近年，廣泛用於遙測影像處理中的線性頻譜混合分解法被應用於屬於多頻譜影像之磁振造影像之分析。線性頻譜混合分解法是一個常用於高頻譜影像(例如遙測影像)中次像素目標物偵測與物質分類之技術。由於在遙測影像中感興趣之目標物尺寸一般都小於地面解析度，此時則必須使用次像素目標偵測法來偵測遙測影像中之目標物，否則目標物太小難以辨識。然而在過去將線性頻譜混合分解法應用於磁振造影像的實驗中，皆是利用電腦模擬出來的腦部影像以及真實的腦部磁振造影成像來做分析，雖然電腦模擬之影像在腦部組織量化的數據上有真實解，但電腦模擬之影像存在沒有達到次像素成像的問題，而真實腦部磁振造影像則因為操作上無法實際將人體大腦取出來做組織的量化，所以影像量化的結果就必須經由專業放射部醫師來判斷結果的正確與否，因此線性頻譜混合分解方法是否能達到次像素的偵測和量化尚未得到證明，故本研究實際製作了假體，取得其經由磁振造影儀器掃描後的次像素成像，藉此來驗證線性頻譜混合分解方法對於次像素目標物偵測和量化的能力。透過與假體的含量真實解比較可以發現，線性頻譜混合分解方法確實能計算出不同組織成份的含量百分比，驗證了其能改善傳統影像處理方法無法達到次像素偵測和量化的缺點，進而有效量化出腦部中的主要組織。 The technique of Linear Spectral Unmixing, which is widely used in remote sensing, has recently been applied to analyze MRI images. Linear Spectral Unmixing is a technique normally used in Hyperspectral images to sense target sub-pixels and classify materials accordingly. In remote sensing, the dimensions of an object of interest are usually smaller than ground resolution. Due to an increased difficulty in identifying objects of a smaller dimension, subpixel object sensing can thus be used to recognize an object. The computer simulation brain images and real brain MRI scans are usually analyzed in the experiments of applying Linear Spectral Unmixing in MRI. Although there exists ground truth on the computer simulation data of brain tissues, an image cannot be created by subpixels in computer simulation images. It is impractical to quantize a real human brain from MRI; as such, the quantization process is completed by a professional physician. Thus, it is not yet verified that whether the method of Linear Spectral Unmixing achieves the sense of subpixels. In this research, a phantom has been made and its subpixels from MRI obtained to verify the sense and quantization capability of subpixel objects by the approach of Linear Spectral Unmixing.Compared with the real ratio in the phantom, the method of Linear Spectral Unmixing is able to obtain the percentages of different tissues. It verifies that the approach is able to resolve the defect associated with traditional image processing methods – that is, an inability to sense subpixel – and result in the effective quantization of main brain tissues.
