靜息態功能性磁振造影之轉譯技術開發：從影像品質檢驗到術前大腦功能定位

Abstract

自 1995 年首度發現大腦網路的自發性同步現象以來，靜息態功能性磁振造影（Resting-state fMRI, rs-fMRI）技術日益受到神經科學以及臨床神經醫學研究的重視。因其非侵入性且無需外在刺激的實驗設計與簡單指令，該技術已被廣泛運用於兒童發育、老化歷程、神經退化以及精神疾病等群組研究。然而rs-fMRI技術並不僅侷限於群組研究，隨著轉譯醫學與個人化醫療技術的快速進展，rs-fMRI亦逐步受到關注，現正興起一波以fMRI進行精準術前大腦功能定位之浪潮。然而，在真正實現以rs-fMRI進行個人化評估之前，仍需面對下列三項困境：（一）評估指標：該技術迄今尚未發展出有效的檢驗指標，難以評估其資料品質；（二）種子點選擇：就分析方法而言，經常使用的種子點相關性分析（seed-based correlation analysis）需要主觀設定一種子點，以分析大腦在休息狀態下的網路連結，並定位個人化之功能網路。 隨探究的大腦網路愈趨向高階特化功能（如語言網路），受試者間的功能網路位置變異就越大，造成種子點選定的困難性；（三）分析平台：儘管rs-fMRI技術已蔚為潮流，目前仍欠缺以臨床應用導向設計的分析平台。 本論文的主旨在針對rs-fMRI資料、種子分析技術做進一步的品質檢驗指標設計與方法改良，並建構以臨床醫事人員為導向之fMRI技術分析平台，期望提供簡易操作的使用者介面與可靠的技術以進行臨床相關研究與應用。本論文共具三項主要目標，分列如下: (一) 本論文第一目標為發展一rs-fMRI影像品質指標（PICSO），估測rs-fMRI資料中蘊含的神經生理訊息承載量。結果顯示功能性連結的強弱與PICSO指標呈現正相關，反之常用快篩型指標－時序信雜比(tSNR)－和大腦功能性連結之間並無明顯關聯。 (二) 準確設定種子點為定位高階功能網路之首要目標，然而實作上存在諸多困難。本論文將以語言功能為例，提出一新式自動化選擇法－結合靜息態區域同質性（regional homogeneity, RH）與統合分析技術（Meta-analysis, MA）－利用RH+MA方法導航種子點的設定，增進術前語言功能定位之精準度。 (三) 倘若病灶已影響腦血管的健康狀態，則直接影響以健康血管為前提假設的fMRI訊號。為評估腦血管對fMRI的影響程度，近期研究建議在以fMRI進行大腦功能區域定位時，同時比對反應腦血管健康狀態的順應力圖譜（CVR mapping），避免誤判大腦重要功能區的範圍。故本論文的第三目標為發展一平台，整併與模組化任務型fMRI（task-fMRI）, rs-fMRI與CVR mapping，並將結果轉換至與臨床閱片系統與手術導航系統相容之DICOM格式，便於醫師進行完整的術前評估與術中導航之用。 本論文的研究目標為轉譯功能性磁振造影分析技術於臨床應用。於技術上，設計實用型的影像品質檢驗指標、發展設定種子點之導航方法；於臨床上，整合多項技術於單一平台，得同時提供影像分析介面、視覺化呈現與直接整併至手術導航系統。在未來發展中，本論文將引入深度學習演算法，自動化解構靜息態功能性網路。這些初步成果顯示功能性造影技術在臨床應用上具有高度潛力。總結而言，本論文在神經科學領域提供了功能性磁振造影的技術改進與整合，預期將能貢獻於國內的醫學エ程產業、個人化醫療以及提昇臨床診斷之精確度。

Since its debut in 1995, the resting-state functional magnetic resonance imaging (rs-fMRI) has received sustained attention from fundamental and clinical neuroscience. Because of its non-invasive mapping of brain network integrity and high clinical feasibility without task engagements, this technique has proliferated amongst the fundamental investigations of development, geriatrics, and psychiatric and neurological disorders. Regarding its clinical practice in personalized medicine, presurgical functional mapping is of increasing importance in clinical management to aid the surgical planning to patients with neurosurgical intervention. Targeting on presurgical mapping, rs-fMRI is occasionally used in clinical practices. The rationale is that this technique still faces several methodological challenges: (1) a practical measure of rs-fMRI data quality to obtain reliable functional networks that has yet to be determined; (2) the inter-subject variability in functional localization and lesion-related functional reorganization makes the seed selection difficult for mapping functional networks on the basis of anatomical landmark alone, and thus affect its clinical use; (3) Despite its importance and usefulness, a specialized clinical software that integrates complementary fMRI techniques for presurgical fMRI workflow is still lacking. Inherited from the rs-fMRI technique, this dissertation targets at three specific aims: (1) quality assurance from physiological contributions, (2) seed guidance for presurgical language mapping with seed-based rs-fMRI, and (3) technique integration in the specialized software for clinical practices. The ultimate goal of this PhD dissertation is to translate and integrate state-of-the-art fMRI techniques for presurgical mapping and clinical studies. In specific aim 1, we proposed a quality-assurance index for rs-fMRI to estimate the physiological contributions in spontaneous oscillations (PICSO). With the calibration through the phantom data, we verified that the PICSO showed a significantly positive correlation with the strength of functional connectivity while the tSNR was not, providing a practical quality-assurance indicator for all existing rs-fMRI data sets. In specific aim 2, we proposed a novel method to guide the seed selection for mapping the rs-fMRI language network by incorporating data-driven regional homogeneity and meta-analysis. The results demonstrated that localization performance on language network was significantly improved comparing to the seed selection based on MNI coordinate and was equivalent to the seed localization guided by task-fMRI activation. These results suggest that the proposed method may be an effective and beneficial approach for rs-fMRI mapping in the clinical practice, especially when patients have difficulties in compliances of task engagements. In specific aim 3, we developed the Integrated fMRI for Clinical Research (IClinfMRI) software package to incorporate advanced fMRI methods of task-fMRI, rs-fMRI, and cerebrovascular reactivity (CVR) mapping. Incorporating CVR technique is to indicate the potential false-negative areas in fMRI results, and to implement data conversion modules for facilitating clinical fMRI researches with the applicability to pre-surgical planning in the treatment of intracranial lesions. In summary, the dissertation was designed for translating fMRI techniques into clinical practice, initiating from the quality examination, seed guidance of rs-fMRI mapping, to platform development. Verifying the positive relation of the PICSO index with the strength of functional connectivity, proposing an effective approach for quality assurance of rs-fMRI mapping, and developing the IClinfMRI software in the clinical workflow. In the future direction, we will develop a purely data-driven approach independent of the needs that enable the identification of functional network through the deep learning algorithm. In conclusion, the proposed techniques and software in this thesis not only facilitate the application of fMRI techniques on daily clinical practices, but also improve the brain-mapping precision in personalized medicine.