Quantum Diagnosis: Russian Scientists Teach AI to Detect Tumors on MRI
Russian scientists have developed a neural network that combines artificial intelligence with a quantum layer. The model can detect brain tumors on MRI scans. In the future, the technology could become an additional tool for physicians.
It is no longer surprising that AI is helping physicians analyze medical images. Neural networks can identify suspicious areas on MRI, CT and X-ray images, classify pathologies and flag regions that require closer attention from a clinician. But these systems face a challenge: medical images are not always high quality. MRI quality can be affected by patient micromovements, breathing, equipment characteristics and other factors that introduce visual noise. The poorer the image, the greater the loss of accuracy.
Scientists at the Research Center for Artificial Intelligence at Lobachevsky University decided to test whether a quantum approach could help address this problem. Their system is built around a conventional neural network augmented with a specialized quantum layer. It works with a highly compressed set of features, allowing the image to be analyzed using fewer parameters than classical models require.
In other words, the researchers did more than add another computational module to a medical AI system. They tested whether a quantum architecture could make the neural network more robust when working with challenging input data.
The results were notable. On standard images, the hybrid model achieved accuracy of up to 98%. Under heavy noise, one of the quantum-classical models achieved 96.6% accuracy, compared with 92.2% for a classical neural network. For medical diagnosis, that is a meaningful difference. The findings were published in the international journal The European Physical Journal Special Topics.
How It Works?
The system first analyzes an MRI scan and flags features that could indicate a pathology. The radiologist then gets an additional point of reference, but evaluates the scan independently, compares the findings with the patient's medical history and makes the clinical decision.
For physicians, the technology is primarily a way to reduce the amount of routine image review. For patients, it adds another layer of screening that could help prevent a suspicious area or an emerging inflammatory lesion from being overlooked.
Research at Scale
The current work builds on a research program that scientists at Lobachevsky University have been developing for several years. In 2024, researchers at the university, working with the Russian Quantum Center, proposed a superconducting neuromorphic processor architecture. The system was designed to operate in both classical and quantum modes.
A year later, the university reported on AI research aimed at diagnosing brain tumors from genetic markers. Machine learning was used to classify gliomas, a type of brain tumor, predict patient survival and analyze gene activity.
The Next Step: Real-World Data
The technology is still at the research stage. The team used images from open medical databases to train and validate the model. The next step is to test it on new datasets. This is necessary to determine how well the model performs on images it was not trained on. The researchers then plan to test the approach using Russia's quantum-computing systems. Participation by specialized medical institutions and research centers is also under discussion. In the foreseeable future, the technology will be tested on actual patients.
What Researchers Want Next
The developers want to teach the neural network not only to detect tumors but also to provide a more detailed description of MRI scans. They also plan to expand the range of diseases the system can recognize.
More broadly, researchers in Nizhny Novgorod are already showing how Russian IT development can move beyond conventional neural networks. By combining AI with quantum technologies, they are expanding what the system can do and making it a potentially more useful tool for physicians. The approach could improve the chances of detecting a pathology when poor image quality might undermine a conventional algorithm. That, in turn, could improve the chances of a healthier outcome for individual patients.
