基於MFC之線狀光頻域式光學同調斷層掃描術即時成像平台開發

Abstract

隨著高階精密製程與智慧化生產的興起，非破壞性檢測在工業品管、半導體製程與生醫診斷等領域扮演日益關鍵的角色。光學同調斷層掃描術(Optical Coherence Tomography, OCT)作為一種高解析度、深度感測能力強的三維斷層光學影像技術，因具備無輻射、非接觸、即時成像等優勢，近年來廣泛應用於視網膜掃描、皮膚組織分析、材料缺陷檢測與微結構檢測等領域。然而，傳統的點掃描式OCT系統為取得影像，需倚賴掃描振鏡(Galvanometer scanner , Galvo)進行逐點掃描，受限於掃描速度、穩定性與樣品運動干擾，難以應對實務上對高速、大視野與高通量掃描的需求。 為改善傳統OCT 系統的限制，線狀光光學同調斷層掃描術(Line-field OCT)近年來逐漸受到重視。此技術透過線型照明與面陣或線陣列偵測器結合，可在單一曝光下取得二維掃描資料(B-scan)，不僅大幅提升影像擷取效率，也能減少機械移動元件，提高系統整體穩定性與耐用性，並特別適合應用於大視野、高通量的快速檢測需求。因此，為使線狀光掃描系統能夠充分發揮效能，開發一套能即時處理大量資料，並能與二維相機相容的操作平台也顯得格外關鍵。 在本篇論文當中，我們以實驗室既有的頻域式光學同調斷層掃描術(Spectral-domain OCT, SD-OCT)平台為基礎，開發整合線狀光頻域式光學同調斷層掃描術(Line-field Spectral-Domain OCT, LF-SD-OCT)成像模組，並結合圖形處理器(Graphics Processing Units, GPU)加速影像處理流程，以縮短資料擷取與影像重建所需時間。平台開發採用微軟基礎類別館(Microsoft Foundation Classes, MFC) C++ 類別庫進行實作，並新增相機控制模組以支援二維資料擷取與處理。系統可依據設定選擇執行一維或二維影像處理，並加入多項專為LF-SD-OCT的功能，例如：從B-scan影像中選擇特定A-scan進行訊號強度分析，或顯示相機擷取之未處理的原始二維影像，作為系統校準與優化依據。GPU加速的部分則使用 C++ 搭配NVIDIA CUDA Toolkit所提供的函式庫實作，並針對LF-SD-OCT 影像特性，修改了函示庫中影像處理的內容。開發完成後，本研究將平台整合至實驗室既有的SD-OCT系統架構中，並比較SD-OCT與 LF-SD-OCT系統在成像速度上的差異，以及 GPU加速前後的影像處理時間改善效果。本研究所建立之平台驗證了LF-SD-OCT的系統效能與GPU加速優勢，於未來應用於高速、高穩定度之非破壞性檢測提供可行方案。

With the rise of advanced precision manufacturing and intelligent production, non-destructive testing (NDT) has played an increasingly critical role in fields such as industrial quality control, semiconductor manufacturing, and biomedical diagnostics. Optical Coherence Tomography (OCT), a high-resolution optical imaging technology with strong depth-sensing capabilities, has been widely applied in recent years due to its advantages of being radiation-free, non-contact, and capable of real-time imaging. It has found broad applications in retinal scanning, skin tissue analysis, material defect inspection, and microstructure evaluation. However, conventional point-scanning OCT systems require the use of a Galvanometer scanner (Galvo) to perform point-by-point scanning in order to acquire images. These systems are limited by scanning speed, stability, and susceptibility to sample motion, making them unsuitable for practical applications involving high-speed, wide-field, and high-throughput imaging. For example, in large-area defect inspection on production lines, insufficient scanning speed may cause production delays; in biomedical applications, even minor patient movements may result in image blurring or distortion, affecting diagnostic accuracy. To overcome the limitations of traditional OCT systems, Line-field Optical Coherence Tomography (LF-OCT) has gained increasing attention in recent years. This technique employs line-shaped illumination combined with area or line-array detectors to acquire two-dimensional scan data (B-scan) in a single exposure. This significantly improves imaging efficiency, reduces the need for mechanical movement components, and enhances overall system stability and durability. It is especially suitable for fast inspection tasks requiring wide fields of view and high throughput. Therefore, in order to fully utilize the performance of line-field scanning systems, it is also crucial to develop an operating platform that can handle large volumes of data in real time and is compatible with two-dimensional cameras. In this study, we developed and integrated a Line-field Spectral Domain Optical Coherence Tomography (LF-SD-OCT) imaging module based on the laboratory's existing SD-OCT platform. Graphics Processing Units (GPUs) were incorporated to accelerate the image processing workflow, thereby reducing the time required for data acquisition and image reconstruction. The platform was developed using the Microsoft Foundation Classes (MFC) C++ library, with a newly added camera control module to support two-dimensional data acquisition and processing. The system allows users to select between one-dimensional or two-dimensional image processing based on configuration settings and includes several features specifically designed for LF-SD-OCT. These include selecting specific A-scans from B-scan images for signal intensity analysis and displaying raw two-dimensional images captured by the camera for system calibration and optimization. GPU acceleration was implemented using C++ together with libraries provided by the NVIDIA CUDA Toolkit, and the image processing routines were modified to fit the characteristics of LF-SD-OCT imaging. After development, the platform was integrated into the laboratory’s existing SD-OCT system architecture, and a comparative analysis was conducted to evaluate the differences in imaging speed between the SD-OCT and LF-SD-OCT systems, as well as improvements in image processing time before and after GPU acceleration. The platform developed in this study demonstrates the performance advantages of LF-SD-OCT and the benefits of GPU acceleration, offering a viable solution for future applications in high-speed, high-stability non-destructive testing.