- 學位論文
應用於食道癌診斷之吸收光譜檢測系統
Absorption-spectrum Measurement System for the Diagnosis of Esophageal Cancer
摘要
本研究目的為評估以反射式和透射式可見光吸收光譜的特徵，來分辨食道檢體為食道癌或正常組織的方法是否可行。由台大醫院提供來自 16 位病人的共 26 個檢體中，取了 350 個透射式光譜與 349 個反射式光譜；分別於不同的可見光譜波段中，以相關性標準差演算法搭配「癌症可能性」為診斷標準、最小相關性標準差演算法以及癌症平均光譜演算法搭配「與標準癌症光譜的相關係數」為診斷標準，計算相對各種診斷標準下的系統效能，也以此得出 ROC 曲線。最後計算與最佳效能表現所對應的 ROC 空間中與理想點之最短距離，並記錄所對應的波段以及用以定義陽性、陰性的診斷標準。計算後發現以相關性標準差演算法有較佳的診斷效能，其中又以反射式的效能更為突出；在光譜波段 549.63- 595.8 nm 之最佳效能為: 敏感度 0.71，特異率 0.73，陽性預測值 0.84，陰性預測值 0.57；這代表以可見光吸收光譜之特徵來分辨食道癌與否確實有可行的潛力。不過必需一提的是，實驗中訊雜比過差的光譜，以及吸收太弱的光譜並未完全從原始檔案中分別並去除，而此些光譜都是無效的，所以這都將導致分析的結果增加誤差。
並列摘要
The purpose of this study is to assess the application of visible light for transmission and reflection absorption-spectrum features in the ability to distinguish esophageal cancer. In 26 samples out of 16 patients National Taiwan University Hospital, we gain 350 transmittance spectrums and 349 reflection spectrums. In different wave bands, by using Standard deviation of correlation algorithm with the “possibility of cancer” as standard of diagnosis, Minimum standard deviation of correlation algorithm with “correlation coefficient with standard cancer spectrum” as standard of diagnosis, and Cancer average spectral algorithm with “correlation coefficient with standard cancer spectrum” as standard of diagnosis, calculate the system efficacy with each standard of diagnosis to determinate ROC curves. After all, the minimum distance of the ideal point which corresponds to the best system performance will be obtained, and then record the standard of diagnosis and the wave bands. The Standard deviation of correlation algorithm has the better performance. Among them, the reflective type outstands the other diagnostic performances; the best performance locates in 549.63- 595.8 nm : sensitivity 0.71, specificity 0.73, positive predictive value 0.84, negative predictive value 0.57; this expresses that using the features of absorption spectrums is potential to diagnose esophageal cancers. But what must be mentioned is some useless data has poor signal to noise ratio and low absorption which are not deleted from the analysis, and these will cause more errors in the analytic results.