定量型雙波段紅外線乳癌診斷系統之研發:乳癌化療反應監控比較研究之熱圖譜特性化與腫瘤邊緣偵測

Abstract

在紅外線影像(Infra-red, IR)中每一個像素點均含有正常與不正常組織所釋放出來的紅外線訊息，藉此建構一套定量型雙波段紅外線乳癌診斷系統(Quantitative Dual-Spectrum IR Breast Imaging System, QDS-IR)。先以無標記點多時間點紅外線影像對位演算法和時間序列熱影像溫度正規化演算法，使不同次拍攝之多時間點影像轉換至相同的座標系統和相似的溫度基準下進行分析。本研究核心之一為開發定量型雙波段熱圖譜分離演算法(Quantitative Dual-Spectrum Heat Pattern Separation algorithm, QDS-HPS algorithm)，此演算法運用盲源分離(Blind source separation，BSS)的概念，先推估出高溫組織區域(qH map)與正常溫度組織區域(qN map)之含量，再經由多次化療後的時間序列影像，定量分析高溫組織隨時間而改變的現象，可藉此更進一步地推估潛藏於高溫組織下可能的癌組織含量變化。本研究以動態對比顯影核磁共振影像(DCEMRI)以及正子斷層掃描(PET)驗證在化療過程中，腫瘤變化和高溫組織區域(qH map) 之結構性和功能性變化的相互關係。初步結果顯示qH map可有效地量化乳癌區域的組織和代謝改變，因此推估QDS-IR系統具有監控和評估化療病患之療效和偵測腫瘤之潛力。為使本系統功能更強大並證明QDS-IR系統的有效性，本研究另一核心為開發電腦輔助診斷系統之超音波影像分割，欲先以此為架構，未來進行3D超音波影像分割，藉以評估腫瘤大小，輔助說明QDS-IR系統所得到的資訊。超音波影像對比度較低且成像過程產生獨特的假影，灰階場不平均造成假影和腫瘤組織形成偽影，導致欲正確地自動分割出腫瘤邊界更加困難。本研究提出一種新的分割演算法，透過偏移場修正，提高超音波影像中腫瘤部分的對比度，期能較準確地判斷腫瘤邊緣。此研究結果顯示經過灰階場修正之影像，對於傳統的等位函數法，亦或本研究開發之Gibbs-weighted K-means分群演算法和模糊區域競爭法，所得到分割結果都較為原本未修正灰階場之影像分割結果來得好。

Breast cancer has become the first cause of death for the female populations in the developed countries. Most modern imaging modalities, e.g., mammogram, breast sonogram, X-ray CT, etc., either do not have a sufficient spatial resolution to detect very small breast cancers or are not sensitive enough to measure the minutia angiogenesis phenomenon in the very earliest stage of breast cancers. MRI and PET are both very costly and not usually used for screening or a first line of defense during the regular checkup. In contrast, a passive medical imaging modality, called quantitative dual-spectrum IR (QDS-IR) breast imaging system is developed. The system roots in two facts: One is cancer cells tend to have higher temperatures than the normal cells because of the angiogenesis. The other is according to the intrinsic Wien's displacement law of the Planck thermal radiation, the radiation in the shorter wavelengths increases much more rapidly than in the longer ones as temperature increases.With energy readings from two IR cameras, middle wavelength IR (MIR, 3-5μm) and long wavelength IR (LIR, 8-9.2μm), IR image decomposes into normal ( map) and high ( map) temperature tissues via the blind source separation (BSS) algorithm. To interpret the information obtained by QDS-IR system, 2D sonogram segmentation methods with intensity inhomogeneity correction algorithm are developed first to further implement 3D segmentation in the future. Based on the visual evaluation of two experienced radiologists, 46 out of 49 breast lesions were considered to have better contrasts on the inhomogeneity-corrected images by both radiologists.The interrater reliability for the radiologists was found to be Kappa = 0.479 (p = 0.001). The mean gradients of the low-gradient boundary points before and after correction of the intensity inhomogeneity were compared by the paired t-test yielding a p-value of 0.000, and the Chan and Vese level set method could derive a much better lesion boundary on the inhomogeneity-corrected image than on the original image (p=0.000). Clinical tests have been carried out with the approval of Institutional Review Board of National Taiwan UniversityHospital. From June 2008 to June 2011, 50 patients of ages between 30 and 68 are recruited. The information of QDS-IR breast imaging system is correlated with DCE-MRI and PET. The results show that the map derived by quantitative dual-spectrum heat pattern separation (QDS-HPS) algorithm indicates not only the structure change but also the function change duringthe chemotherapy and prove that QDS-IR breast imaging system has potential to monitor the chemotherapy response and detect the minutia angiogenesis phenomenon.