第四圖,舉例而言,第四圖係本發明對區塊 影像,分配麵之結果影像6的示意圖。先於「等級數量」 插作指令鍵13中輸入另一個數字「32」,表示 可分配給所有的區塊影像22。後續,自二 ❹ ^(Self:organ1Z1ng Map Neural Network ; S〇MNN)^# , 將所預疋的複數種等級,對應分類予每—無塊影像^。 ☆ j充⑽7自我組織特徵映射轉經網路係經由競 (competitive learning) (Kohonen, 1984; rossberg’ 1987)的訓練,將高維度的輸 事The object of the present invention is to provide a pathological detection system and method for detecting breast fluoroscopic images. In addition to quickly and accurately discriminating pathological images, the parameters can be adapted to different databases, such as autologous fat. Among the more oriental physiques, the #5 points clusters are selected to eliminate the lack of current practices in Western standard databases to make more accurate judgments. The present invention relates to a pathological image for detecting a breast cancer visual image by using a pathological detection system for detecting images of breast milk. The pathology detection system is an in-step inclusion-perspective image capture device, a feature value acquisition module, a self-organizing feature mapping module, and the same capture module. The fluoroscopic image capturing device is configured to capture the breast fluoroscopic image. / The characteristic value of the practice pool first (four) milk money 郷 · divided into a plurality of block images, Lai, miscellaneous levy _ _ _ block image, the generation of a variety of parameters. The mother of the secret, the self-organizing feature mapping module first pre-receives several levels, followed by the er, _ self-group, special mapping neural network (such as the chattering Map Neural image, take at least a level The corresponding block shadow is the pathological image, and the sample age module is at least "equal to 201000074, and the fuzzy theory method can be further used to capture the block image corresponding to all relevant levels, so The acquisition of pathological images can be more regrettable. Therefore, by using the pathological detection system for detecting breast fluoroscopy images and the method, the self-organizing feature mapping neural network is used for calculus interpretation, which can be fast and accurate. In addition to discriminating pathological images, it is also possible to adapt to different databases by modulating parameters. For example, in the oriental physique with more autologous fat, 'pick up the 35 screaming set, and eliminate the lack of the current practice of using the Western fresh (four) treasury. In order to make a more accurate determination, the advantages and spirit of the present invention can be further understood from the following detailed description of the invention and the accompanying drawings. Figure 1 is a schematic view of a breast fluoroscopic image displayed by a Graphical User Interface (GUI) 8 of the present invention. The breast fluoroscopic image 1 at the left is the original image 4, and the breast fluoroscopic image 10 at the right. For the processed image result 6, the graphical user interface 8 is a plurality of operation command keys 13 for the user to select and control. The breast fluoroscopic image 10 is formed by X-ray imaging of the breast perspective, which can be seen in the figure. The image of the contours of the breast 12 and various grayscale values in the outline, the image 14 with a large grayscale value may be a pathological image of the calcified spot 26, but the grayscale value of the body fat is also quite large, which may cause interference to the judgment. Referring to the second diagram, the second diagram is a schematic diagram of the pathology detection system 30 of the present invention. The present invention relates to a pathology system 30 for detecting a breast fluoroscopic image 10 for screening a breast fluoroscopic image. Pathological image 201000074 2j (pathological reading 26 detailed level sixth. Pathological examination period is the main package;:; visual side extraction device 32, - feature value acquisition module 34, - self-organizing features and coffee and 36, and - Sample stencil Group 38. The visual image capturing device 32 is used to take a breast fluoroscopic image, and is a 5-pack s X-ray image detector 3202, a pre-processing module 3204, and a long-image enhancement module 3206. 1〇彡The side hall is the self-contained viewing age. (10) The fluoroscopic image module 3204 adjusts the chyme migration image 10 captured by the X-ray image detector 32〇2 to the appropriate picture ruler. In the first figure, referring to the second figure, the operation command key 13 of the name processing in the touch screen user interface 8 will operate the pre-processing module 3204, which can be based on the automatic dumping group _. The secret module should be = ^The light image detector is used to take the breast fluoroscopy image 1 () adjusted to the ruler's secret 敝 32G4 and Wei Wei Na like 10 conversion touch graphic user interface 8 in the name of "image enhancement" operation finger 13 The system will operate the image enhancement 敝32{)6, or it will be reported as the default automatic = image enhancement module. The image intensification module 3 is applied to enhance the image of the grayscale value in the image 10 of the insult. Touching the operation in the user interface 8 with the name "characteristic value capture" ^ 13 is to operate the feature value module 34, or the module 34 can be retrieved according to the preset feature value. The feature value extraction method 34 firstly divides the breast fluoroscopy image 1 that is completed by the perspective image into a plurality of block images 1 , and subsequently, the feature value extraction module 34 is directed to the block image 22 201000074 Each block image 22 produces a plurality of feature parameters. The feature side capture module 34 is further comprised of an image cutting module 3402 and a feature value generating module 34〇4. When the operation command key 13 of the feature extraction is touched, the graphic user interface 8 asks the user to enter the operation command key 13 named "the pixel number of the block image" on the right side. The cut-off group will be classified into a fluoroscopic image 1〇 according to the quilt image. The profile generation module is compliant with the grayscale values of the block image 22, and generates four characteristic parameters such as an average value, an entropy (Ent bribe), a standard deviation, and a discrete cosine transform. = Steps Referring to the third figure', for example, a third schematic diagram of the resulting image 6 of the cutting zone W of the present invention. The image 10 is cut into a block image 22 before the input of the number 7 in the block image of the block image, which represents the image of each block image 22 ^ : the smallest unit of the pixel unit image. Then, the 'eigen value generation module 3404 respectively performs the gray scale value of each block image and the pixel number (four), and in each block image 2, the average value, the job (10) y), and the difference are produced. And the discrete discrete cosine transform ^ four characteristic parameters. Among them, the mean, standard deviation and entropy are statistical features, and the discrete cosine transform is the frequency domain feature. Back to the second figure 'self-organizing feature mapping module 36 is step-by-step Including a = level classification module 3602; referring to the first figure, when the operation command key 13 of "self-organizing ^ mapping" is touched, the user interface 8 will be called "number of levels" on the right side of the user interface 8 The operation command key 13 is rotated into another number. The self-organizing feature mapping module 36 classifies each per-block image 22 into a plurality of levels according to the other number 201000074. The system is based on another number of the input person. , @〜& in the number of grade cutter modules, due to For example, the fourth figure is a schematic diagram of the result image 6 of the block image and the distribution face of the present invention. Another number "32" is input before the "number of levels" is inserted into the command key 13, indicating that all of the block images 22 can be assigned. Subsequently, from the second ❹ ^ (Self:organ1Z1ng Map Neural Network; S〇MNN)^#, the pre-existing plural grades are correspondingly classified into each-no-block image ^. ☆ j charge (10)7 self-organizing feature mapping through the network through the training of competitive (Kohonen, 1984; rossberg' 1987) training, high-dimensional output

