全域數位乳房攝影影像之腫塊自動偵測系統

Automatic Masses Detection System of Full-Field Digital Mammography Image

饒至剛(Chi-Kung Jao)

指導教授 ： 蘇振隆

中原大學/工學院/生物醫學工程學系/碩士(2009年)

https://doi.org/10.6840/cycu200900371

摘要

隨著國人癌症比率逐年升高，乳癌已成女性的頭號殺手。如何早期偵測與診斷已成為治療癌症之重要手段。除了觸診檢查，乳房X光攝影為目前應用最廣泛的乳房檢測工具。隨著數位乳房攝影系統之使用，本研究參考國內外腫塊偵測之相關研究，建立一套全域數位乳房攝影(FFDM)影像之自動腫塊偵測系統。
本研究發展之腫塊偵測系統，以數位式乳房攝影之原始影像為輸入影像。使用向量域收斂濾波器及分割演算法，找出腫塊的位置以及形狀，並擷取腫塊圈選後之特徵參數(包括形狀、邊緣及密度上的9個特徵參數)，以輔助區分是否為腫塊或是偽腫塊。另外，為了評估本系統的好壞，本研究以FROC曲線來表示系統的效能。而本研究所使用之乳房影像有19張，每張影像皆有1個腫塊，並透過專業醫師確認其所在位置。
在研究成果上，本系統提供友善化之介面，可直接輸入DICOM格式之全域數位式乳房攝影影像。19張影像在經影像處理與統計分析後，當系統偵測靈敏度為84.2%、73.7%；其平均每張影像的偽腫塊數量相對為2.36個、0.315個。另外，若就病例區分，每一個帶有腫塊的病例，本研究皆可偵測到正確腫塊位置在一張影像以上，即100%的偵測靈敏度。在腫塊位置上，本研究所找出之腫塊位置與醫師所圈選之腫塊位置平均重疊面積可達63.7%。
本研究運用影像處理的技術已發展一套全域數位式乳房攝影影像之腫塊自動偵測系統，能夠找出腫塊概略的位置，並有較低的誤判個數，因此可輔助醫師對於乳房在腫塊上的診斷，減少未找到腫塊的機率。初步已獲臨床醫師之肯定。未來若經過更多資料的測試及程式的驗證後，可以成為開發電腦輔助診斷系統之基礎。

關鍵字

全域數位式乳房攝影影像 ； 向量域收斂濾波器 ； 區域成長 ； 電腦輔助偵測系統 ； 乳房腫塊

並列摘要

Recently, the rate of patient with cancer increasing and the breast cancer has already become women's No.1 killer in Taiwan. How to detect and diagnose in early stage has become an important task for the cancer treatment. At present, except that palpation, the mammography applies the most extensive breast measure tools. With the use of the digital mammography, an Automatic Masses Detection System of Full-Field Digital Mammography (FFDM) Image was developed.

In this study the raw image of digital mammography was used as the input image, and the Convergence Index Filter for Vector Field and division algorithm were used as detection method to find out the positions and shape of the masses. After the masses selected, the characteristic parameters which include 9 parameters on the shape, edge and density were then chosen as an auxiliary tool to differentiate whether for masses perhaps false mass. In addition, in order to evaluation the efficiency of the system, we show efficiency of system as Free-Receiver Operating Characteristic (FROC) curve. And totally 19 raw images are used in this study, and each image all has a mass and confirm its position through the professional doctor.

On the results of this study, this system offers a friendly interface, and it can direct input FFDM image with DICOM format. After the image processing and the statistical analysis, as system sensitivity goes to 84.2 % and 73.7 %, there are 2.36 and 0.315 FP/Image, respectively. However, the sensitivity of this system nearly 100% as if a patient has both MLO and GC view images. Moreover, the average overlap area of mass detected by system and selected by radiologist can be possible to reach 63.7%.
Automatic Masses Detection System of Full-Field Digital Mammography Image has been developed which use the technology of the image processing, it can find out the position of the masses, and it has lower FP/Image. Therefore may assist doctor regarding breast on mass diagnosis, to reduce the probability not finding the mass. This system has already preliminarily obtained the clinician's affirmation. It can become a foundation for developing the computer aid diagnoses system after testing more data and verification of the program in the future.

並列關鍵字

Full Field digital mammography image ； Convergence Index Filter for Vector Field ； computer Aid detection system ； Breast masses ； region growing

參考文獻

[2] Lihua Li, Robert A. Clark, MD, Jerry A. Thomas, MSc, Computer-aided

Diagnosis of Masses with Full-Field Digital Mammography, Academic

for directional feature extraction in medical Image,IEEE International

Conference on Image Processing. Santa Barbara, 500-503

[5] Nicholas Petrick, Heang-Ping Chan, Berkman Sahiner, Mark A. Helvie,