RCS - Dynex Quantum Tech. - Peer-Reviewed Study Shows Quantum-Improved AI

Dynex Capital

18/09/2026 07:00

RNS Number : 3602V
  
Dynex Quantum Technologies AG
  
18 September 2026
  
 


Dynex Peer-Reviewed Study Demonstrates Quantum-Driven Optimisation Improves Healthcare AI
Baar, Switzerland - September 2026 Dynex Quantum Technologies AG (www.dynex.co), an award-winning global technology leader and solutions provider headquartered in Switzerland focused on the development and deployment of end-to-end quantum-driven solutions at industrial scale, powered by its proprietary Quantum-as-a-Service (QaaS) technology, today announced the publication of a peer-reviewed study in the Journal of Medical Internet Research (JMIR) evaluating the use of quantum-enhanced artificial intelligence for medical text anonymisation.
Read the full peer-reviewed study in the Journal of Medical Internet Research (JMIR): https://www.jmir.org/2026/1/e91513
The study is also now listed on PubMed, operated by the U.S. National Library of Medicine (NLM), the world's largest and most comprehensive biomedical library and a global leader in biomedical informatics and computational health data science.
Read the listing on PubMed, operated by the U.S. National Library of Medicine (NLM): https://pubmed.ncbi.nlm.nih.gov/42735382/
This study was conducted in partnership with two leading institutions: InCor, one of the largest heart transplant hospitals in the world and the No. 1-ranked hospital in Latin America for cardiology, and the University of São Paulo Medical School HCFMUSP, the largest university hospital complex in Latin America. It builds on a partnership that has been in place since 2024.
Medical records contain enormous amounts of information with potential value for clinical research and healthcare innovation. However, much of this information exists as unstructured text containing patient names, dates, identifiers, organisations and geographic locations. Before such data can be used for secondary research, sensitive information must be reliably identified and anonymised while preserving as much clinically useful information as possible.
The study evaluated Dynex's quantum-enhanced machine learning architecture against standalone large language models (LLMs) in the anonymisation of sensitive clinical data and found that the quantum-enhanced architecture significantly improves medical text anonymisation, reducing false positives for patient names by 95% while maintaining high sensitivity.
Researchers constructed a gold standard dataset of 1,000 Portuguese outpatient clinical notes, manually annotated by five domain experts across five categories of protected information. Four anonymisation approaches were evaluated, including two standalone Llama models and two quantum-enhanced Dynex-QML architectures. Performance was evaluated on a held-out test set of 500 clinical notes.
The results showed that the Dynex-QML-70B model achieved the highest overall performance, with a macro-F1 score of 0.855, compared with 0.726 for the standalone Llama-3.3-70B model. The quantum-enhanced architecture also achieved a macro-precision of 0.794 compared with 0.636 for Llama-3.3-70B, while maintaining macro-sensitivity of 0.971.
One of the most significant improvements was observed in the identification of patient names. The standalone Llama-3.3-70B model generated 202 false positives, while Dynex-QML-70B reduced this to just 10 - a reduction of approximately 95%. Precision for patient names increased from 0.624 to 0.971, while sensitivity remained high at 0.951.
Similar improvements in precision were recorded for dates and geographic locations. The study concludes that the quantum optimisation layer acts as a corrective mechanism that can constrain an LLM's tendency to over-identify borderline entities, reducing unnecessary removal of information while preserving high recall.
The Dynex-QML-70B architecture also recorded a slightly lower total processing time than the standalone 70B model, at 7.97 ± 2.87 seconds compared with 8.52 ± 3.32 seconds per text, demonstrating that the improvement in anonymisation quality did not come at the expense of processing performance in the 70B configuration.
Quantum-Driven Optimisation for Healthcare AI
Rather than relying solely on the sequential token generation used by conventional LLMs, Dynex-QML integrates the LLM with quantum optimisation. The architecture reformulates the final stage of the anonymisation task as a Quadratic Unconstrained Binary Optimisation (QUBO) problem, enabling decisions across the output to be optimised globally rather than only sequentially.
This approach is designed to preserve the contextual capabilities of the underlying LLM while applying quantum-driven optimisation to improve the final result. In the study, this was particularly effective in reducing over-identification - instances where clinically useful information is incorrectly classified as sensitive and removed.
This balance between privacy and data utility is important for healthcare organisations operating under data protection frameworks including the U.S. Health Insurance Portability and Accountability Act (HIPAA), the European General Data Protection Regulation (GDPR) and Brazil's Lei Geral de Proteção de Dados Pessoais (LGPD). The paper notes that false negatives can create privacy and compliance risks, while excessive false positives can reduce the usefulness of anonymised datasets for research.
According to the authors, the research represents one of the first practical applications of quantum-enhanced methods to a clinical natural language processing task.
Daniela Herrmann, Founder and CEO of Dynex, said:
"This peer-reviewed study demonstrates what quantum-driven technology can deliver when applied to a real and highly relevant healthcare challenge. 'Quantum is now' means: applying it to real-world challenges where accuracy, privacy and scalability are imperative and can save lives. Healthcare is one of the areas where that combination has enormous potential."
What This Means for the Larger AI and Healthcare Industry
The significance of the study extends beyond medical text anonymisation. It demonstrates how quantum-driven optimisation can improve artificial intelligence systems that organisations already use, rather than requiring them to replace their existing AI infrastructure. Dynex-QML builds on an established Llama model by adding a quantum-driven optimisation layer, while Dynex's Quantum Bridge enables integration with widely used deep-learning frameworks including PyTorch and TensorFlow with minimal code modifications. This points towards a hybrid future in which quantum capabilities can be integrated directly into existing enterprise AI workflows.
For healthcare, this approach could help address one of the fundamental challenges of using AI in high-stakes environments: models can produce outputs that appear confident while still being wrong. In medical anonymisation, this can result in sensitive information being missed or clinically useful information being unnecessarily removed. The study found that the quantum optimisation layer acted as a corrective mechanism, constraining the LLM's tendency to over-identify borderline entities. For patient names, false positives fell from 202 with Llama-3.3-70B to just 10 with Dynex-QML-70B, while maintaining high sensitivity.
The implications extend to the enormous volumes of clinical information contained in electronic health records (EHRs). These datasets have significant potential for medical research and healthcare innovation, but their secondary use is constrained by the need to protect patient privacy. More accurate anonymisation could help healthcare organisations preserve more clinically valuable information while reliably protecting sensitive patient data, making datasets more useful for medical research, drug development, population health analysis and the development of healthcare AI. The paper specifically identifies this balance between privacy protection and data utility as critical to enabling the secure secondary use of healthcare data.
Ultimately, improving the ability to use clinical data safely is not simply a data-management challenge. Greater access to high-quality, privacy-protected healthcare data can support research, accelerate medical innovation and improve the AI systems used to understand disease and support healthcare delivery. By helping unlock more of this information without compromising patient privacy, quantum-enhanced AI has the potential to contribute to better research, better healthcare and, ultimately, save lives.
 
