以類神經網路建立乳房切開式切片結果分類之輔助預測模組

乳癌自民國八十五衛生署所公佈的資料中已超越子宮頸癌成為女性死亡率的第一位，在民國九十七年資料中罹患乳癌的女性有8,136人，而該年因乳癌死亡人數有1,654人，所以乳癌的早期診斷與治療是國內醫學界目前重要議題之一。有鑒於此，本研究以台北市某醫學中心之乳房攝影分類為BI-RADS 4 A乳癌病患資料，收集其相關乳癌致病危險因子與乳房X光攝影結果上之特徵，希望可以透過類神經網路能仿效生物神經網路可以分析非線性資料、不斷訓練和不斷學習的特性，與邏輯斯迴歸此種統計上的迴歸分析模型，這兩種模式建構預測模型，比較其預測的績效。最後將以此預測模型實際應用於臨床的切開式切片的結果預測，使臨床在診斷的準確度上能有所提升，達到提高在BI-RADS 4 A此種分類下，切開式切片的陽性預測率(Positive predict value rate)。如此不僅能降低病人生理與心理上的傷害，也相對的在醫療資源的妥善運用上也有所裨益。本研究共蒐集280筆資料中，使用乳癌致病危險因子與乳房X光攝影結果上之特徵，包括:年齡、初經年齡、哺餵母乳、荷爾蒙使用、Coarse Heterogeneous、Plemorphic、Fine linear Branching、超音波下鈣化點的有無，共8個變數，所建構的邏輯斯迴歸模型的正確度為87.0％、敏感度為50.0％、特異度為96.2％、AUC為0.731。類神經網路模型，經輸入驗證組的資料後，在切開式切片結果的預測效能方面，得到的正確度為97.9％、敏感度為97.0％、特異度為100.0％、AUC為0.972。相較之下類神經網路模型較邏輯斯迴歸模型有較好的預測效能。

關鍵字

類神經網路；乳癌；切開式切片

並列摘要

According to the report published by Department of Health in 1996, the breast cancer is beyond the cervical cancer to become the first female mortality. In 2008, there are 8136 females suffering from breast cancer and 1654 of them died in the same year. Regarding to this significant number of death, the research of breast cancer becomes one of the important topics in the medical profession. The data, which contains the result of mammography with some risk factors, in category BI-RADS 4A collected from one of the medical research center is used for training Artificial Neural Network module that can emulate biological neural network in the characteristics of analysing non-linear data, continuous training and learning. The result of this prediction module will be used to compare with the result from logistic regression analysis model in the aspect of performance of prediction. Finally, this module will be applied to predict the result of clinical excision biopsy, in order to increase the accuracy and reduce the positive predict value rate in the classification of BI-RAD 4A. This research can not only avoid of psychological and physical harm but help the using of medical resource more efficiently. This study were collected in the 280 cases in the result of the characteristic of breast cancer risk factor and mammography feature which includes a total of eight variables as ages、menarche age、breast-feeding、hormone use、Coarse heterogeneous、Plemorphic、Fine linear Branching、Ultra-sound visible calcification。The accuracy of logistic regression model is constructed of 87.0%，sensitivity was 50%，specificity was 96.2%，AUC 0.731 。After inputting the data of the validation group in the neural network model，the accuracy of predicting performance in the excision biopsy results，obtained for 97.9%，sensitivity was 97% and a specificity of 100.0%，AUC 0.972。Compare to the neural network model than the logistic regression model has better prediction performance。

並列關鍵字

artificial neural network ； breast cancer ； excision biopsy

參考文獻

癌症登記年度報告（97年版）【資料檔】。台北市：行政院衛生署國民健康局

郭守仁, 陳守棟, 張正雄, 劉幕台, 陽光道, 葉坤土, et al. (2003). 乳房醫學. 彰化市: 彰化基督教醫院.

Lipnick, R. J., Buring, J. E., Hennekens, C. H., Rosner, B., Willett, W., Bain, C., et al. (1986). Oral contraceptives and breast cancer. A prospective cohort study. [Research Support, U.S. Gov't, P.H.S.]. JAMA : the journal of the American Medical Association, 255(1), 58-61.

Ayer, T., Alagoz, O., Chhatwal, J., Shavlik, J. W., Kahn, C. E., Jr., & Burnside, E. S. (2010). Breast cancer risk estimation with artificial neural networks revisited: discrimination and calibration. [Evaluation Studies

Ayer, T., Chhatwal, J., Alagoz, O., Kahn, C. E., Jr., Woods, R. W., & Burnside, E. S. (2010). Informatics in radiology: comparison of logistic regression and artificial neural network models in breast cancer risk estimation. [Comparative Study

國際替代計量

以類神經網路建立乳房切開式切片結果分類之輔助預測模組

全文下載

主題瀏覽