數位乳房X光影像之電腦輔助微鈣化群偵測與分割系統

A Computer-Aided Diagnostic System for Detection and Segmentation of Clustered Microcalcification in Digital Mammograms

楊勝智(Sheng-Chih Yang) ； 李三剛(San-Kan Lee) ； 詹寶珠(Pau-Choo Chung) ； 楊晴雯(Ching-Wen Yang) ； 李覃(Tain Lee) ； 許居誠(Giu-Cheng Hsu) ； 張建禕(Chein-I Chang) ； 羅見順(Chien-Shun Lo)

《中華放射線醫學雜誌》 27卷3期 (2002/06) Pp. 89-101

摘要

乳房X光攝影影像的微鈣化群對於乳癌提供了一個早期的警訊，許多的不可觸知的原位腺管癌及微小癌可由乳房X光攝影的微鈣化得知。微鈣化群的組織微小難以辨認，因此不經過醫師細心的觀察，往往容易忽略而造成病人日後延誤醫療，然而隨著乳癌發生率的增加與衛生教育的普及，乳房X光攝影已經日漸普遍，加上此項診斷需要專業的訓練，使得醫師的人力已不足以應付這逐漸增加的工作負擔。本研究提出發展乳癌電腦輔助診斷系統，協助醫師偵測與分割出微小的鈣化群，減少錯判，提高診斷品質。我們依照分割乳房區域、偵測與分割微鈣化群的步驟，找出有意義的微鈣化群，再使用荷蘭Nijmegen大學附屬醫院的乳房數位影像資料庫作為訓練與測試，根據使用不同的碎形幾何維度容許誤差值所產生的ROC圖，其偵測效能可以達到Az=0.96，再使用本院資料庫的乳房X光攝影影像，比較三位放射線專業醫師對微鈣化群的診斷與本研究系統的測試結果，對於易辨認的影像，診斷的正確率較高（真陽性比率98%），而全部的診斷真陽性比率為86%。
因為本院所使用的數位影像是由X光片掃描所得到，經由人工處理過程容易產生假影影響測試結果，未來使用數位X光乳房攝影設備，可以改進此一缺點，依照此研究成果，將繼續發展微鈣化群的良惡性分析、腫瘤的偵測與分割、腫瘤的良惡性分析等，完成一個完整的電腦輔助診斷系統。

關鍵字

乳房 ； 鈣化 ； 乳房腫瘤 ； 診斷 ； 乳房攝影 ； 比較研究 ； 技術

並列摘要

Clustered microcalcification screened from mammograms provides an early sign of breast cancer. Many impalpable in situ ductal carcinomas and minimal carcinomas can be identified by using X-ray mammography. Generally, microcalcifications are tiny clustered particles and probably smallest structures within the breast, which are difficult to detect. Therefore, microcalcifications are generally overlooked by physicians if they do not carefully screen the mammograms. Consequently, it may cause the delay of medical treatment. So far, X-ray mammography is the only effective screening procedure to detect breast cancer in early stage. However, due to the increasing incidence rates of breast cancer and public awareness, mammography has been also increasingly used by physicians for screening purpose. As a result, a large volume of mammograms will be required to be read by radiologists. Due to shortage of experienced physicians, this tremendous workload creates dilemma to maintain quality of medical diagnosis.
This paper presents a computer-aided diagnostic system for detection of clustered microcalcifications, which can help physicians reduce errors in medical diagnosis while improving the quality of medical service. The proposed system includes three stages. The first stage extracts the breast region from a digital mammogram, and the second stage detects the suspicious area in the extracted breast region. Finally, the last stage segments microcalcifications from the suspicious area. In order to evaluate the designed system, a preliminary study was conducted using the public Nijmegen database provided by the Department of Radiology at the Nijmegen University Hospital, Netherlands. According to the different tolerance of Fractal. The experimental results show that the AZ of ROC distribution can achieve 0.96. When the database of TCVGH is used, three categories of mammograms (Obvious, Possibly neglected and Difficult to be identified) were studied according three radiologists' reports. In obvious cases, the rate of true positive could achieve as high as 98 %. For all cases, the rate of true positive could also achieve 86%. The results of this paper will provide a further development of mass detection used in computer-aided diagnosis system.

並列關鍵字

Breast ； calcification ； Breast neoplasms ； diagnosis ； Breast nadiography comparative studies ； Breast radiography technology