基於嵌入式平台之可攜式光學同調斷層掃描系統之整合設計與實作

Abstract

光學同調斷層掃描術(Optical coherence tomography, OCT)做為一種高解析度且非侵入式的成像技術，現今已被廣泛應用於眼科、皮膚科以及齒科等醫療診斷領域，提供組織微結構的即時觀察，對精準診斷至關重要。然而傳統OCT設備因成本高昂與體積龐大，應用受限於大型醫療中心，難以普及至小型醫療機構。為提升可攜性與降低成本，手持式OCT系統成為研究焦點。為提升可攜性與降低建置成本，手持式OCT系統近年成為研究焦點。雖有研究採用如Raspberry Pi與Intel NUC等單板電腦進行系統控制與訊號處理，但OCT高頻寬資料的即時處理對硬體效能要求極高，仍普遍存在運算能力不足、功耗過高及整合困難等問題，需更高效的硬體架構來克服這些限制。 本研究延續本實驗室先前開發之基於現場可程式化邏輯閘陣列(Field programmable gate array, FPGA)的OCT處理架構，結合芯聖科技(OPXION Technology, Inc.)所開發之手持式皮膚掃描儀，進行OCT掃描與影像處理，並進一步於NVIDIA Jetson Xavier NX平台上開發具備即時預覽、三維資料擷取與量化分析功能之圖形化使用者介面(Graphical User Interface, GUI)。其中FPGA專責影像處理與傳輸，發揮其高速且低延遲之硬體優勢。Jetson則負責圖形化介面與深度學習分析，善用其圖形處理器(Graphics processing unit, GPU)與機器學習之運算能力。透過兩者分工協作，大幅提升系統整體效能與即時性，亦保留後續擴充之彈性。 本系統以Qt為開發框架，設計了兩種主要操作模式，分別對應於即時顯示與完整三維資料擷取，並以多執行緒方式確保介面穩定與資料同步性。傳輸介面採用通用序列匯流排(Universal serial bus, USB) 3.0串接USB控制器模組，接收資料後進行動態記憶體緩衝與管理，並提供使用者瀏覽、儲存與後續分析之介面。此外亦為FPGA回傳之32位元浮點格式資料進行資料轉換，將系統實作資料解析與強度映射流程，考量位元序對資料正確性的影響，設計自定轉換函式，確保接收資料能正確對應至灰階影像，供使用者觀測OCT水平及垂直切面影像。而後亦設計測試樣本驗證資料格式處理正確性，並進行影像品質指標評估，以量化強度轉換對視覺細節之保留程度，驗證本系統之穩定性與準確性。整體系統介面操作簡易，適合應用於攜帶式OCT裝置，提供偏鄉基礎醫療應用。

Optical coherence tomography (OCT), as a high-resolution and non-invasive imaging technique, has been widely applied in medical diagnostic fields such as ophthalmology, dermatology, and dentistry. It enables real-time observation of tissue microstructures, which is crucial for accurate diagnosis. However, conventional OCT systems are often limited to large medical centers due to their high cost and bulky size, making them difficult to adopt in smaller clinics. To improve portability and reduce costs, handheld OCT systems have become a key research focus in recent years. Although some studies have utilized single-board computers like Raspberry Pi and Intel NUC for system control and signal processing, the real-time processing of OCT’s high-bandwidth data imposes demanding requirements on hardware performance. Common challenges still include insufficient computing power, high power consumption, and integration difficulties. Therefore, a more efficient hardware architecture is needed to overcome these limitations. This study builds upon a previously developed Optical Coherence Tomography (OCT) processing architecture based on a Field-Programmable Gate Array (FPGA) by our laboratory. It integrates a handheld skin scanner developed by OPXION Technology, Inc. to perform OCT scanning and image processing. Furthermore, a graphical user interface (GUI) was developed on the NVIDIA Jetson Xavier NX platform, providing real-time preview, 3D data acquisition, and quantitative analysis capabilities. In this system, the FPGA is dedicated to image processing and data transmission, leveraging its advantages of high speed and low latency. The Jetson handles the GUI and deep learning-based analysis, utilizing its Graphics Processing Unit (GPU) and machine learning computational power. Through this division of labor, the system achieves significantly enhanced overall performance and real-time responsiveness, while also maintaining flexibility for future expansion. The system is developed using the Qt framework and features two primary operating modes: real-time display and full 3D data acquisition. A multi-threaded architecture ensures interface stability and data synchronization. The data transmission interface utilizes USB 3.0, which connects to a USB controller module. Upon receiving data, the system performs dynamic memory buffering and management, providing users with functions for viewing, saving, and further analysis. To handle the 32-bit floating-point data returned from the FPGA, custom data conversion functions were implemented. These functions account for byte order (endianness) to ensure correct mapping of received data into grayscale images, allowing users to observe both horizontal and vertical OCT cross-sectional images. Test samples were designed to verify the correctness of data format processing, and image quality metrics were evaluated to quantify the preservation of visual detail during intensity conversion, thereby validating the system’s stability and accuracy. The overall interface is user-friendly and well-suited for portable OCT applications, making it ideal for use in primary care in resource-limited settings.