先決定的低維度帥中(通常維度科2),絲訓練H 、自我組織特徵映射_、_路的輸出單元表現所有的輸入資 料’並產生類似功能的輪出單元彼此群聚的現象。換言之, 類似的輸人資料項會映射到相同的自我組織特徵映射二神經 =路輸出單元,相鄰近的輸出單元表現相似的輸入資料項特 因此’自我組織特徵映_神_路可以將高維度的 ^資料’以低維度的空間卿表現,避免了統計階層集群分 ^法運用於大量資料時,難關讀的情況。自我組織特徵映 射類神_路已經是現行朗的理論,自餘晴徵映射模 j 36係將自她織魏崎類神棚路倾其巾,在此 贅述。 201000074 接著,觸動第一圖中圖形使用者介面8中名稱為「樣本 選取」之操作指令鍵13,係會運作樣本擷取模組兕,或也可 根據預設自動運作樣本擷取模組38。樣本擷取模組38係搁 取至少一種等級所對應之區塊影像22,其中組成樣本操取模 組38所擷取之區塊影像22,則係後續可組成為所述之病理' 影像26。配合第二圖所示,樣本擷取模組38進一步包含一 影像套接模組3802、一病理點過遽模組3804、以及一標示模 組 3806。 Λ 影像套接模組3802係將所擷取該種等級對應之區塊影 像22,於原始影像4的乳房透視影像1〇中之對稱位置,擷 取組合出一樣本影像24。 進一步請參閱第五圖,舉例而言,第五圖係本發明操取 特定等級區塊影像22之結果影像6的示意圖。本實施例樣本 擷取模組38係取預設之「第31等級」(根據過去的實驗印證 所累積之經驗選取),因此如圖示,所有第31等級的區塊影 像22被選取出來,並套接在原始影像4上,將原始影像4 中相同位置之區塊影像22顯示出來,此等區塊影像22所組 成的即為樣本影像24。 觸動第一圖中圖形使用者介面8中名稱為「病理點過濾」 之操作指令鍵13係會運作病理點過濾模組3804,或也可根 據預設自動運作病理點過濾模組38〇4。在觸動「病理點過濾」 之操作指令鍵13時’圖形使用者介面8會要求使用者在其右 方名為「灰階門播值」之操作指令鍵13中輸入再一個數字, 此輸入之再一個數字即為灰階門檻值。病理點過濾模組38〇4 12 201000074 係根據此再-錄字,自所之區塊影㈣巾勒灰階值 在灰階f灘之上之晝素。所以,病理點過_組期可根 =6灰階門楹值’於樣本影像24中擷取出所述真正的病 ❹ ❹ 進一步參閱第六圖,舉例而言,第六圖係本發明自特定 等級區塊雜22中擷取病理影像26之結果影像6的示意 圖。於「紐Η檻值」之操作指令鍵13巾輸人灰階門捏值 「220」,則如圖示—般’影像結果6會顯示出灰階值在灰階 門檻值之上之晝素,該等晝素即構成病理影像%。 參閱第七圖,第七圖係本發明最後結果影像6之示意 圖。觸動第一圖中圖形使用者介面8中名稱為「標示」之操 作指令鍵13 ’係會運作標示模組讓,或也可根據預設自動 運作標示歡3麵。標示模組讓相較人性化的方式, 以至少-標示框4G來框標出所述之病理影像26,以方便醫 師以及病人或其他研究人員以目視直接觀看。 此外,請翔第八圖,第八圖係本發縣—實施例之功 能方塊示意圖。前述預設等級係根據經驗來進行選取,另一 實施例係引進模糊理論法(Fuzzy^進行樣本選取,與前實施 例不同的财單單以預設的丨個或2個等級來擷取可疑的區 塊影像22,目為絲可能具有病理影像% _塊影像22會 跨及好幾辦級’卿利闕糊理論可將可疑的數種等級所 對應的區塊影像22統統擷取出來,則所得到的病理影像沈 就會更精確。 依圖示,如前述之病理檢測系統3〇 一般,其中較不同的 13 201000074 疋樣本擷取模組38進一步包含一模糊化模組5〇、一影像套 接模組3802、一病理點過濾模組38〇4、以及一標示模組洲⑽。 模糊化模組50係以模糊理論法擷取所有相關等級所對 應之區塊影像22。 。影像套接模組3802係將所擷取所有相關等級所對應之 區塊衫像22,於乳房透視影像1〇中之對稱位置,擷取組合 出一樣本影像24。病理點過濾模組38〇4係根據預定之灰階 ❹ 門根值,於樣本影像24中擷取出所述之病理影像26。標示 模組3806係以至少一標示框4〇來框標出所述之病理影像 26 ° - 模糊化模組50會偵測自組性類神經網路(SOMNN)所分類 料級中’ S否有足夠蝴化點,如果在__倾塊影像22 中有足夠的鈣化點,並且此區塊影像22鈣化點的叢集狀態符 合,則視此區塊影像22為所需選取之等級。補充說明,以模 糊理論對相關分級所做出之選擇以及篩選,至少有二實施例 ❿ 可陳述於後。 第一例如下,以灰階值以及超過特定灰階值所佔百分比 數量,分別分為「低」、「正常」、「高」、「很高」、「非常高」 等五個區等。其中,預先定義出語義學(linguistic),語義 學可例如下述四點: 1. 當一個區塊影像22中灰階值的平均值較「很高」區 等為高,且此區塊影像22中百分比數量為「很高」區等,則 可視此區塊影像22之等級為所欲選取之等級。 2. 當一個區塊影像22中灰階值的平均值低於「高」區 201000074 等,且此區塊影像22中百分比數量為「正常」區等,則可視 此區塊影像22之等級為所欲選取之等級。 3. 當一個區塊影像22中灰階值的平均值高於「很高」 區等,並且其中99%的晝素單位20也是屬於「很高」區等, 則此區塊影像22之等級應視為不被選取之等級。 4. 其他狀況的區塊影像22之等級,都被視為不被選取 之等級。 © 然後針對每一個等級(如等級1〜等級32)的區塊影像 22,分別計算出每一個等級的平均灰階值以及超過特定灰階 值所佔百分比數量(如超過灰階值18〇晝素單位2〇所佔百分 *數量)。接著,比對預定之語義學與第九圖之歸屬函數 (membership f_i〇n) ’第九圖之歸屬函數可根據經驗值來 取得。若此等級之區塊影像22符合歸屬函數與語義學中肯定 所需求等級的條件,則做此區塊影像22為所需選取之等 級。 ❹ 例=:「等、級28」的平均灰階值為150,百分比數量為 二8(18/〇,則「等級28」於灰階值歸屬函數位於「正常」 「高」區等)’且「等級28」於百分比數量歸屬函 厂耸;正常」區荨,此係符合上述語義學之第二點。所以, 驗選被判定為所需選取之等級。如此,相較先前以經 的莖如疋等級之方法’以模糊理論可進一步選出更多相關 的等級’靴魏影像26更為精確。 t卜盔例如下’同樣的,歸屬函數(不論是灰階值或是百分 ’皆分別分為「低」、「正常」、「高」、「很高」、「非常 15 201000074 问」等五個區等,也比較_之語義學。 例如級?大:圍的等級為我們所要的等級範圍, 有的區塊影像選取此等級範圍所 ^==平__砂嫌量,再將每一個 ilUf ^平均灰階值與百分比數量,依據前例所 ❹ 蘇上1,匕較分類’則可以自此大範圍的等級 判〜所雷::不狀區塊影像22,剩下的區塊影像22,即為 之區塊影像22。如此,較前例之方法,以模糊 f两可進—步選出貼切的區塊影像22,以使病理影像26更 為快速。 %參閱第十圖’第十圖係本發明病理檢測方法之流程 圖。本發明也係-麵用讀測制透娜像之病理檢測方 法,用以自-乳房透視影像中篩選出—病理影像26,該病理 檢測方法係包含下列步驟: 步驟S02 :以X光擷取乳房透視影像1〇。 步驟S04 :將乳房透視影像1〇轉換為灰階值。 步驟S06 :將所擷取之乳房透視影像10調整為適當畫面 尺規。 步驟S08 :於乳房透視影像1〇轉換為灰階值後,強化該 乳房透視影像10中灰階值較大之影像。 步驟S10 :分割乳房透視影像1〇為複數個區塊影像22。 201000074 ^驟S12 :針對該等區塊影像22之每一個區塊影像22, 產生複數種特徵參數,所述產生複數種特徵參數,於本發明 實施例中主要係針對該等區塊影像22之灰階值,產生平均 值、熵(Entropy)、標準差、以及離散餘弦轉換等四個特徵參 數。 步驟S14 :預設出複數種等級的數量。 y驟S16 .掏取所述該等特徵參數,以自我組織特徵映 ❹ 射類神經網路(Self-organizing Map Neural Ne1:work ; NN) ’將所預定的複數種等級,對應分類予每一個區塊影 像22。 步驟S18 :擷取至少一種等級所對應之區塊影像22,於 該乳錢郷像t之對獅置,娜組合丨―樣本影像24。 步驟S20 :根據預定之灰階門檻值,於樣本影像%中擷 取出所述之病理影像26。 步驟S22 .以至少一標示框4〇來框標出所述之病理影像 从 26〇 此外’請參閱第十-圖’第十—圖係本發明另一實施例 =流程圖。另一例係以模糊理論法來進行樣本擷取,其中與 前例所不一樣的是該樣本擷取方法係於「以自 . 射類神經網路將所預定的複數種等級,對應二予= 塊影像22」之步驟後,進一步包含下列步驟: 步驟S30 :以模糊理論法擷取所有相關等級所對應之區 塊影像22。 〜 17 201000074 步驟S32 .將所擷取所有相關等級所對應之區塊影像 22 ’於該乳房透視影像中之對稱位置,擷取組合出—樣本影 像24。 步驟S34 :根據預定之灰階門檻值,於樣本影像%中掏 取出所述之病理影像26。 步驟S36 .以至少一標示框4〇來框標出所述 26。 豕 ❻First decided on the low-dimensional handsome (usually dimension 2), the silk training H, the self-organizing feature map _, the output unit of the _ road represents all the input data' and the phenomenon that the round-out units with similar functions are clustered with each other. In other words, similar input data items will be mapped to the same self-organizing feature map two nerves = road output units, and adjacent output units will behave similarly to input data items. Therefore, 'self-organizing feature maps _ god _ roads can be high-dimensional The ^ data's performance in the low-dimensional space, avoiding the situation that the statistical hierarchy clusters are difficult to read when applied to large amounts of data. The self-organizing feature mapping god _ road is already the current theory of Lang, since the Yu Qingzheng mapping module j 36 series will pour the towel from her weaving Weiqi class Shenshen Road, hereby repeat. 201000074 Next, the operation command key 13 named "sample selection" in the graphical user interface 8 in the first figure is activated, and the sample capture module is operated, or the sample capture module 38 can be automatically operated according to the preset. . The sample capture module 38 is configured to capture at least one level of the block image 22 corresponding to the level, wherein the block image 22 captured by the sample operation module 38 can be subsequently formed into the pathological image 26 . As shown in the second figure, the sample capture module 38 further includes an image socket module 3802, a pathology point overmodulation module 3804, and a labeling module 3806.