About Dynex
Dynex (www.dynex.co) is an award-winning global technology solutions provider focused on the development and deployment of end-to-end quantum-driven solutions at industrial scale, powered by its proprietary Quantum-as-a-Service (QaaS) technology. The technology is embedded in a unique cloud-based, qubit-agnostic platform architecture, which gives customers access to Dynex's emulation technology supporting up to 1,000,000 algorithmic qubits or Dynex's quantum-driven neuromorphic Apollo chip at room temperature, designed to operate with up to 10,000 physical p-qubits, and Zeus, a breakthrough full-quantum processor prototype built on a nitrogen-vacancy (NV) nanodiamond spin-qubit architecture, operating at room temperature.
Dynex delivers customised, production-grade solutions to global enterprises through seamless API integration and access to the platform, leveraging quantum-driven computational capabilities within established technology stacks. By combining scalable infrastructure with domain-specific implementation, Dynex advances practical accessibility and cost-efficient adoption of next-generation computational paradigms.
The transition towards mainstream adoption with Dynex's quantum-driven computing has been underpinned by the ethical integrity of Dynex. This has been led and safeguarded by Dynex Moonshots (www.dynexmoonshots.com), Family Office of Dynex, steward of the Ethical Committee and investor in pioneers shaping the future across health, society, nature and space by fostering significant breakthroughs.
For the Advanced Quantum Education initiative, Dynex Moonshots acts as a founding investor of the Q platform (www.qedu.co), which, through its programmes, is designed to cultivate emerging talent, foster interdisciplinary research and promote responsible technological progress.
About JMIR
The Journal of Medical Internet Research (JMIR) is a pioneering open-access eHealth journal and the flagship journal of JMIR Publications, ranked No. 1 on Google Scholar in Medical Informatics. JMIR publishes peer-reviewed research on emerging technologies and digital health innovations, including medical devices, artificial intelligence, telehealth, health informatics and technology-enabled clinical care. The journal is indexed in major international databases including MEDLINE, PubMed, Scopus and Web of Science. In 2025, JMIR received an Impact Factor of 8.2, ranking in the first quartile (Q1) in both Medical Informatics and Health Care Sciences & Services, as well as a Scopus CiteScore of 10.4.
About PubMed
PubMed is one of the world's most widely used resources for searching biomedical and life sciences literature, providing access to more than 40 million citations and abstracts spanning biomedicine, health, life sciences, behavioural sciences, chemical sciences, bioengineering and related disciplines. Available online to the public since 1996, PubMed is developed and maintained by the National Center for Biotechnology Information (NCBI) at the U.S. National Library of Medicine (NLM), part of the National Institutes of Health (NIH). NLM is the world's largest biomedical library, producing electronic information resources that are searched billions of times each year by millions of people around the world.
Media Contact
 Dynex Quantum Technologies AG
 Zoë van Horen
 media@dynex.ch  | www.dynex.co




This information is provided by Reach, the non-regulatory press release distribution service of RNS, part of the London Stock Exchange. Terms and conditions relating to the use and distribution of this information may apply. For further information, please contact rns@lseg.com or visit www.rns.com.

RNS may use your IP address to confirm compliance with the terms and conditions, to analyse how you engage with the information contained in this communication, and to share such analysis on an anonymised basis with others as part of our commercial services. For further information about how RNS and the London Stock Exchange use the personal data you provide us, please see our Privacy Policy.
 
  END
 
 
NRAFIFVSAVITLIR