針對小型動物磁粒子造影系統的磁源定位研究

Abstract

磁性奈米粒子（MNPs）已經被廣泛用於生物醫學領域，它可作為藥物治療的載體或是用於生物感測器的追踪介質。由於它們的低毒性和多功能性，它們可以注射到生物體中並追踪病灶。因此，了解磁性奈米粒子的位置資訊對於其在生醫應用上是很重要的。磁粒子造影（MPI）系統是一種可以追踪磁粒子位置的成像系統。 MPI系統在磁性奈米粒子（MNPs）上施加外部磁場，而磁性奈米粒子的感應磁場可以透過系統的感測器檢測到。然而，MPI系統所得到的影像是磁性奈米粒子的感應磁場空間分布圖，無法直接被獲得實際的磁性奈米粒子分布位置。本工作研究藉由MPI 系統追蹤小動物體內磁性奈米粒子的反演算問題。 首先透過模擬的方式以優化 MPI 系統的運行。建立MPI系統的正向模型和逆向模型，並用線性Bregman迭代算法求解小動物磁粒子影像的磁源分佈問題。模擬中使用的磁源設置為高斯分佈源並放置在小鼠大腦的上側。 MPI 感測器在固定激發磁場下與小鼠大腦上方進行平面掃描。由遠離鼠腦的多平面掃描組合計算得出的磁源與最近掃描平面計算得出的磁源相較於設置的磁源的平均偏差分別為2.78 × 10-3和2.84 × 10-3。多平面掃描和最近平面掃描結果的差異非常小，額外的掃描平面幾乎不會提高計算磁源的準確性。為了改善此結果，梯度磁場被用來掃描小鼠的大腦。梯度磁場的梯度會由它們引導場矩陣的一階微分決定，而所得到的梯度磁場將會使引導場矩陣在不同位置有最大值。當模擬磁源設置在鼠腦底部，梯度場掃描方法結果與固定直流磁場結果的平均偏差分別為4.42 × 10-2和5.05 × 10-2，而位置誤差為2.24 × 10-1 cm和3.61 × 10-1 cm。在模擬中，梯度場掃描方法比固定直流磁場具有更好的結果，尤其是檢測較深遠離感測器磁源的時候。 同時，我們也進行實際的小鼠實驗。藉由MPI追蹤溯源小鼠體內被注入磁奈米粒子的位置。分析由MPI所量測的小鼠體內磁奈米粒子的磁場分布圖。 MPI影像會先與MRI影像整合定位，然後透過反演算模型來計算磁源在小鼠體內得分布。所計算得出的磁源的位置與實驗時注入磁性奈米粒子的位置和MRI中所顯示的粒子位置皆為一致。此外，我們還建立了一個圖形使用者介面，用於處理MPI的數據。GUI整合了MPI分析時所用到的諸多功能，降低了處理MPI數據的複雜度。透過所建立的GUI，我們成功處理 MPI 圖像並追蹤磁奈米粒子的位置。

Magnetic nanoparticles (MNPs) have been widely used in biomedicine as the drug carrier for treatment or the tracer for biosensing. Due to their low toxicity and multi-functionality, they can be injected into the organism and track the lesion. So, it is essential to know the location of the particles for bio application. The magnetic particle imaging (MPI) system is an imaging system that can track the location of the magnetic particles. MPI system applies an external magnetic field on the magnetic nanoparticles (MNPs), and the sensors of the MPI system can detect the induced magnetic field of the MNPs. However, the information MPI obtained is the magnetic field map of the MNPs. The location of the MNPs cannot be obtained directly. The inverse problem for the location of the MNPs inside small animals tracking by the MPI system is studied in this work. The simulation work has first been done to optimize the MPI system’s operation. The forward and inverse models were built. The source distributions of the magnetic particles are inversely calculated from the field maps measured by our small animal's Magnetic Particle Imaging (MPI) system. And this kind of inverse problem is solved by the linearized Bregman iterative algorithm. The source used in the simulation was set up as a Gaussian distribution source and placed at the upper side of the mouse brain. MPI sensors were plane scanning above the mouse brain with a constant external magnetic field. Compared to the set-up source, the average deviation of the calculated source from the combination of multiplane scanning away from the mouse brain and the closest scanning plane are 2.78 × 10-3 and 2.84 × 10-3, respectively. The difference between the result of the multiplane scanning and the closest plane scanning is very small. The accuracy of the source is hardly improved with extra plane scanning. The gradient magnetic fields are applied to improve the result. The gradients of the gradient magnetic fields are decided by their first differential of the lead field matrixes. The gradient fields can make lead field matrixes have a maximum at different positions. With the simulation source setting at the bottom of the mouse brain, the average deviations of the results from the method based on multiple scans with gradient fields and the constant field are 4.42 × 10-2 and 5.05 × 10-2. And the location errors are 2.24 × 10-1 cm and 3.61 × 10-1 cm. In the simulation, the gradient scan method has a better result than the constant field, especially with the deeper source away from the sensors. Besides, the actual experiments tracking the MNPs in the sides mouse are conducted. The field maps of the MNPs inside the real mouse obtained from MPI have been analyzed. The images of MPI were first registered with the MRI images, and the source distribution then been calculated with the inverse model. The results showed that the location of the calculated sources consisted with the location where the MNPs were injected into. A graphical user interface GUI was also established. The functions required for processing the MPI are integrated into GUI. The complexity of the image processing is reduced. With the GUI established, MPI images can be successfully processed and locate the location of the MNPs.