影像 The image socket module 3802 extracts the same image 24 from the symmetrical position in the breast fluoroscopic image 1 of the original image 4 by capturing the block image 22 corresponding to the level. Further referring to the fifth figure, for example, the fifth figure is a schematic diagram of the resulting image 6 of the particular level block image 22 of the present invention. The sample capture module 38 of the present embodiment takes the preset "31st level" (according to the experience accumulated in the past experimental verification), so as shown, all the block images 22 of the 31st level are selected. The patch image 22 of the same position in the original image 4 is displayed on the original image 4, and the image 22 of the block image is the sample image 24. The operation command key 13 named "Pathological point filtering" in the graphical user interface 8 in the first figure is activated to operate the pathological point filtering module 3804, or the pathological point filtering module 38〇4 can be automatically operated according to the preset. When the operation command key 13 of the "pathological point filter" is touched, the graphical user interface 8 asks the user to input another number in the operation command key 13 named "gray level homing value" on the right side. Another number is the grayscale threshold. The pathological point filter module 38〇4 12 201000074 is based on this re-recorded word, from the block shadow (four) towel gray scale value on the gray scale f beach. Therefore, the pathological point _ group period can be root = 6 gray scale threshold value 'to extract the true disease 样本 in the sample image 24 ❹ further refer to the sixth figure, for example, the sixth picture is the invention self-specific A schematic diagram of the result image 6 of the pathological image 26 taken from the rank block. In the operation button of the "New Η槛 value", the input value of the grayscale door is "220", as shown in the figure, the image result 6 will show the gradation of the grayscale value above the grayscale threshold. These alizarins constitute % of pathological images. Referring to the seventh drawing, the seventh drawing is a schematic view of the final result image 6 of the present invention. Touching the operation command button 13 of the graphic user interface 8 in the first figure named "Marking" will operate the indicator module, or it can automatically operate the indicator according to the preset. The labeling module allows the pathological image 26 to be framed by at least the label box 4G in a manner that is more user-friendly, so that the doctor and the patient or other researcher can directly view it visually. In addition, please draw the eighth picture, the eighth picture is a functional block diagram of the county. The foregoing preset level is selected according to experience, and another embodiment introduces a fuzzy theory method (Fuzzy^ performs sample selection, and the financial bills different from the previous embodiment use the preset one or two levels to draw suspicious Block image 22, the target silk may have pathological image% _ block image 22 will span several levels of 'Qing Li paste theory can extract the block image 22 corresponding to suspicious several levels, then the The obtained pathological image is more accurate. As shown in the figure, as described above, the pathological detection system is generally the same, wherein the different 13 201000074 疋 sample capture module 38 further comprises a fuzzification module 5 〇, an image set The module 3802, a path point filter module 38〇4, and a label module continent (10). The fuzzification module 50 captures the block image 22 corresponding to all relevant levels by a fuzzy theory method. The module 3802 extracts the same block image 22 corresponding to all relevant levels, and extracts the same image 24 from the symmetrical position in the breast fluoroscopic image 1〇. The pathological point filtering module 38〇4 is based on Scheduled grayscale ❹ The threshold value is obtained by extracting the pathological image 26 from the sample image 24. The indicator module 3806 marks the pathological image 26° with at least one frame 4〇 - the fuzzification module 50 detects In the classification class of the self-organizing neural network (SOMNN), there is enough vanishing point if there is enough calcification point in the __dump image 22, and the cluster state of the calcification point of this block image 22 If it is met, then the block image 22 is regarded as the level to be selected. Supplementary explanation, the selection and screening of the relevant classification by the fuzzy theory, at least two embodiments may be stated later. First, for example, The grayscale value and the percentage of the percentage of the specific grayscale value are divided into five zones, such as "low", "normal", "high", "very high", and "very high". Among them, the semantics are defined in advance. Linguistic, semantics can be, for example, the following four points: 1. When the average value of the grayscale values in a block image 22 is higher than the "very high" area, and the percentage of the image in the block 22 is " Very high, etc., you can see the level of this block image 22 as desired 2. When the average value of the grayscale value in a block image 22 is lower than the "high" zone 201000074, etc., and the percentage of the image in the block image 22 is "normal" zone, etc., the block image can be visualized. The level of 22 is the level to be selected. 