Inception-ResNetV2-SE Model Performance Under Alternative CT Preprocessing Methods

9.3 KB of tabular data compares the performance of a deep learning model for CT scan analysis under four preprocessing variants. The data, authored by Chao Zhang and last updated in May 2026, includes discrimination metrics like AUROC and PR-AUC, as well as threshold-based metrics and statistical significance tests. It shows that all tested alternative CT window settings and normalization methods performed worse than the standard 40/350 HU window with z-score normalization.

Use Cases

• Benchmarking CT scan preprocessing pipelines based on reported AUROC and PR-AUC values.
• Evaluating the impact of window width and level settings on model sensitivity and specificity.
• Comparing normalization methods (z-score vs. min-max) for deep learning in medical imaging.

Strengths

• Includes statistical significance tests (DeLong P values) for performance comparisons.
• Directly compares four specific preprocessing variants against a defined baseline.
• Released under a permissive CC-BY-4.0 license for reuse.

Limitations

• The dataset is very small at 9.3 KB, indicating limited scope.
• Row count and column-level documentation are unknown, requiring manual inspection after download.

Provenance

Source

figshare

Freshness

Last updated 2026-05-14 17:44:32; freshness should be verified.