3D C-Vit Model: Pediatric Brain Tumor Grading Performance on 340 MRI Cases

A retrospective study of 340 pediatric brain tumor cases (143 low-grade, 197 high-grade) used to validate a 3D hybrid deep learning model. The model, developed by Huixin Wu, was trained on five MRI sequences and evaluated using metrics like AUC and accuracy. The dataset was last updated on 2026-04-27.

Use Cases

• Benchmarking new tumor grading models against the reported 3D C-Vit performance metrics (e.g., 91.36% AUC).
• Analyzing the impact of specific model components (CAEFF, MSFE, MHSA) on grading accuracy based on ablation study results.
• Comparing deep learning model performance to traditional clinical and radiomics methods (e.g., SVM with 77.55% ACC).
• Investigating feature importance for pediatric brain tumor grading using the 59 key radiomics features identified by LASSO regression.

Strengths

• Model performance is quantitatively validated on a test set with reported metrics: 91.36% AUC, 86.53% accuracy, and 89.29 F1-score.
• Dataset includes 340 clinically annotated cases with a defined split (70:15:15) for training, validation, and testing.
• Input data is multi-modal, incorporating five standardized MRI sequences (CE-T1WI, T1WI, T2WI, FLAIR, ADC).
• Ablation experiments provide specific performance contributions for model components (e.g., MSFE increased ACC by 11.67%).

Limitations

• The dataset is very small (39.8 KB), suggesting it contains summary results or a document, not the raw image data.
• Row count and column-level documentation are unknown, limiting assessment of the underlying data structure.
• Data freshness should be verified as the last update date is in the future (2026-04-27).

Provenance

Source

figshare, authored by Huixin Wu.

Collection Method

Retrospective study with tumor regions annotated by two senior radiologists.

Freshness

Last updated 2026-04-27 04:12:02; freshness should be verified.