3. When the average value of the gray level value in a block image 22 is higher than the "very high" area, and 99% of the elemental units 20 are also in the "very high" area. Etc., the level of the block image 22 should be regarded as the level that is not selected. 4. The level of the block image 22 in other conditions is regarded as the level that is not selected. © Then for each level (such as level The block image 22 of 1~level 32) calculates the average grayscale value of each level and the percentage of the specific grayscale value exceeding the grayscale value (if the grayscale value is less than 18% of the prime unit) Quantity). Next, the predetermined semantics of the comparison and the attribution function of the ninth graph of the membership function (membership f_i〇n) of the ninth graph can be obtained based on empirical values. If the block image 22 of this level meets the conditions of the attribution function and the level of affirmation required in semantics, then the block image 22 is made to be the selected level. ❹ Example =: "Equivalent, Level 28" has an average grayscale value of 150, and the percentage number is two 8 (18/〇, then "Level 28" in the grayscale value attribution function is in the "normal" "high" area, etc." And "Level 28" is attributed to the percentage of the number of the factory; the normal "zone", which is in line with the second point of the above semantics. Therefore, the selection is judged to be the level of selection required. Thus, the more relevant level of the 'Wei Wei image 26' can be further refined by the fuzzy theory than the previous method of stalks such as sputum. For example, the same, the attribution function (whether grayscale value or percent) is divided into "low", "normal", "high", "very high", "very 15 201000074 question", etc. Five districts, etc., also compare the semantics of _. For example, level? Large: the level of the surrounding is the range of levels we want, and some block images select this level range ^== 平__砂量量, and then each An ilUf ^ average gray scale value and percentage number, according to the previous example, Su Shang 1, 匕 category ' can be judged from this wide range of grading ~: Thunder block image 22, the remaining block image 22, that is, the block image 22. Thus, compared with the method of the previous example, the image 22 of the block can be selected by blurring, so that the pathological image 26 is faster. The invention is a flow chart of the pathological detection method of the invention. The invention also relates to a pathological detection method for reading and measuring the image of the lens, and is used for screening the pathological image 26 from the breast-seeing image, the pathological detection method comprises The following steps: Step S02: Capture the breast fluoroscopic image 1 by X-ray. Step S04: Converting the breast fluoroscopic image 1〇 to a grayscale value. Step S06: Adjusting the captured breast fluoroscopic image 10 to an appropriate picture ruler. Step S08: Strengthening the breast after the breast fluoroscopic image is converted to a grayscale value The image of the grayscale value in the image 10 is fluoroscopy. Step S10: segmenting the breast fluoroscopic image 1 into a plurality of block images 22. 201000074 ^Step S12: for each block image 22 of the block image 22, generated A plurality of characteristic parameters, wherein the plurality of characteristic parameters are generated. In the embodiment of the present invention, the gray level values of the image 22 of the block are mainly generated, and an average value, an entropy, a standard deviation, and a discrete cosine transform are generated. Four characteristic parameters: Step S14: Presetting the number of the plurality of levels. yStep S16. Extracting the characteristic parameters to self-organizing Map Neural Ne1:work NN) 'Classify the predetermined plurality of levels to each block image 22. Step S18: Capture the block image 22 corresponding to at least one level, and set the lion to the lion, Na combination Sample image 24. Step S20: extracting the pathological image 26 from the sample image % according to a predetermined grayscale threshold value. Step S22. Frame the pathological image from at least one label frame 4〇 from 26 〇In addition, please refer to the tenth-figure tenth--the other embodiment of the present invention=flow chart. Another example is to use the fuzzy theory method to perform sample extraction, wherein the sample is not the same as the previous example. The method is further characterized by the following steps: "Steps to the predetermined number of levels from the radio network, corresponding to the two predictions = block image 22", further comprising the following steps: Step S30: extracting all relevant levels by fuzzy theory Corresponding block image 22. ~ 17 201000074 Step S32. Combine the block image 22' corresponding to all relevant levels to the symmetrical position in the breast fluoroscopic image, and extract the sample image 24. Step S34: extracting the pathological image 26 from the sample image % according to a predetermined grayscale threshold value. Step S36. The 26 is indicated by at least one frame 4〇.豕 ❻

