Open dataset of annotated DICOM images of head CTA for intracranial aneurysm detection
Authors/Creators
- 1. Gammamed-Soft, Ltd
-
2.
Pirogov Russian National Research Medical University
- 3. 3Research and Clinical Institute for Pediatrics Named after Yuri Veltischev, Pirogov Russian National Research Medical University
- 4. Moscow Multidisciplinary Clinical Center «Kommunarka», Moscow Healthcare Department
- 5. N.V. Sklifosovsky Research Institute for Emergency Medicine, Moscow Healthcare Department
- 6. Clinical Medical Center, Russian University of Medicine, Ministry of Healthcare of Russia
- 7. Laboratory of Medical Computing System, Scobeltsyn Nuclear Physics Research Institute, Lomonosov Moscow State University
Description
Overview
We release a collection of computed tomography angiography (CTA) DICOM studies, including confirmed intracranial aneurysm (IA) cases and aneurysm-free controls. This dataset contains 172 anonymized CTA scan series (90 controls, 82 IA cases) with expert annotations of aneurysms and metadata. The data collected from three clinical centers addresses the critical shortage of high-quality, diverse cerebrovascular imaging data. A consecutive sampling method was used to collect the data. The dataset is characterized by a wide variety of pathology shapes and sizes, including a large number (24%) of difficult-to-diagnose miliary aneurysms.
Features
Image series
- DICOM format
- Full DICOM anonymized
- Face removal (Defacing)
- Slice thickness 0.5-2 mm
Annotations
- Curation by two experts: a neurosurgeon and neuroradiologist, each with over 20 years of experience
- Threshold-based segmentation after manual aneurysm isolation
- STL format
Metadata
- Table in CSV format
- Contains:
· Patient ID
· Presence of pathology
· Aneurysm ID
· Sex
· Age
· Parent artery
· Shape (saccular/nonsaccular)
· Position classification (lateral/bifurcation/unclassifiable)
· Type of nonsaccular aneurysm (blister-like/fusiform/eccentric-fusiform/fusiform-saccular/unclassifiable)
· Size classification (miliary (<3mm)/middle (3-15mm)/large (15-25mm)/giant (≥25mm))
· Measurements (neck size, maximum length, maximum width, aneurysm volume)
· Complexity factors (rupture, bleb, thrombus, calcification, vessel arising from the dome)
· Series description (number of images, slice thickness, inclusion of the carotid bifurcation area)
Applications
- Training, validation and testing of artificial intelligence models for IA detection/segmentation
- Hemodynamic risk simulation (computational fluid dynamics)
- Morphometric analysis
Usage notes
Image data for each case is stored in a folder named with its unique identifier (`CaseID`). Segmentation masks are named `an_CaseID` for single aneurysms and `an_CaseID_N` for multiple aneurysms, where N is the specific aneurysm identifier from the metadata (e.g., an_123_1.stl, an_123_2.stl).
This study involving human data was conducted in accordance with the Declaration of Helsinki and approved by the Biomedical Ethics Committee of the N.V. Sklifosovsky Research Institute for Emergency Medicine (Protocol No. 2-23, March 28, 2023). The dataset is publicly accessible under CC-BY 4.0. We kindly request that any publication using this dataset cites the original paper describing its creation: https://doi.org/10.3390/data11040074.
Files
Dataset.zip
Additional details
Related works
- Is described by
- Data paper: 10.3390/data11040074 (DOI)