Non-Contrast Brain CT Scans for AI-Based Cerebral Venous Thrombosis Detection

Iran Cerebral Venous Thrombosis Registry provided a dataset of 692 patients, including 258 with CVT and 434 controls. The study used NCCT images processed by a pre-trained CT-Foundation model to generate embeddings for training machine learning classifiers. The research was published on figshare by author figshare admin karger, with a last update in April 2026.

Use Cases

• Training AI models for CVT detection based on non-contrast CT scan embeddings.
• Benchmarking classifier performance (e.g., SVM, Random Forest) for a medical imaging classification task.
• Developing ensemble models to improve sensitivity in flagging suspicious cases for expert review.

Strengths

• Dataset includes 692 patient cases with a defined split of 258 CVT and 434 control subjects.
• Model performance metrics, such as a specificity of 0.856 for Random Forest and an AUC of 0.718 for SVM, are provided.
• Images were processed using a pre-trained CT-Foundation model, suggesting a standardized feature extraction method.

Limitations

• Row count and column-level documentation for the underlying image data are unknown.
• Data may reflect geographic bias inherent to the Iran Cerebral Venous Thrombosis Registry source.
• The dataset size is 1.3 MB, which is small for image-based AI tasks; the actual image files are not directly described.

Provenance

Source

Iran Cerebral Venous Thrombosis Registry (ICVTR), code 9001013381.

Collection Method

Retrospectively collected dataset of patient NCCT scans.

Freshness

Last updated 2026-04-18 07:55:34

Geography

Iran (inferred from registry name).