請參閱第十二圖’第十二圖係比對習知以及本發明所分 析之病理影像之示意圖。圖左塊為f知電腦輔助診斷系統& CAD所自動化檢測之病理影像,圖右塊為本發明所自動化分 析之病理影像’觀對,習知之技術會旨職判為病理影 像’而本發明之技術所觸之_影像賴為鮮。’ 因此’藉由本發明利用以檢測乳房透視影像之病理檢蜊 =統30以及方法,_自餘織類映_神軸路進行演 算判讀’除能快速、準確的判別出病理影像26之外,更可藉 由調變參數來it應不_資料庫,如自體贿較多的東 質中’挑_化點叢集’消除用西方標準資料庫現行作 缺失,以做出更準確的判定。 、 、、藉由以上較佳频實施例之詳述,係希望能更加清 述本發明之概與精神’而並非以上述__較佳: 施例來對本發明之鱗加以關。減地,其目的是^ 涵蓋各種改變及具相等性的安排於本發明所欲申 = 圍的範略内。 κ寻利範 .201000074 【圖式簡單說明】 一第一圖係本發明圖形使用者介面所顯示乳房透視影 之示意圖; 第二圖係本發明病理檢測系統之示意圖; 第一圖係本發明切割區塊影像之結果影像的示意圖; i第四圖係本發明對區塊影像分配等級之結果影像的示 思、圖; .第五圖係本發明擷取特定等級區塊影像之結果影像的 示意圖; 第圖係本發明自特定等級區塊影像中擷取病理影像 之結果影像的示意圖; 第七圖係本發明最後結果影像之示意圖; 第八圖係本發明另一實施例之功能方塊示意圖; 第九圖係本發明灰階值與百分比數量之歸屬函數之示 意圖; 第十圖係本發明病理檢測方法之流程圊; 第十一圖係本發明另一實施例之流程圖;以及 — 第十二圖係比對習知以及本發明所分析之病理影像之 不意圖。 【主要元件符號說明】 圖形使用者介面8 乳房透視影像1〇 19 201000074 ❹ 原始影像4 乳房12 操作指令鍵13 透視影像擷取裝置32 前處理模組3204 特徵值擷取模組34 特徵值產生模組3404 特徵等級分類模組3602 影像套接模組3802 標示模組3806 區塊影像22 病理影像26 模糊化模組50 影像結果6 影像14 病理檢測系統30 X光影像偵測器3202 影像強化模組3206 影像切割模組3402 自我級織特徵映射模組36 樣本梅取模组38 病理點過濾模組3804 晝素單位20 樣本影像24 標示框4〇Referring to Figure 12, the twelfth image is a schematic representation of a conventional pathology and a pathological image analyzed by the present invention. The left block of the figure is the pathological image of the computer-aided diagnosis system & CAD automatic detection. The right block of the figure is the pathological image of the automated analysis of the invention, and the conventional technology is judged as the pathological image. The image that the technology touches is fresh. 'Therefore, by using the present invention to detect the pathological examination of the breast fluoroscopic image and the method, _ from the residual weaving type _ _ _ axis to calculate the interpretation of 'dissociation quickly and accurately discriminate the pathological image 26, It is also possible to make a more accurate judgment by using the modulating parameters to determine whether it should be _ database, such as the self-bashing of the East, and the selection of the Western Standard Database. And the detailed description of the preferred embodiments of the present invention is intended to provide a more detailed description of the present invention. The purpose of reducing the land is to cover the various changes and equivalence arrangements within the scope of the invention. κ寻利范.201000074 [Simplified illustration of the drawings] A first diagram is a schematic diagram of a breast perspective image displayed by the graphical user interface of the present invention; the second diagram is a schematic diagram of the pathological detection system of the present invention; A schematic diagram of a result image of a block image; i is a diagram and a diagram of the result image of the block image distribution level of the present invention; and the fifth figure is a schematic diagram of the result image of the image of the specific level block of the present invention; The figure is a schematic diagram of the result image of the pathological image taken from the image of the specific level of the present invention; the seventh figure is a schematic diagram of the final result image of the present invention; the eighth figure is a functional block diagram of another embodiment of the present invention; 9 is a schematic diagram of the attribution function of the gray scale value and the percentage quantity of the present invention; the tenth diagram is the flow chart of the pathological detection method of the present invention; the eleventh figure is a flow chart of another embodiment of the present invention; and - the twelfth The drawings are intended to be inconsistent with conventional and pathological images analyzed by the present invention. [Main component symbol description] Graphic user interface 8 Breast fluoroscopic image 1〇19 201000074 原始 Original image 4 Breast 12 Operation command key 13 Perspective image capture device 32 Pre-processing module 3204 Characteristic value acquisition module 34 Characteristic value generation mode Group 3404 Feature Level Classification Module 3602 Image Socket Module 3802 Label Module 3806 Block Image 22 Pathology Image 26 Fuzzy Module 50 Image Results 6 Image 14 Pathology System 30 X-ray Image Detector 3202 Image Enhancement Module 3206 Image Cutting Module 3402 Self-leveling Feature Mapping Module 36 Sample Metering Module 38 Pathology Point Filtering Module 3804 Elementary Unit 20 Sample Image 24 Marking Box 4〇

