Baseline PET/CT Habitat Radiomics for Predicting Lymphoma Treatment Response

Yu He's dataset supports a comparative study of baseline 18F-FDG PET/CT habitat radiomics versus dual-channel deep learning for predicting early metabolic response in diffuse large B-cell lymphoma. It contains data from 148 patients (101 EMR, 47 non-EMR) retrospectively enrolled between December 2018 and August 2024. The dataset was last updated on June 4, 2026.

Use Cases

• Training machine learning models to predict interim PET response based on habitat radiomics features.
• Comparing the performance of radiomics models (Habitat_MLP) against deep learning architectures (DL_DenseNet161).
• Developing decision-support tools for pretreatment risk stratification in DLBCL patients.
• Analyzing the relationship between intratumoral heterogeneity, characterized by voxel-wise habitat subregions, and treatment outcome.

Strengths

• Dataset includes 148 patient records with ground truth labels (EMR vs. non-EMR) based on Deauville scores.
• Model performance metrics are reported, including an AUC of 0.871 for the habitat radiomics model on the test set.
• Study design includes a dedicated test set of 45 patients used exclusively for final model evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying feature data is unknown, which may limit suitability assessment.
• Data is from a single-center retrospective study, which may limit generalizability.

Provenance

Source

figshare, author Yu He.

Collection Method

Retrospective collection from patients who underwent baseline PET/CT and R-CHOP or R-CHOP-like chemotherapy.

Time Range

December 2018 to August 2024.

Freshness

Last updated 2026-06-04 04:30:21; freshness should be verified.