Processed ECG Scalogram Dataset, Metadata, and Preprocessing Workflow for Multi-Class ECG Classification using a Hybrid DCNN–BiLSTM–SE–MHSA model

This repository contains processed ECG scalogram images, metadata files, preprocessing code, and supporting documentation associated with the study entitled: "A Multi-Class ECG Classification Model using Hybrid DCNN–BiLSTM with SE and Multi-Head Self-Attention Network" Raw ECG recordings were obtained from publicly available PhysioNet databases, including: MIT-BIH Arrhythmia Database (ARR) BIDMC Congestive Heart Failure Database (CHF) MIT-BIH Normal Sinus Rhythm Database (NSR) The preprocessing workflow implemented in this repository includes: Butterworth bandpass filtering (0.5–50 Hz) Resampling to 128 Hz Signal standardization to 65,536 samples Amplitude normalization to the range [-1,1] Fixed-length segmentation (4096 samples) with 50% overlap Morlet-wavelet-based Continuous Wavelet Transform (CWT) Pseudo-colour scalogram generation Image resizing to 227 × 227 × 3 pixels The repository contains: Processed ECG scalograms from ARR, CHF, and NSR classes Metadata files and combined dataset metadata Preprocessing notebook (PLOS_ECG_Data_Preparation.ipynb) Documentation files describing data sources, preprocessing steps, and dataset statistics These materials are provided to support transparency, reproducibility, and reuse of the methodology described in the associated manuscript.