Ming03/DSI-SCG-ECG

--- license: cc-by-nc-4.0 language: - en size_categories: - 100B<n<1T --- # Dataset Description This repository contains the dataset for the paper: **"Cardiac 3D Mechanical and Electrical Signal Reconstruction via Defocused Speckle Imaging"**. It provides dual-camera Defocused Speckle Imaging (DSI) videos and cardiac signals. The dataset is collected from two distinct cohorts to ensure robustness and clinical relevance. * **Total Participants:** 50 * **Lab Cohort:** 20 healthy subjects. * **Clinical Cohort:** 30 patients from the ultrasound department. * **Data Modalities:** Dual-camera DSI videos, Seismocardiogram (SCG), Gyrocardiogram (GCG), and Electrocardiogram (ECG). # Data Structure The dataset includes raw videos and processed signal files in Python `.pkl` format: * `Lab_data.pkl`: Processed signals for the 20 healthy subjects. * `Clinic_data.pkl`: Processed signals for the 30 ultrasound department patients. ## Sampling Rates * **DSI Optical Flow & Mechanical Signals (SCG/GCG):** 250 Hz * **ECG Reference Signals:** 500 Hz # Signal Dictionary (Processed Data) When loading the `.pkl` files, the data for each subject is stored as dictionaries containing numpy `ndarray` objects. Most 2D arrays (e.g., `(N, 500)`) represent `N` cardiac cycles resampled to a fixed length of 500. ## 1. Reference Signals (Ground Truth) * `ECG`: Reference Electrocardiogram signal. * `SCG`: Reference Seismocardiogram signal. * `GCGx`: Reference Gyrocardiogram signal (x-axis). * `GCGy`: Reference Gyrocardiogram signal (y-axis). ## 2. Reconstructed Signals (via Physical Model) Signals reconstructed using the physical model from two tracked points (`g` and `r`): * `SCG_g` / `SCG_r`: Reconstructed SCG signals from point `g` and point `r`. * `GCGx_g` / `GCGx_r`: Reconstructed GCGx signals from point `g` and point `r`. * `GCGy_g` / `GCGy_r`: Reconstructed GCGy signals from point `g` and point `r`. ## 3. Camera Motion Signals (DSI Optical Flow) Raw motion signals captured by the two cameras at two specific points (`g` and `r`) across the x and y axes: * **Camera 1:** `cam1_gx`, `cam1_gy`, `cam1_rx`, `cam1_ry` * **Camera 2:** `cam2_gx`, `cam2_gy`, `cam2_rx`, `cam2_ry` ## 4. Raw Signal Lengths 1D arrays recording the original signal length of each cardiac cycle before length normalization: * `ECG_Raw`: Original length of ECG signals per cardiac cycle. * `SCG_Raw`: Original length of SCG signals per cardiac cycle. * `GCGx_Raw`: Original length of GCGx signals per cardiac cycle. * `GCGy_Raw`: Original length of GCGy signals per cardiac cycle. # Usage Example Here is a quick example of how to load and explore the `.pkl` files in Python: ```python import pickle # Load the Lab data with open('Lab_data.pkl', 'rb') as f: lab_data = pickle.load(f) subject_id = 'Lab_1' video_id = 0 subject_data = lab_data[subject_id][video_id] # Extract reference ECG and reconstructed SCG ecg_ref = subject_data['ECG'] # shape: (N_cycles, 500) scg_reconstructed = subject_data['SCG_g'] # shape: (N_cycles, 500) raw_lengths = subject_data['ECG_Raw'] # shape: (N_cycles,) print(f"Number of cardiac cycles: {ecg_ref.shape[0]}") ``` # License Our dataset is CC-BY-NC 4.0 licensed, as found in the LICENSE file.--- 许可证：CC-BY-NC-4.0 语言： - 英语 数据规模类别： - 100B < 总数据量 < 1T --- # 数据集概述 本仓库配套于论文《基于散焦散斑成像的心脏三维机械与电信号重建》，存储对应数据集。本数据集包含双摄像头散焦散斑成像（Defocused Speckle Imaging, DSI）视频与心脏信号，为确保模型鲁棒性与临床相关性，数据来自两个独立队列。 * **总受试者数：50名** * **实验室队列：20名健康受试者** * **临床队列：30名来自超声科的患者** * **数据模态：双摄像头DSI视频、心震图（Seismocardiogram, SCG）、心陀螺图（Gyrocardiogram, GCG）与心电图（Electrocardiogram, ECG）** # 数据结构 本数据集包含原始视频与Python格式的`.pkl`处理后信号文件： * `Lab_data.pkl`：20名健康受试者的处理后信号数据 * `Clinic_data.pkl`：30名超声科患者的处理后信号数据 ## 采样率 * **DSI光流与机械信号（SCG/GCG）：250 Hz** * **ECG参考信号：500 Hz** # 处理后数据信号字典 加载`.pkl`文件时，每名受试者的数据以字典形式存储，其中包含NumPy的`ndarray`对象。多数二维数组（例如`(N, 500)`）表示经过重采样至固定长度500的`N`个心动周期。 ## 1. 参考信号（真值标签） * `ECG`：参考心电图信号 * `SCG`：参考心震图信号 * `GCGx`：参考心陀螺图x轴信号 * `GCGy`：参考心陀螺图y轴信号 ## 2. 物理模型重建信号 通过物理模型从两个追踪点位（`g`与`r`）重建得到的信号： * `SCG_g` / `SCG_r`：分别来自点位`g`与点位`r`的重建心震图信号 * `GCGx_g` / `GCGx_r`：分别来自点位`g`与点位`r`的重建心陀螺图x轴信号 * `GCGy_g` / `GCGy_r`：分别来自点位`g`与点位`r`的重建心陀螺图y轴信号 ## 3. 相机运动信号（DSI光流） 两台相机分别在两个特定点位（`g`与`r`）的x、y轴上采集得到的原始运动信号： * **相机1：** `cam1_gx`、`cam1_gy`、`cam1_rx`、`cam1_ry` * **相机2：** `cam2_gx`、`cam2_gy`、`cam2_rx`、`cam2_ry` ## 4. 原始信号长度 一维数组，记录每一个心动周期在长度归一化前的原始信号长度： * `ECG_Raw`：每名受试者单心动周期心电图信号的原始长度 * `SCG_Raw`：每名受试者单心动周期心震图信号的原始长度 * `GCGx_Raw`：每名受试者单心动周期心陀螺图x轴信号的原始长度 * `GCGy_Raw`：每名受试者单心动周期心陀螺图y轴信号的原始长度 # 使用示例 以下为Python语言中加载与探索`.pkl`文件的快速示例： python import pickle # 加载实验室队列数据 with open('Lab_data.pkl', 'rb') as f: lab_data = pickle.load(f) subject_id = 'Lab_1' video_id = 0 subject_data = lab_data[subject_id][video_id] # 提取参考心电图与重建心震图信号 ecg_ref = subject_data['ECG'] # 形状：(N_cycles, 500) scg_reconstructed = subject_data['SCG_g'] # 形状：(N_cycles, 500) raw_lengths = subject_data['ECG_Raw'] # 形状：(N_cycles,) print(f'心动周期总数：{ecg_ref.shape[0]}') # 许可证 本数据集采用CC-BY-NC 4.0许可证，详细条款请参见LICENSE文件。