2020

Claims (1)

201000074 十、申請專利範圍: 種利用以檢測乳房透視影像之病理檢測系 統,用以自一乳房透視影像中篩選出一病理影 像,該病理檢測系統係進一步包含: —透視影像擷取装置,係用以擷取該乳房透視 影像; 特徵值擷取模組,將該乳房透視影像區分為 複數個區塊影像,該特徵值擷取模組係針對 該等區塊影像之每一個區塊影像,產生複數 種特徵參數; 自我纟且織特徵映射模組,係預定複數種等 級,該自我組織特徵映射模組係擷取所述 該等特徵參數,以自我組織特徵映射類神 經網路(Self-organizing Map Neural Network ; SOMNN),將所預定的複數種等 級’對應分類予每一個區塊影像;以及 一樣本擷取模組,係擷取至少一種等級所對應 之區塊影像,其中組成該樣本擷取模組所擷 取之區塊影像,則係組成為所述之病理影 像。 21 201000074 2、 如申請專利範圍第丨項所述之病理檢測系統,其 中該透視影像擷取裝置進一步包含: 一前處理模組,係將該乳房透視影像轉換為灰 階值。 3、 如申請專利範圍第2項所述之病理檢測系統,其 中該透視影像擷取裝置進一步包含: 一 X光影像偵測器,係擷取該乳房透視影像, 該前處理模組係將該X光影像偵測器所擷取 之乳房透視影像調整為適當畫面尺規;以及 一影像強化模組,係強化該乳房透視影像中灰 階值較大之影像。 4、 如申請專利範圍第2項所述之病理檢測系統,其 〇 中該特徵值擷取模組進一步包含: 一影像切割模組,該影像切割模組係將該乳房 透視影像區分為複數個區塊影像;以及 一特徵值產生模組,係針對該等區塊影像之灰 階值,產生平均值、熵(Entropy)、標準差、 以及離散餘弦轉換等四個特徵參數。 5、 如申請專利範圍第1項所述之病理檢測系統,其 22 201000074 中該自我組織特徵映射模組係進一步包含· 一特徵等級分類模組,係根據所輸人之數字, 預定為所需要分類等級的數量。 6、如中請專利㈣第2項所述之病理檢喝統,其 中β亥樣本掏取模組進一步包含: ❹ -影像套接模組’係將所擷取該種等級對應之 區塊影像,於該乳房透視影像中之對稱位 置,擷取組合出一樣本影像;以及 一病理點過濾模組,係根據預定之灰階門檻 值,於該樣本影像中擷取出所述之病理影 像。 ⑹申請專利範圍第6項所述之病理檢測系統,其 中該樣本擷取模組進一步包含: —標示模組,係以至少一標示框來框標出所述 之病理影像。 如申明專利範圍第2項所述之病理檢測系統,其 中該樣本操取模組進一步包含: —模糊化模組,係以模糊理論法擷取所有相關 等級所對應之區塊影像; 23 201000074 p像套接n係將所擁取所有相關等級所 對應之區塊影像,於該乳房透視影像中之對 稱位置,指員取組合出一樣本影像;以及 病理點過濾、模組,係根據預定之灰階門播 值’於該樣本影像中擷取出所述之病理影 像。 、 如申"月專利圍第8項所述之病理檢測系統,其 中該樣本_取模組進一步包含: -標示模組,係以至少一標示框來框標出所述 之病理影像。 10 ❹ 種利用以檢測乳房透視影像之病理檢測方 法,用以自一乳房透視影像中篩選出一病理影 像’該病理檢測方法係包含下列步驟: 擷取該乳房透視影像; 分割該乳房透視影像為複數個區塊影像; 針對該等區塊影像之每一個區塊影像,產生複 數種特徵參數; 預設出複數種等級的數量; 擷取所述该等特徵參數,以自我纟且織特徵映射 24 201000074 類神經網路(Self-organizing Map Neural Network ; SOMNN),將所預定的複數種等 級,對應分類予每一個區塊影像;以及 擷取至少一種等級所對應之區塊影像,其中組 成所擷取之區塊影像,則係為所述之病理影 像。 ❹ 11 12 13、 如申請專利範圍第1 〇項所述之病理檢測方法, 其中該透視影像擁取方法於擁取該乳房透視影 像後,進一步包含下列步驟: 將該乳房透視影像轉換為灰階值。 如申請專利範圍第11項所述之病理檢測方法, 其中針對擷取該乳房透視影像,係進一步包含下 列步驟: 以X光擷取該乳房透視影像; 將所擷取之乳房透視影像調整為適當畫面尺 規;以及 於該乳房透視影像轉換為灰階值後,強化該乳 房透視影像中灰階值較大之影像。 如申請專· ®第U項所述之病理檢測方法, 25 201000074 其中所述產生複數種特徵參數,係針對該等區塊 影像之灰階值,產生平均值、熵(Entr〇py;)、標準 差、以及離散餘弦轉換等四個特徵參數。 14 ❹ 15、 ❹ 16、 、如申請專利範圍第U項所述之病理檢測方法, 其中該樣本擷取方法於擷取至少一種等級所對 應之區塊影像後,進一步包含下列步驟: 將所擷取該種等級對應之區塊影像,於該乳房 透視影像中之對稱位置,擷取組合出一樣本 影像;以及 根據預定之灰階門襤值’於該樣本影像中梅取 出所述之病理影像。 如申請專利範圍第14項所述之病理檢測方法, 其中該樣本擷取方法於擷取出所述之病理影像 後’進一步包含下列步驟: 以至少-標示框來框標出所述之病理影像。 如申請專利範圍第11項所述之病理檢測方法, 其中該樣本擷取方法於以自我組織特徵映射類 神經網路將所預定的複數種等級,對應分類予每 一個區塊影像後,進一步包含下列步驟: 26 201000074 以模糊理論法擷取所有相關等級所對應之區塊 影像; 將所掏取所有相關等級所對應之區塊影像,於 "亥礼房透視影像中之對稱位置,擷取組合出 一樣本影像;以及201000074 X. Patent application scope: A pathological detection system for detecting breast fluoroscopic images for screening a pathological image from a breast fluoroscopic image, the pathological detection system further comprising: - a fluoroscopic image capturing device, used The breast value image is captured by the feature value capture module, and the breast perspective image is divided into a plurality of block images, and the feature value capture module is generated for each block image of the block images. a plurality of characteristic parameters; a self-destructive feature mapping module, which is predetermined to be a plurality of levels, the self-organizing feature mapping module extracting the characteristic parameters to self-organize a feature-like neural network (Self-organizing) Map Neural Network (SOMNN) classifies the predetermined plurality of levels into corresponding image of each block; and the same capture module, which captures block images corresponding to at least one level, wherein the sample is composed. The block image captured by the module is composed of the pathological image. The method of claim 1, wherein the fluoroscopic image capturing device further comprises: a pre-processing module for converting the breast fluoroscopic image into a grayscale value. 3. The pathological detection system of claim 2, wherein the fluoroscopic image capturing device further comprises: an X-ray image detector that captures the breast fluoroscopic image, the pre-processing module The breast fluoroscopic image captured by the X-ray image detector is adjusted to an appropriate picture ruler; and an image enhancement module is used to enhance the image with a large grayscale value in the breast fluoroscopic image. 4. The pathological detection system of claim 2, wherein the feature value capture module further comprises: an image cutting module, wherein the image cutting module divides the breast perspective image into a plurality of The block image; and a feature value generating module generate four characteristic parameters such as an average value, an entropy, a standard deviation, and a discrete cosine transform for the grayscale values of the image of the block. 5. The pathological detection system according to claim 1, wherein the self-organizing feature mapping module further includes a feature level classification module, which is predetermined according to the number of the input person. The number of classification levels. 