請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/92184
標題: | 單時間點雙波段紅外線乳癌偵測系統 A Dual-Spectrum Infrared System for Detection of Breast Cancer Based on Single Visit |
作者: | 林若真 Ruo-Jhen Lin |
指導教授: | 陳中明 Chung-Ming Chen |
關鍵字: | 乳癌,影像對位,雙波段紅外線影像,薄板仿樣法,血管提取,支持向量機,邏輯回歸,ROC曲線, breast cancer,Thin-plate spline,Dual-Spectrum infrared image,Hessian Filter,Support Vector Machine,ROC curve, |
出版年 : | 2016 |
學位: | 碩士 |
摘要: | 根據世界衛生組織(WHO)統計,女性中最常見的癌症之一是乳癌且其發生率有逐年增加的趨勢,目前仍沒有可以完全預防或治療的方法,只有早期發現並診斷且接受治療才能降低乳癌的威脅。紅外線攝影具有無放射性、無侵入性、成像快速並可重複成像等優點且研究顯示紅外線影像可以觀察出癌細胞之表現,過去有許多利用傳統紅外線影像來偵測乳癌的研究,在嚴謹的拍攝條件下取得的紅外線影像並直接分析,但由於傳統紅外線影像的變異性大,容易受到外在環境或是生理因素的影響,因此只利用溫度來診斷是較不足的。
本研究提出一套單時間點雙波段乳癌偵測系統,受測者是到門診就醫的病患,並且排除已完成治療或治療中的病患,受測者會同時拍攝中波及長波紅外線影像並將兩張影像利用薄板仿樣法(Thin-plate spline)對位,再利用Hessian Filter找出影像中的血管結構,接著透過雙波段熱圖譜分離(Dual-Spectrum Heat Pattern Separation, DS-HPS)演算法計算出每個pixel點的高溫組織含量(Nq_H map)與常溫組織含量(Nq_N map),最後利用癌細胞的血管增生狀態明顯且部分血管溫度會較高的特性,找出合適的特徵值透過支持向量基(Support Vector Machine)得到上述特徵所得到的分類結果,並與受測者現有的檢查結果相互比較,來評估此套系統的診斷效果。 結果顯示,透過單時間點雙波段紅外線乳癌偵測系統的AUC為90.8%,最佳正確為86.3%,在最佳正確率下的敏感度與特異度分別為77%和94.8%;然而在特異度為85%的情況下,正確率為83.2%而敏感度為81.2%;當敏感度為85%時,其分類正確率為82.9%,特異度為81.4%,透過上述不同情況下其評估的數據顯示,本研究所提出的單時間點雙波段紅外線乳癌偵測系統對於偵測乳癌有良好的效果。 According to the statistic from the World Health Organization, breast cancer is one of the most common cancer in women and its incidence ration is increasing year by year. However, there is no methods of prevention or treatment completely. The only two ways to reduce the threat are as follows detecting breast cancer early and monitoring the effect of treatment. Infrared technology has such advantages as: non-radioactive, non-invasive, fast and repeating imaging. And there are many studies have shown that the infrared images can observe the representation of the cancer cells. There are many studies of detecting breast cancer by using infrared image have been proposed. These researches take the infrared images under stringent conditions and then analyzed these images without any processing. The variability in the traditional infrared image is large. It was affected easily by external environmental or physiological factors. Therefore, only used temperature as feature to diagnosis breast cancer is insufficient. This study proposed a Dual-Spectrum Infrared System for Diagnosis of Breast Cancer Based on Single Visit. The subjects in this study were the outpatient without any treatments. First, the subjects were captured both the LIR and MIR information by the infrared cameras. Then it used Thin-plate spline for image registration and the corresponding position was found. In order to find the vascular structures on the infrared image, it used Hessian Filter to achieve this purpose. The fourth step is to calculate the tissue area of high temperature (Nq_H map) and normal temperature (Nq_N map) in each pixel by Dual-Spectrum Heat Pattern Separation (DS-HPS) algorithm. According to the angiogenesis of cancer cells was significant and the vessels of cancer cells had high temperature to find the right feature. Finally it can get the classification results by using Support Vector Machine (SVM). Compared the results with other way diagnosing breast cancer to evaluate the effect of this diagnostic systems. The results of using Dual-Spectrum Infrared System for Diagnosis of Breast Cancer Based on Single Visit show that: the AUC was 90.8%, the best accuracy 86.3% as for sensitivity and specificity rate is 77% and 94.8%; when specificity was 85%, the accuracy was 83.2% and sensitivity was 81.2%; on the other hand sensitivity was 85%, the accuracy was 82.9% and specificity was 81.4%. Through these different assessments show that Dual-Spectrum Infrared System for Diagnosis of Breast Cancer Based on Single Visit for detecting breast cancer has a good effect. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/92184 |
DOI: | 10.6342/NTU201603183 |
全文授權: | 未授權 |
顯示於系所單位: | 醫學工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-104-2.pdf 目前未授權公開取用 | 2.25 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。