Data and code: Prediction of Paroxysmal Atrial Fibrillation based on Time- frequency Analysis Network and related studies (Manuscript)

These data and codes were used in the submitted manuscript "Prediction of Paroxysmal Atrial Fibrillation based on Time- frequency Analysis Network and related studies", which includes all data and codes for this study. The AFPDB database used in this study is a public dataset and therefore not uploaded here, the download link is https://physionet.org/content/afpdb/1.0.0/. Data: It contains the clinical data from Shandong Provincial Hospital (SPHD) used in this study, and the data have been removed from sensitive information. It includes the original .edt data file and the .csv annotation file, all 24h ECG records. Since Mendeley only has 10G of space available, only a portion of the raw data were uploaded. To avoid lack of space and to ensure that all data is uploaded, we also upload the processed .m data files, which are all the data for this study. 'data_n.m' and 'data_p.m' are data from the AFPDB database, and 'data_paf.m' and 'data_nor.m' are the data in SPHD. ' T-Fmaps' folder is the generated ECG time-frequency map and RR interval time-frequency map. Code: 'edtReader.m' is the code to read the .edt file; 'ConstructTFANet.mlx' is the code to construct TFANet; 'TFANet.m ' is the TFANet network; 'PAFpredict.m' is the code for prediction experiments; 'PAFpredictcross' is the code for cross-patient experiments. No 5 cross-patient experiments were uploaded separately, and only the input data needed to be changed at runtime. The remaining files are functions for denoising and normalization, functions for generating time-frequency maps, code for predicting PAF using RR intervals, predicting PAF based on different time durations, predicting PAF using three classical CNN models, deep learning visualization, and so on. The running environment is Matlab R2022b.

本研究相关数据与代码已用于已投稿论文《"Prediction of Paroxysmal Atrial Fibrillation based on Time-frequency Analysis Network and related studies"》，包含本研究所需的全部数据与代码。本研究使用的AFPDB数据库为公开数据集，故未在此处上传，其下载链接为https://physionet.org/content/afpdb/1.0.0/。
数据部分：包含本研究使用的山东省立医院（Shandong Provincial Hospital, SPHD）临床数据，已完成敏感信息脱敏处理。数据包括原始.edt格式文件与.csv格式注释文件，均为24小时心电（Electrocardiogram, ECG）记录。由于Mendeley平台仅提供10GB存储空间，仅上传了部分原始数据。为避免存储空间不足并确保完整上传本研究所需全部数据，本次同时上传了处理后的.m格式数据文件。其中，data_n.m与data_p.m对应AFPDB数据库数据集，data_paf.m与data_nor.m对应山东省立医院数据集。`T-Fmaps`文件夹为生成的ECG时频图与RR间期时频图。
代码部分：edtReader.m为.edt文件读取代码；ConstructTFANet.mlx为构建时频分析网络（Time-frequency Analysis Network, TFANet）的代码；TFANet.m为TFANet网络定义代码；PAFpredict.m为阵发性心房颤动（Paroxysmal Atrial Fibrillation, PAF）预测实验代码；PAFpredictcross为跨患者实验代码。本次未单独上传5折跨患者实验代码，仅需在运行时修改输入数据即可。其余文件涵盖降噪与归一化函数、时频图生成函数、基于RR间期预测PAF的代码、基于不同时长数据的PAF预测代码、使用三种经典卷积神经网络（Convolutional Neural Network, CNN）模型预测PAF的代码、深度学习可视化工具等。本项目运行环境为Matlab R2022b。