使用紅外線熱影像之乳癌分類評估

The Assessment of Breast Cancer Classification by Using Infrared Thermal Images

吳玉洺(Yuh-Ming Wu)

指導教授 ： 蔡育秀

中原大學/工學院/生物醫學工程學系/碩士(2008年)

https://doi.org/10.6840/cycu200900101

摘要

紅外線熱影像可用於評估關節炎、疼痛、心臟血管手術及乳癌輔助診斷等方面之應用。乳癌篩檢為乳癌防治之基本策略，紅外線熱影像如可作為乳癌篩檢之工具，可藉由其低成本、及無輻射傷害的特性造福大眾。在早期之乳房紅外線影像研究，一般均是以影像的模式進行研判，但是以影像模式進行研判之方式較無法擬定一個客觀的標準，近年來亦有一些研究將影像進行數值統計分析及類神經網路之分類應用，本研究擬在探討合併使用過去研究的特徵及新增不同的統計特徵，使用類神經網路及決策樹之分類研究，祈能達到輔助診斷之需求。
本研究蒐集，乳癌診斷為正常者有107名、良性腫瘤者有104名、惡性腫瘤有138名樣本之乳房熱影像及診斷資料。在特徵方面取正面乳房紅外線熱影像，以手動分割之方式區分出兩個橢圓，乳房影像區塊及乳頭位置，進行溫度特徵統計。統計特徵包含最高溫度、平均溫度、左右乳房最高溫度差、平均溫度差、溫度分佈標準差、偏態係數、峰態係數、亂度、變化梯度等17個特徵。將合格的受試者特徵樣本以診斷進行分組，再以各組樣本之三分之二的特徵及診斷進行決策樹及倒傳遞類神經網路訓練及驗證，最後再以剩餘的三分之ㄧ樣本進行最終測試。決策樹測試結果正確率(Accuracy)為 0.546至0.718不等；倒傳遞類神經網路測試結果正確率(Accuracy)為0.587至0.644不等。
由於預測效能結果並不理想，說明在17個溫度統計參數再加以倒傳遞類神經網路或決策樹之分類無法做出有效的分類。檢討主要的原因是在不同診斷族群的特徵值，有著嚴重重疊的情形，致使在分類上使用線性或非線性的分類方法，均無法將不同族群予以有效的分類，而得到令人滿意的結果。本研究以紅外線攝影儀及電腦運算能力的持續進步為基礎進行，以所使用的特徵參數預測乳癌或非乳癌之正確率、靈敏度、特異度，在評估作為單獨適用的診斷工具上並未達到診斷的基本要求，對於正常及腫瘤之熱影像分佈在特徵上之表現，無法明確的切割兩者之差異。此一結果限制它在將來作為乳癌之診斷工具的用途，它只能作為X光乳房攝影及其他診斷工具的輔助，仍無法提作為可靠的篩檢方式，至於合併其他診斷工具使用，對於乳癌診斷之幫助仍有待後續研究之驗證。

關鍵字

紅外線熱影像 ； 乳癌 ； 決策樹 ； 倒傳遞類神經網路

並列摘要

Infrared Thermography can be used as a tool for the evaluation of arthritis, pain modeling, and cardiac surgery. It also can be used for breast cancer evaluation. In the past, making use of thermography to evaluate the condition of beast cancer was mainly done by visual inspection. Besides, nowadays there is an increase in putting temperature statistical features and classified it by neuron network to evaluate the breast cancer by thermography. The study used some unique features from other studies, and classified them to breast cancer positive and negative groups by backpropgation neuron network and decision tree techniques, hopefully to build some classify model for to help the diagnosis of breast cancer.
The study had its beginning in collecting breast thermoghrapy images. At that time, the total cases of the collection were 422. There were 107 cases diagnosed normal,104 cases diagnosed benign, and 138 cases diagnosed malignant. And the rest cases were about bad images, breast mastectomy, and insufficient diagnosis information and had been excluded. In addition, according to the research, the 17 extracted features were based on the different temperature distribution statistics of the two breasts. The classified work was proceeding by backpropagation neuron network and decision tree techniques. The 2/3 eligible samples were separated for training set, and the remaining samples were kept for the final performance test. The decision tree test accuracies are between 0.546 and 0.718. The backprojection neuron network test accuracies are between 0.597 and 0.644.
The study was proceeding on the basis of high speed computer calculation and new infrared camera technology. But the result is still not satisfactory for the breast cancer diagnosis. The main reasons are that the features of normal and malignant tumor are overlapped, and the features of benign tumors are distributed amid normal and malignant groups. These reasons limited the possibility of thermography as a screen tool for breast cancer. However, under the trend of breast cancer multi modality diagnosis strategy, the application of combining thermography with other diagnostic modality still needs for further assessment.

並列關鍵字

Decision tree ； Thermogram ； Breast cancer ； Neuron network

參考文獻

[4] 行政院衛生署, "癌症登記年報,"2004.

[1] Hisayoshi Suma, Tadashi Isomura, Taiko Horii, and Toru Sato, "Intraoperative coronary atery imaging with infrared camera in off-pump CBAG," The Annals of Thoracic Surgery, vol. 70, pp. 1741-1742, 2000.

[2] Preston B.Rich, Georage R.Dulabon, Christtle D.Douillet, and Todd M.et.al, "Infrared thermography: a rapid, portable, and accurate techmique to detect experimental pneumothorax," Journal of Surgical Research, vol. 120, no. 2, pp. 163-170, Jan.2004.

[7] J. R. Keyserlingk, P. D. Ahlgren, E. Yu, N. Belliveau, and M. Yassa, "Functional infrared imaging of the breast," Engineering in Medicine and Biology Magazine, IEEE, vol. 19, no. 3, pp. 30-41, 2000.

[8] Isard HJ, Becker W, and Shilo R, "Breast Thermography after four years and 10000 studies," AJR, vol. 115, pp. 811-821, Aug.1972.