雙波段紅外線乳癌診斷系統之研製與評估

Abstract

乳癌已連續八年居於台灣女性惡性腫瘤死因第四位，惟有早期發現早期治療能提升治癒率。目前常用的乳癌篩檢工具為乳房X光攝影及乳房超音波影像，然而這兩種方法有空間解析度與敏感度不足的問題，無法偵測極早期的乳癌；雖然乳房磁振造影與正子斷層掃描都具有早期偵測乳癌的潛力，但單次檢測價格昂貴使其無法用於第一線的乳癌早期偵測。因此本研究發展了雙波段紅外線影像（DS-IR spectrogram），希望評估此被動式醫學影像技術用於偵測乳癌的效能。 DS-IR spectrogram是由美國海軍研究院斯華齡教授所提出的嶄新概念，其獨特能力來自一突破性的想法：以紅外線光譜比值（DS-IR spectral ratio）作為區分癌細胞與正常細胞的指標。由於癌細胞生長時多半會有血管增生現象，使其溫度高於正常細胞；根據普朗克輻射定律與維恩位移定律，溫度上升時，光譜高峰會由長波往中波移動，且中波紅外線（MIR，波長為3-5μm）的增加量會比長波紅外線（LIR，波長為8-12μm）要來得多，故只要癌細胞釋放出MIR跟LIR照相機可以偵測到的MIR跟LIR光子，就能藉由DS-IR spectral ratio與時俱增的傾向作為乳癌特徵。由此可知DS-IR spectrogram並非單憑體表溫度的改變去決定病灶位置，能提供比傳統紅外線熱影像更可靠的資訊。 獲得DS-IR spectrogram的主要步驟如下：首先是不同波段的紅外線影像對位，本研究使用了線性的仿射轉換跟非線性的薄板仿樣法進行對位。其次，根據對位後的MIR與LIR影像，以核心演算法（Deterministic neighborhood-based BSS algorithm）計算出每一像素隱含的惡性腫瘤比例，藉此偵測乳癌。 為了評估DS-IR spectrogram偵測乳癌的效能，本研究使用一台MIR照相機（波長為3-5μm）跟一台LIR照相機（波長為8-9.2μm）建構出第一套雙波段紅外線乳房攝影系統，自2007年7月起於台大醫院公館院區進行大型臨床試驗，並從招收到的樣本中選出100個進行分析。實驗結果顯示DS-IR spectrogram具有偵測乳癌的潛力，惟需更清楚掌握其隱含資訊，未來將有機會以此被動式的醫學影像系統偵測乳癌。

Breast cancer has been ranked the fourth leading cause of cancer deaths for females in Taiwan for the past eight years. Common imaging modalities used to detect breast cancers at present are mammogram and breast sonography. However, they both do not have a sufficient spatial resolution and are not sensitive enough to detect breast cancers. Although the MRI and the PET have the potential for early detection of breast cancers, they are both very costly and not usually used as the first-line defense. Therefore, this thesis proposes to develop a passive medical imaging modality, called dual-spectrum IR (DS-IR) spectrogram, for detection of malign tumors in breast. The DS-IR spectrogram was invented by Dr. Harold Szu. The unique capability of the DS-IR spectrogram comes from using spectral ratio as the signature to distinguish cancer cells from normal ones. Since the cancer cells tend to have higher temperatures than the normal cells because of the angiogenesis process involved in the growth of cancer cells, the spectrum peak up-shifts from long wavelength infrared toward medium wavelength infrared due to Planck radiation physics and Wien’s displacement law. As long as there were a few amount of MIR photons emitting from the malign tumor cells and detected by the dual IR cameras, one may identify the malign tumor cells by using the trend of DS-IR spectral ratio over time as the feature. Two essential steps are involved in deriving the DS-IR spectrograms. One is registration of the MIR and LIR images accomplished by using linear affine transformation and nonlinear thin-plate spline. The other step is detection of the breast cancers based on the registered MIR and LIR images. Deterministic neighborhood-based BSS algorithm proposed by Dr. Harold Szu is realized to derive the probability distributions of cancer cells and normal cells. To evaluate the performance of the DS-IR imaging system, we have built the first clinical DS-IR imaging system using one MIR(3~5