6. The pathological inspection system described in item 2 of the patent (4), wherein the β-hai sample extraction module further comprises: ❹ - image socket module is to capture the image of the block corresponding to the level The symmetrical position in the breast fluoroscopic image captures the same image; and a pathological point filtering module extracts the pathological image from the sample image according to a predetermined grayscale threshold value. (6) The pathological detection system of claim 6, wherein the sample capture module further comprises: - a labeling module, wherein the pathological image is framed by at least one indicator box. The pathological detection system of claim 2, wherein the sample manipulation module further comprises: - a fuzzification module, which uses a fuzzy theory method to capture block images corresponding to all relevant levels; 23 201000074 p Like the socket n image, the block image corresponding to all relevant levels is captured, and the symmetrical position in the breast fluoroscopic image is taken by the finger to combine the same image; and the pathological point filtering and module are according to the predetermined The grayscale homing value 'extracts the pathological image in the sample image. The pathological detection system of claim 8, wherein the sample_taking module further comprises: - a labeling module, wherein the pathological image is framed by at least one indicator box. 10 ❹ a method for detecting a pathological detection of a breast fluoroscopic image for screening a pathological image from a breast fluoroscopic image. The pathological detection method comprises the steps of: capturing the breast fluoroscopic image; dividing the breast fluoroscopic image into a plurality of block images; generating, for each of the block images of the block images, a plurality of feature parameters; presetting a plurality of levels; and extracting the feature parameters to map the features 24 201000074 Self-organizing Map Neural Network (SOMNN), which classifies a predetermined plurality of levels into each block image; and captures block images corresponding to at least one level, wherein the composition The image of the block captured is the pathological image. The method of detecting a pathology according to the first aspect of the invention, wherein the fluoroscopic image capturing method further comprises the following steps after the breast fluoroscopic image is captured: converting the breast fluoroscopic image into a grayscale value. The method for detecting a pathology according to claim 11, wherein the method for extracting the breast fluoroscopic image further comprises the steps of: capturing the breast fluoroscopic image by X-ray; adjusting the captured breast fluoroscopic image to appropriate The screen ruler; and after the breast fluoroscopic image is converted into a grayscale value, the image with a larger grayscale value in the breast fluoroscopic image is enhanced. The pathological detection method described in the application of the U.S. U, 25 201000074, wherein the plurality of characteristic parameters are generated, and the average value and entropy (Entr〇py;) are generated for the grayscale values of the image of the block. Four characteristic parameters, such as standard deviation and discrete cosine transform. 14 ❹ 、 、 、 、 、 、 、 、 、 、 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理 病理Taking the block image corresponding to the level, in the symmetrical position in the breast fluoroscopic image, extracting the same image; and extracting the pathological image in the sample image according to the predetermined grayscale threshold value . The pathological detection method according to claim 14, wherein the sample extraction method comprises the following steps after the removal of the pathological image: the pathological image is framed by at least a label. The pathological detection method according to claim 11, wherein the sample extraction method further comprises: after the self-organizing feature mapping neural network classifies the predetermined plurality of levels, correspondingly to each block image, further including The following steps: 26 201000074 Obtain the block image corresponding to all relevant levels by the fuzzy theory method; take the block image corresponding to all relevant levels, and take the symmetrical position in the perspective image of the "Haifang room Combine the same image; and 17 根據預定之灰階門檻值’於該樣本影像中操取 出所述之病理影像。 其申明專利範圍第16項所述之病理檢測系統, 後^樣本擁取方法於掏取出所述之病理影像 進一步包含下列步驟:17 The pathological image is processed in the sample image according to a predetermined grayscale threshold value. The pathological detection system described in claim 16 of the patent scope, the method for extracting the pathology after extracting the pathological image further includes the following steps: 以至少一 標示框來框標出所述 之病理影像 27The pathological image is indicated by at least one frame.
TW97122633A 2008-06-18 2008-06-18 Pathologic detecting system by using the detection of perspective image of breast and the method thereof TW201000074A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW97122633A TW201000074A (en) 2008-06-18 2008-06-18 Pathologic detecting system by using the detection of perspective image of breast and the method thereof

