Synthetic Benchmark Dataset for OCT Similarity Metrics: Controlled Micro-, Meso-, and Macro-Structural Shifts
收藏资源简介：
Overview This dataset contains synthetic Optical Coherence Tomography (OCT) B-scans and associated physical parameter maps generated via a rigorous physics-based simulation. It serves as a benchmark for evaluating the performance of similarity metrics (e.g., MSE, SSIM, MS-SSIM, LPIPS, PSNR, VIF) in the context of optical coherence tomography. The dataset was generated to support the MICCAI 2026 submission: "Benchmarking OCT Scans Similarity Metrics via Physics-Based Simulatory Analysis with Controlled Micro-, Meso-, and Macro-Structural Shifts." Dataset Structure The data is organized into three hierarchical categories representing different biological scales: 1. Micro-structure (Scatterer Statistics) Tests sensitivity to sub-resolution scatterer density changes under constant backscattering amplitude. Dens_200000: Reference standard tissue (200,000 scatterers). Dens_50000: Low-density tissue (50,000 scatterers). 2. Meso-structure (Structural Inclusions) Tests sensitivity to discrete inclusions (e.g., vessels, cysts etc) embedded in the reference tissue. Meso_Dens: Inclusions defined by reduced scatterer density (0.1x of background). Meso_Amp: Inclusions defined by reduced backscattering amplitude (0.1x of background). Meso_Both: Inclusions defined by both reduced density (0.2x) and amplitude (0.2x). 3. Macro-structure (Layer Morphology) Tests sensitivity to morphological layer changes (e.g., epithelial thickness) on a complex heterogeneous background. Macro_Thin: Two-layer tissue with a thin top layer (Boundary at 240 µm). Macro_Thick: Two-layer tissue with a thick top layer (Boundary at 480 µm). Both macro sets feature a dim top layer (0.1x amplitude) and underlying meso-inclusions in the bottom layer to simulate complex clinical scenarios. File Contents Each scan folder contains: Scan_X.txt: Ground truth coordinates and amplitudes of discrete scatterers. Scan_X.png: The synthetic OCT structural B-scan. Scan_X_OAC.png: Optical Attenuation Coefficient map. Scan_X_SC.png: Speckle Contrast map. Scan_X_RSC.png: Refined Speckle Contrast map. Additional dataset for sensitivity evaluation - Metrics_Sensitivity_Benchmark_Grading_structural_changes.zip:This dataset contains physics-based synthetic OCT B-scans designed to evaluate the sensitivity of metrics to structural changes against a background of speckle noise. The archive is organized into three main sections. First, a 'Baseline' folder containing 20 independent speckle realizations of a fixed structure (400 micron layer thickness, 20 void inclusions) to establish the noise floor. Second, 'Series1_Macro_Thickness', which contains 10 steps of graded macro-structural changes where the epithelial layer thickness decreases from 380 microns to 200 microns (approaching a 2x reduction). Third, 'Series2_Meso_Count', which contains graded steps of meso-structural changes where the number of inclusions increases from 20 to 39 (approaching ~2x increase). Each step in these series contains 20 independent scans to allow for statistical analysis of metric distributions. This dataset is made freely available for any reuse. Use of this dataset requires appropriate citation of both this dataset (https://doi.org/10.5281/zenodo.18670884) and the associated papers: Methodology References [1] Sovetsky, A., Matveyev, A., Chizhov, P., Zaitsev, V., Matveev, L. (2025). OCT Scans Simulation Framework for Data Augmentation and Controlled Evaluation of