UNN scientists integrate quantum technologies and artificial intelligence to analyse MRI scans
The UNN Institute of Information Technology, Mathematics and Mechanics. "Classical deep models A hybrid quantum-classical neural network trained by UNN researchers has recognised brain tumours on MRI images with an accuracy of 98%. Brain images were obtained from open databases. This is one of the first successful attempts to combine the potential of standard neural networks and quantum circuits. With the use of the quantum layer, scientists succeeded in obtaining an in-depth analysis of images using fewer parameters than required by a classical neural network. This means that a hybrid neural network can be a more effective tool for analysing medical data.
"Our approach enhances classification accuracy through neural network architecture optimisation," noted Marina Bastrakova, Senior Researcher from the Laboratory "Artificial Intelligence in Preventive Medicine" at the Research Centre for Artificial Intelligence of contain billions of trainable parameters and require significant computing resources, whereas our quantum layer works with a highly compressed feature representation. At the same time, it is the interaction of the classical and quantum parts of the model that is crucial."
The hybrid quantum-classical neural network also confidently diagnosed MRI scans with noise and defects. For instance, it maintained 96.6% accuracy with strong Gaussian noise, compared to 92.2% for classical networks. This is particularly significant for medical imaging, where image quality can be affected by various factors, including patient positioning, equipment specifications and scanning protocols.
"We have seen that the quantum layer is not merely an additional element of the architecture: its design significantly impacts the outcome. In particular, models with more advanced repeating quantum blocks in our experiments better preserved accuracy under image distortions, indicating that quantum feature transformations play a key role in creating a stable data representation. At the same time,further research and testing on real quantum hardware are needed to confirm this effect," Marina Bastrakova said.
"It is important to note that any neural network solutions do not replace a diagnostic doctor. They act as assistants in the initial diagnosis, help the doctor not to miss important signs and give a signal to pay attention to critical values. The probability of error decreases and the specialist's diagnosis becomes more reliable,"emphasisedDr. Mikhail Ivanchenko, Chief Researcher at the Laboratory of Artificial Intelligence in Preventive Medicine.
Moving forward, the researchers plan to test the approach on new medical datasets using Russian quantum computers. Collaboration with specialised institutions and research centres aims to integrate these hybrid methods into practical medical applications. The developers intend to train the neural network fordescribing images in more detail and expand image classification.
"The development of quantum technologies and artificial intelligence is a strategic priority forour University. Our scientists have developed an advanced tool for medical diagnostics that allows accurate analysis of complex data and opens new avenues for early detection of pathologies and improving healthcare quality in general. We aim to further integrate these algorithms into clinical practice and adapt them for Russian quantum computers to address key challenges in the field of technological leadership," stated Oleg Trofimov, Rector of Lobachevsky University.
This research was conducted by scientists of the Artificial Intelligence Research Centreat theUNNInstitute of Information Technology, Mathematics and Mechanics under the "Priority 2030" federal programme and published in The European Physical Journal Special Topics.
