#United States #San Francisco #AI or Not #Deepfake X-Ray #Radiology Study

AI or Not Achieves Breakthrough in Deepfake X-Ray Detection

In a landmark development for the field of medical imaging, AI or Not has demonstrated significant advancements in detecting deepfake X-rays. A recent independent evaluation revealed that the company achieved an impressive 100% detection rate for synthetic X-rays, coupled with an overall accuracy of 95%. These results outshine both the performance of qualified radiologists and leading multimodal large language models (LLMs) in a recent study published in the journal Radiology.

The study, titled "The Rise of Deepfake Medical Imaging," was conducted by researchers from the Icahn School of Medicine at Mount Sinai, who provided a curated dataset comprising authentic and AI-generated radiographs for review. During blinded tests, AI or Not's detection technology proved effective, correctly identifying all synthetic X-rays while maintaining a false positive rate of just 7.8% on genuine images. This level of precision positions AI or Not's technology as a powerful tool for enhancing the integrity of medical imaging and safeguarding patient safety.

The Risk of Deepfake Medical Imaging

As the prevalence of deepfake technology grows, the implications for medical imaging become increasingly critical. The researchers highlighted several risks associated with synthetic medical images:

• - Insurance Fraud: Artificial imaging can support the fabrication of false claims.
• - Legal Evidence: Manipulated images may distort diagnoses introduced into legal litigation or disability cases.
• - Research Integrity: Synthetic visuals could contaminate training datasets and published studies, leading to compromised research outcomes.
• - Patient Safety: Altered scans can significantly impact treatment decisions based on inaccurate information.

Alarmingly, the findings revealed no correlation between a radiologist's experience level and their ability to successfully detect deepfake images. The study cited that radiologists achieved only 41% detection accuracy without prior warning about the presence of AI-generated images, which only improved to 75% when prompted to be vigilant. This highlights a concerning gap in expert readiness against this emerging form of medical fraud.

A Comparative Analysis: AI or Not vs. Radiologists and LLMs

In the same study, four leading LLMs—GPT-4o, GPT-5, Gemini 2.5 Pro, and Llama 4 Maverick—exhibited variable accuracy rates ranging from 57% to 85%. This fluctuation underscores AI or Not's reliability compared to both human and conventional AI counterparts. According to Anatoly Kvitnitsky, CEO and Founder of AI or Not, the findings emphasize the need for a multifaceted approach to combating deepfake threats. He stated, "No single layer solves this. It takes clinician training, watermarking, dataset governance, and detection working together."

The Path Forward for Medical Imaging

AI or Not's results showcase the importance of targeted detection mechanisms that can complement existing medical practices. By validating the efficacy of specialized detection technology, the study encourages further dialogue on integrating safeguards such as clinician training and encoding authenticating measures like watermarks.

AI or Not's advanced detection API is designed to empower various stakeholders—be it developers, businesses, or healthcare systems—by embedding synthetic media detection directly into their products and workflows. The company has positioned itself as a leader in AI detection, delivering industry-leading accuracy rates of 98.9% across various media formats.

For a detailed understanding of the benchmarking methodology employed and further statistics, the complete results from this pivotal study can be accessed for review. As the challenges around deepfake medical imaging continue to evolve, staying abreast of technological developments such as AI or Not's capabilities will be crucial for safeguarding the future of medical diagnostics and patient care.