Signal Processing Approaches. In: Fernandez, V., Wolterink, J.M., Wiesner, D., Remedios, S., Zuo, L., Casamitjana, A. (eds) Simulation and Synthesis in Medical Imaging. SASHIMI 2024. Lecture Notes in Computer Science, vol 15187. Springer, Cham.https://doi.org/10.1007/978-3-031-73281-2_12 [2] Nikoshin, D., Mikhailenko, D., Sovetsky, A., Matveyev, A., Zaitsev, V., Matveev, L. (2026). From Tissue-Mimicking Phantoms to Physics-Based Scans: Synthetic OCT for Few-Shot Foundation Model Training. In: Fernandez, V., Wiesner, D., Zuo, L., Casamitjana, A., Remedios, S.W. (eds) Simulation and Synthesis in Medical Imaging. SASHIMI 2025. Lecture Notes in Computer Science, vol 16085. Springer, Cham.https://doi.org/10.1007/978-3-032-05573-6_5 [3] MICCAI 2026 Submission (will be availiable soon) Repo: https://github.com/SynthOCTChallenge/SynthOCT_Baseline FundingThis dataset was curated and published with the support of the Russian Science Foundation (RSF):RSF Grant № 25-12-20032: "New Approaches to the Development of Algorithms for Analyzing OCT Scans: Modification and Optimization of Large Models Based on Physical Principles and Conditions of OCT Signal Formation" (https://rscf.ru/en/project/25-12-20032/)
概述 本数据集包含通过严谨的基于物理的仿真生成的合成光学相干断层扫描（Optical Coherence Tomography, OCT）B型扫描图像及相关物理参数图，旨在作为基准数据集，用于评估光学相干断层扫描场景下各类相似性指标（如均方误差MSE、结构相似性SSIM、多尺度结构相似性MS-SSIM、学习感知图像块相似度LPIPS、峰值信噪比PSNR、视觉信息保真度VIF）的性能。 本数据集的生成用于支撑MICCAI 2026投稿论文：《通过基于物理的仿真分析控制微观、介观与宏观结构偏移，实现OCT扫描相似性指标基准测试》。 数据集结构 数据按三个层级分类，对应不同的生物尺度： 1. 微观结构（散射体统计特性） 测试在恒定背散射振幅下，对亚分辨率散射体密度变化的敏感度。 - Dens_200000：参考标准组织（含200,000个散射体）。 - Dens_50000：低密度组织（含50,000个散射体）。 2. 介观结构（结构内含物） 测试对嵌入参考组织中的离散内含物（如血管、囊肿等）的敏感度。 - Meso_Dens：以降低的散射体密度（为背景的0.1倍）定义的内含物。 - Meso_Amp：以降低的背散射振幅（为背景的0.1倍）定义的内含物。 - Meso_Both：以同时降低的密度（为背景的0.2倍）与振幅（为背景的0.2倍）定义的内含物。 3. 宏观结构（层状形态） 测试在复杂异质背景下，对形态学层状变化（如上皮层厚度）的敏感度。 - Macro_Thin：双层组织，顶层较薄（边界位于240 µm处）。 - Macro_Thick：双层组织，顶层较厚（边界位于480 µm处）。 两类宏观结构数据集均设置了暗淡的顶层（振幅为背景的0.1倍），并在底层嵌入介观内含物，以模拟真实临床场景。 文件内容 每个扫描文件夹包含以下文件： - Scan_X.txt：离散散射体的地面真实坐标与振幅。 - Scan_X.png：合成OCT结构B型扫描图像。 - Scan_X_OAC.png：光学衰减系数（Optical Attenuation Coefficient, OAC）图。 - Scan_X_SC.png：斑点对比度（Speckle Contrast, SC）图。 - Scan_X_RSC.png：精细化斑点对比度（Refined Speckle Contrast, RSC）图。 用于敏感度评估的附加数据集——Metrics_Sensitivity_Benchmark_Grading_structural_changes.zip： 本数据集包含基于物理的合成OCT B型扫描图像，用于评估指标在斑点噪声背景下对结构变化的敏感度。该归档文件分为三个主要部分： 1. 「Baseline」文件夹：包含固定结构（400微米层厚度、20个空隙内含物）的20组独立斑点噪声实现，用于确定噪声基准线。 2. 「Series1_Macro_Thickness」：包含10个分级宏观结构变化步骤，上皮层厚度从380微米降至200微米（降幅接近2倍）。 3. 「Series2_Meso_Count」：包含10个分级介观结构变化步骤，内含物数量从20个增至39个（增幅接近2倍）。 上述每个步骤均包含20组独立扫描结果，可用于开展指标分布的统计分析。 本数据集可免费用于任何用途。使用本数据集时，需正确引用本数据集（https://doi.org/10.5281/zenodo.18670884）及相关论文： 方法学参考文献 [1] Sovetsky, A., Matveyev, A., Chizhov, P., Zaitsev, V., Matveev, L. (2025). 用于数据增强与信号处理方法受控评估的OCT扫描仿真框架. 见：Fernandez, V., Wolterink, J.M., Wiesner, D., Remedios, S., Zuo, L., Casamitjana, A. (eds.) 医学影像中的仿真与合成. SASHIMI 2024. 计算机科学讲义丛书, 第15187卷. Springer, Cham. https://doi.org/10.1007/978-3-031-73281-2_12 [2] Nikoshin, D., Mikhailenko, D., Sovetsky, A., Matveyev, A., Zaitsev, V., Matveev, L. (2026). 从组织模拟体模到基于物理的扫描：用于少样本大语言模型训练的合成OCT. 见：Fernandez, V., Wiesner, D., Zuo, L., Casamitjana, A., Remedios, S.W. (eds.) 医学影像中的仿真与合成. SASHIMI 2025. 计算机科学讲义丛书, 第16085卷. Springer, Cham. https://doi.org/10.1007/978-3-032-05573-6_5 [3] MICCAI 2026投稿论文（即将上线） 代码仓库：https://github.com/SynthOCTChallenge/SynthOCT_Baseline 资助信息 本数据集的整理与发布得到了俄罗斯科学基金会（Russian Science Foundation, RSF）的支持：RSF资助项目№25-12-20032：「基于OCT信号形成的物理原理与条件，开发OCT扫描分析算法：大语言模型的改进与优化」（https://rscf.ru/en/project/25-12-20032/）