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW97122633A TW201000074A (en) 2008-06-18 2008-06-18 Pathologic detecting system by using the detection of perspective image of breast and the method thereof

Publications (1)

Publication Number Publication Date
TW201000074A true TW201000074A (en) 2010-01-01

Family

ID=44824389

Family Applications (1)

Application Number Title Priority Date Filing Date
TW97122633A TW201000074A (en) 2008-06-18 2008-06-18 Pathologic detecting system by using the detection of perspective image of breast and the method thereof

Country Status (1)

Country Link
TW (1) TW201000074A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI790689B (en) * 2021-07-21 2023-01-21 財團法人資訊工業策進會 Method and electric device for processing breast tomosynthesis images
TWI823897B (en) * 2018-02-12 2023-12-01 美商史柯比人工智慧股份有限公司 System and method for diagnosing gastrointestinal neoplasm

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI823897B (en) * 2018-02-12 2023-12-01 美商史柯比人工智慧股份有限公司 System and method for diagnosing gastrointestinal neoplasm
TWI790689B (en) * 2021-07-21 2023-01-21 財團法人資訊工業策進會 Method and electric device for processing breast tomosynthesis images

Similar Documents

Publication Publication Date Title
CN109886179B (en) Image segmentation method and system of cervical cell smear based on Mask-RCNN
CN108898595B (en) Construction method and application of positioning model of focus region in chest image
Coburn et al. A multiscale texture analysis procedure for improved forest stand classification
US7483919B2 (en) Object based image retrieval
US8699769B2 (en) Generating artificial hyperspectral images using correlated analysis of co-registered images
US20090082637A1 (en) Multi-modality fusion classifier with integrated non-imaging factors
CA2554162A1 (en) Method and system for intelligent qualitative and quantitative analysis of digital radiography softcopy reading
CN112669254A (en) Deep learning prostate cancer bone metastasis identification system based on whole-body bone scanning image
CN110796659B (en) Target detection result identification method, device, equipment and storage medium
Yang et al. Skull sex estimation based on wavelet transform and Fourier transform
CN115393351B (en) Method and device for judging cornea immune state based on Langerhans cells
US20220374785A1 (en) Machine Learning System
CN105718947B (en) Lung Cancer Images sophisticated category method based on LBP and wavelet moment fusion feature
CN103246888A (en) System and method for diagnosing lung disease by computer
CN112927215A (en) Automatic analysis method for digestive tract biopsy pathological section
Bator et al. Detection of cancerous masses in mammograms by template matching: optimization of template brightness distribution by means of evolutionary algorithm
TW201000074A (en) Pathologic detecting system by using the detection of perspective image of breast and the method thereof
Guo et al. Pathological detection of micro and fuzzy gastric cancer cells based on deep learning
Al-Darabsah et al. Breast cancer diagnosis using machine learning based on statistical and texture features extraction
CN115423806B (en) Breast mass detection method based on multi-scale cross-path feature fusion
DE112019004112T5 (en) SYSTEM AND PROCEDURE FOR ANALYSIS OF MICROSCOPIC IMAGE DATA AND FOR GENERATING A NOTIFIED DATA SET FOR TRAINING THE CLASSIFICATORS
WO2017089953A1 (en) Process for processing medical images of a face for recognition of facial dysmorphisms
CN116402756A (en) X-ray film lung disease screening system integrating multi-level characteristics
CN113989588A (en) Self-learning-based intelligent evaluation system and method for pentagonal drawing test
Sun et al. Automatic diagnosis for prostate cancer using run-length matrix method