# MedForge: Interpretable Medical Deepfake Detection via Forgery-aware Reasoning

Source: [MedForge: Interpretable Medical Deepfake Detection via Forgery-aware Reasoning](<https://www.sota2.com/research/paper/69bcd4770636e5bf4afc297d>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-05T05:00:12.194Z

Authors: Zhihui Chen, Kai He, Qingyuan Lei, Bin Pu, Jian Zhang, Yuling Xu, Mengling Feng

Published: 2026-03-19T07:38:11.000Z

## Abstract

Text-guided image editors can now manipulate authentic medical scans with high fidelity, enabling lesion implantation/removal that threatens clinical trust and safety. Existing defenses are inadequate for healthcare. Medical detectors are largely black-box, while MLLM-based explainers are typically post-hoc, lack medical expertise, and may hallucinate evidence on ambiguous cases. We present MedForge, a data-and-method solution for pre-hoc, evidence-grounded medical forgery detection. We introduce MedForge-90K, a large-scale benchmark of realistic lesion edits across 19 pathologies with expert-guided reasoning supervision via doctor inspection guidelines and gold edit locations. Building on it, MedForge-Reasoner performs localize-then-analyze reasoning, predicting suspicious regions before producing a verdict, and is further aligned with Forgery-aware GSPO to strengthen grounding and reduce hallucinations. Experiments demonstrate state-of-the-art detection accuracy and trustworthy, expert-aligned explanations.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Forgery Detection on MedForge Real Forgery (In-Domain) 90K](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-real-forgery-in-domain-90k.md>): Forgery Detection; MedForge Real Forgery (In-Domain) 90K; Accuracy: 99.24.

- [Forgery Detection on MedForge-90K Real Forgery Cross-Forgery](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-real-forgery-cross-forgery.md>): Forgery Detection; MedForge-90K Real Forgery Cross-Forgery; Accuracy: 95.24.

- [Forgery Detection on MedForge Real Forgery Cross-Model 90K](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-real-forgery-cross-model-90k.md>): Forgery Detection; MedForge Real Forgery Cross-Model 90K; Accuracy: 92.86.

- [Forgery Detection on MedForge-90K Implant Forgery Cross-Forgery](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-implant-forgery-cross-forgery.md>): Forgery Detection; MedForge-90K Implant Forgery Cross-Forgery; Accuracy: 93.39.

- [Forgery Detection on MedForge-90K Implant Forgery Cross-Model](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-implant-forgery-cross-model.md>): Forgery Detection; MedForge-90K Implant Forgery Cross-Model; Accuracy: 94.86.

- [Forgery Detection on MedForge-90K Remove Forgery (In-Domain)](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-remove-forgery-in-domain.md>): Forgery Detection; MedForge-90K Remove Forgery (In-Domain); Accuracy: 99.21.

- [Forgery Detection on MedForge-90K Remove Forgery Cross-Forgery](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-remove-forgery-cross-forgery.md>): Forgery Detection; MedForge-90K Remove Forgery Cross-Forgery; Accuracy: 99.15.

- [Forgery Detection on MedForge-90K Remove Forgery Cross-Model](<https://www.sota2.com/research/sota/forgery-detection-on-medforge-90k-remove-forgery-cross-model.md>): Forgery Detection; MedForge-90K Remove Forgery Cross-Model; Accuracy: 94.09.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.18577>)

[PDF](<https://arxiv.org/pdf/2603.18577.pdf>)