MedForge-90K: Real and Forged Medical Scans for Interpretable Deepfake Detection

MedForge-90K is a large-scale benchmark pairing real clinical-style images with high-fidelity forged counterparts for medical deepfake detection. The dataset provides pixel-level edit localization and expert-aligned, visually grounded chain-of-thought supervision for forgery-aware reasoning. It was created by RichardChenZH and last updated on Hugging Face in April 2026.

Use Cases

• Training deepfake detection models based on paired real and forged medical scans.
• Developing interpretable AI systems based on the provided chain-of-thought supervision.
• Benchmarking lesion localization algorithms based on the pixel-level edit masks.
• Studying realistic text-guided lesion editing and removal in authentic medical images.

Strengths

• Large-scale benchmark with 90,000 entries suggested by the title.
• Provides pixel-level edit localization for precise forgery analysis.
• Includes expert-aligned, visually grounded chain-of-thought supervision for reasoning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific collection of authentic scans used.

Provenance

Source

Hugging Face dataset by author RichardChenZH.

Collection Method

Pairs real clinical-style images with high-fidelity forged counterparts via text-guided lesion editing.

Time Range

null

Freshness

Last updated 2026-04-12 19:39:38; freshness should be verified.

Geography

null