PET/CT Feature Clusters for 289 Fever of Unknown Origin Patients

A retrospective cohort of 289 adult patients with fever or inflammation of unknown origin who underwent 18F-FDG PET/CT imaging. The dataset, created by Sule Ceylan and last updated in May 2026, contains features extracted from PET/CT scans, including organ involvement patterns and SUVmax values, which were analyzed using four unsupervised clustering algorithms. The consensus clustering identified four distinct imaging phenotypes, which were compared to final clinical diagnoses.

Use Cases

• Benchmarking unsupervised clustering algorithms based on real-world medical imaging features.
• Exploring the relationship between PET/CT imaging phenotypes and clinical etiologies like infections or rheumatologic diseases.
• Developing feature extraction pipelines for medical imaging data involving reticuloendothelial, parenchymal, and vascular involvement.

Strengths

• Includes data from 289 patients with a defined mean age of 59 ± 17 years and gender distribution.
• Features were extracted using a standardized protocol covering multiple organ systems and uptake patterns.
• Analysis employed a consensus clustering approach combining four different unsupervised algorithms to derive robust clusters.

Limitations

• The dataset is very small at 14.4 KB, suggesting limited scope or summary-level data rather than raw patient records.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain modeling tasks.

Provenance

Source

figshare, author Sule Ceylan.

Collection Method

Retrospective cohort study; data extracted from PET/CT scans of patients meeting FUO/IUO criteria.

Freshness

Last updated 2026-05-29 05:11:48; freshness should be verified.