對於乳癌而言,微鈣化點是可以提供有意義判斷的資訊,分佈或型態都是能夠當成良惡性辨別的準則。乳房攝影即利用鈣化點與乳房組織在X光攝影下,所產生不同的效應而成像。然而這些資訊有可能因為判讀者的經驗或主觀看法,產生差異的結果,尤其是在非典型的例子上。因此本研究的目的在於提出一套自動化的電腦輔助系統,藉由影像上所呈現的特徵分析,提供客觀的數據,而希望能夠協助醫師於判斷病人的狀況。 由於微鈣化點在整體背景上是一個特殊的位置,本研究中我們將結合不同的小波轉換結果來尋找影像的微鈣化點,並藉由設定門檻以及不同角度的影像對應關係來消除誤判雜訊。在現實的情況下,病人接受立體攝影系統將有可能會有偏移的產生,我們將會建立判斷式去消除移動影響並還原回原始空間位置,而建立出三維座標的鈣化點分佈空間位置,之後針對這些空間對應的資訊,找出群聚的鈣化點。 最後藉由前饋選擇來分析不同種類的特徵,依據這些特徵利用羅吉斯迴歸來分類影像資料與良惡性的判定,而希望達到電腦輔助診斷的目標。 本實驗所擁有的資料來源是由台大醫院影像醫學部提供BI-RAID 4 level cases,利用Senographe DS替病人於切片檢查過程所拍攝,從最初的定位影像,到最後切片結束,每位病人將拍攝數十不等的DICOM影像。而本研究將以最初定位的三角度影像資料來偵測鈣化點以及後續分析動作。 分析結果顯示,當增加了深度資訊的特徵後,對於電腦輔助診斷效能是有些微提升。
The identification of micro-calcifications can be useful in the effective detection of breast cancer. Their various location, shape, and size can reveal malignant cancers from benign cysts. The procedure of mammography, a specific type of X-ray radiograph, can digitally capture the contrasting imageries of both micro-calcifications as well as normal breast tissues. However, different examiners of the same digitized images may arrive at divergent diagnoses due to the varying experience and background of each examiner as well as his/her own subjectivity. Especially the mammographic films are atypical cases. The present thesis, therefore, endeavors to introduce an automatic computer-aided diagnosis (CAD) that will provide additional, more objective data based on certain characteristics within the images from mammography. It is hoped that this data will ultimately assist physicians in their diagnosis of breast diseases in a more objective way. Since micro-calcifications, on mammograms, show up as bright spots against the predominantly black background, we shall combine different types of wavelet transforms as our methodology to identify them in our study. Subsequently we shall devise two primary methods to eliminate false traits of micro-calcification: 1). by establishing thresholds within the computer-aid system, and 2). by examining the images from different angles. In reality, patients might exhibit slight movements during the stereo imaging procedure. In our study a formula will be established to normalize discrepancies resulting from these extraneous patient movements. The normalized results will then be used to derive three-dimensional coordinates. These coordinates will then ultimately be used to show clusters of micro-calcification. Finally, the specific characteristics of the micro-calcifications (shape, pattern, and distance) can be revealed through the process of forward select. These characteristics are then subjected to logistic regression function to aid in the useful interpretation of visual data, and to ultimately help in determining whether the cancer is malign or benign. The data contained in this present study are provided by the NTU Hospital-- BI-RAID 4 level cases--. The data is photographed from biopsy of patients using the Senographe DS. The present study analyzes the visual data from the three initially fixed angles/positions.