This dataset provides myocardial perfusion scintigraphy (MPS) images in both DICOM (raw input) and NIfTI (processed) formats, supporting research on diagnostic automation, segmentation of the left ventricular wall, and image quality analysis. All relevant demographic and clinical metadata, such as patient age and gender, are embedded within the anonymized DICOM headers and can be programmatically extracted using standard DICOM parsing tools.

Images can be accessed using ITK-SNAP, 3D Slicer, or Python libraries such as pydicom, nibabel, and SimpleITK. NIfTI masks, where available, use the label 1 to indicate the left ventricular wall and 0 for background. The dataset structure is compatible with standard medical image processing pipelines and is suitable for training segmentation models, including nnU-Net.

To support reproducible segmentation workflows, a companion software capsule has been developed and made publicly available. It provides the necessary files and instructions to train and apply the nnU-Net 2D model using this dataset. The nnU-Net framework is a widely adopted and self-configuring architecture for medical image segmentation, known for its adaptability and high performance across diverse imaging modalities. The version included in the capsule performs training on fold 0 only, as a simplified alternative to the default 5-fold cross-validation, due to computational constraints. For full training, the user may uncomment the corresponding commands in the exec.sh script and adjust prediction parameters accordingly.

The capsule includes a Dockerfile for environment setup, the exec.sh automation script, a metadata.yml file with project information, and a readme.txt detailing installation and execution. The source code and setup instructions are accessible at: https://doi.org/10.6084/m9.figshare.27047815.v1.

For further technical guidance, users may refer to the related publication and contact the authors. When reusing this resource, please cite the dataset and acknowledge the authors as follows:

Wesley Pacheco Calixto [ORCID: 0000-0002-1928-4432] e-mail: wesley.pacheco@ufg.br

Thiago Fellipe Ortiz de Camargo [ORCID: 0000-0002-6917-8339]

Solange Amorim Nogueira [ORCID: 0000-0001-7481-7142]

Fernanda Ambrogi Barbosa da Luz [ORCID: 0009-0004-8007-0670]



### Demographic Metadata and DICOM Parsing

This dataset includes a file named `demographics.csv`, which provides basic clinical metadata for each study, including sex, age, and BMI (when available). These values were harmonized using information stored in REDCap and verified through manual inspection of anonymized DICOM headers.

To facilitate reproducibility, we provide the script `extract_dicom_metadata.py`, which can be used to extract metadata from DICOM files. The current `demographics.csv` was generated by applying this script to a single representative exam. All studies in the dataset follow the same acquisition protocol and DICOM structure.

Users may apply the script to other DICOM files if they wish to verify consistency or extract additional metadata fields.

### Versioning and Citation
This is version 1.0.0 of the dataset. Future releases may include additional annotations (e.g., diagnostic outcomes or artifact labels), subject to further ethical approvals.

If you use this dataset, please cite it as:

Calixto, W., Nogueira, S., Luz, F., & Ortiz de Camargo, T. F. (2025). Myocardial perfusion scintigraphy image database (version 1.0.0). PhysioNet. RRID:SCR_007345. https://doi.org/10.13026/*****


### Ethics Statement
This study was approved by the institutional ethics committee under protocol number CAAE number 65628422.7.0000.0071. All exams were retrospectively collected and anonymized in accordance with the Basic Application Confidentiality Profile (DICOM PS3.15 Annex E), ensuring removal of all personally identifiable information. Patient consent was waived due to the retrospective and anonymized nature of the study.

### Directory Structure

/
├── DICOM/                 # Raw SPECT images in DICOM format  
├── NIfTI/                 # Converted images in NIfTI format  
├── segmentation_masks/   # Optional — predicted masks (if included)  
├── README.txt             # This file