新穎影像重建演算法暨其系統晶片設計應用於連續波式擴散光學斷層掃描系統

Abstract

本論文針對連續波式擴散光學斷層掃描術之影像重建演算法，提出擁有較低運算複雜度的區塊重建模式，並對此新穎的重建演算法進行模擬與精確度分析，最後將之實現於超大型積體電路。 擴散光學斷層掃瞄為一非侵入式醫學造影術，使用光源為650nm-950nm之近紅外光波段的擴散光源，利用不同生物組織對近紅外光光譜不同的分佈與相關資訊，偵測生物體內組織之光學特性與成分濃度在空間與時間的變化或是絕對量，如:吸收係數、擴散係數、含氧血紅素濃度、非含氧血紅素濃度以及血氧飽和度等…。 近年來，擴散光學斷層掃描系統在臨床應用上有，大腦功能性造影、胸部腫瘤檢測以及其他醫學相關領域。此技術有非侵入、及時檢測、非輻射源等優點，其中連續波式系統擁有良好的可攜性、造價低廉與快速檢測等優點，如可將影像重建演算法複雜度降低並實現於系統晶片之中，並配合手持式裝置與無線模組，可將此醫學影像技術更廣泛的使用。 本研究目的分為兩大部分，第一部分著重於不同影像重建模式的模擬與分析，以求在降低運算量與使用有限的資訊來源限制下重建出精確的影像，模擬的模型為兩不同吸收係數的非均勻吸收體位於均勻的吸收體中，在幾何形狀與位置不同的條件情況進行影像重建，並利用Mean Square Error 分析影像的精確度。第二部分，選擇 Jacobi奇異值分解演算法執行影像重建過程中的反向解並將Jacobi奇異值分解演算法實現於電路，應用MATLAB做硬體資源需求的評估與考量，最後使用FPGA做為功能的驗證。 未來，此醫學影像重建引擎可配合其他醫學訊號處理引擎，像是腦波訊號與心電訊號共同實現於系統晶片。使系統縮小化成為醫療用途的可攜式裝置，並應用於緊急救護、醫療檢測儀器、臨床檢測與遠端照護，為人類帶來更多福祉。

This paper proposes a low computational overhead image reconstruction algorithm of continuous wave diffuse optical tomography (CW-DOT) which is called sub-frame reconstructed mode. The novel reconstruction algorithm was simulated and its accuracy analyzed. Moreover, it was implemented on very large scale integrated circuit. DOT is a non-invasive medical imaging technique and uses near infrared optical source with wavelength within 650nm-950nm. By the different distribution of near infrared spectroscope (NIRS) within distinct biological tissue, the spatial and temporal change and absolute values of biological characteristics can be obtained such as absorption coefficient, scattering coefficient, concentration of oxy-hemoglobin and concentration of de- oxy-hemoglobin. Recently, the clinical applications of DOT are in functional image reconstruction of the brain, detection of tumor in breast, and others. Non-invasive, real-time and not a source of radiation are the beneficial features of DOT. Furthermore the CW-DOT systems have small volume, low cost, high speed and other advantages. If the image reconstruction algorithm can be decreased the complex and implemented on SoC, then through cooperation with handheld instrument with wireless module, the DOT system can be more convenient to put in use in real life applications. There are two focuses in this research. The first one aims to reduce the complexity of algorithm and reconstruct accurate image with the finite information by simulating and analyzing the different reconstructed way. Distinct geometry and location of inhomogeneous absorbers were allocated in homogenous absorbers to do simulation. Mean Square Error (MSE) was used to compare the accuracy of images. The second issue is choosing the algorithm to implement. After survey, the Jacobi Singular value decomposition was chosen to realize the inverse solution of reconstruct process and realize on hardware description language (HDL). MATLBA is used to evaluate the hardware resources. Finally, FPGA will be used to do the function identification of the image reconstruction processor. In the future, the image reconstruction processor can be accompanied with other biomedical signal processors such as electroencephalography (EEG) or electrocardiograph (ECG) and be implemented on SoC. The urgent image modality, clinical diagnosis and bed-side monitor, real-time inspection and remote care can be made by minimizing these medical systems to handheld apparatus and bring the more welfare to our human life.