PET-FM-Bench: Patient-Level Splits for Nine PET Imaging Tasks

Patient-level train, calibration, test, and test-retest splits for nine Positron Emission Tomography (PET) tasks. The dataset is hosted on Kaggle and appears designed for benchmarking machine learning models in medical imaging. The specific source institution, collection date, and data volume are not provided in the available metadata.

Use Cases

• Benchmarking model generalization across patient cohorts based on the provided patient-level splits.
• Training machine learning models for PET image analysis using the defined training set.
• Calibrating model confidence or uncertainty using the dedicated calibration split.
• Evaluating model performance on held-out data with the test split.
• Assessing model consistency or reliability using the test-retest split.

Strengths

• Provides structured patient-level splits for nine distinct PET imaging tasks, which can facilitate standardized evaluation.
• Includes dedicated calibration and test-retest splits, which suggests a design for robust model validation.

Limitations

• Row count and total data volume are unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Collection Method

Likely derived from medical imaging studies, but the specific collection method is not detailed.