Application of machine learning in neuroradiology (literature review)

Abstract

Neuroradiology is a branch of medical imaging focused on diagnosing diseases of the brain, spine, and nervous system using radiological methods (computed tomography, magnetic resonance imaging, positron emission tomography, etc.). In recent decades, artificial intelligence – particularly machine learning (ML) – has rapidly advanced and is increasingly applied for automated analysis of medical data in interventional neuroradiology. In neuroradiology, this approach has the potential to increase diagnostic accuracy and speed, help «see the invisible» on scans, and support clinical decision-making. This review highlights the history of ML applications in neuroradiology, the main machine learning methods used, key research areas (stroke, tumors, neurodegenerative diseases, etc.), and recent achievements with an emphasis on clinical implementation.A search of scientific publications published primarily over the last five years (2019–2024), along with key earlier works, was conducted in PubMed, IEEE Xplore, arXiv, and Google Scholar. The review includes original research, meta-analyses, and systematic reviews devoted to the application of machine learning and deep learning methods in neuroradiology. Inclusion criteria were: focus on brain pathologies (stroke, tumors, neurodegenerative diseases), use of standard imaging modalities (CT, MRI, PET), and availability of quantitative algorithm performance metrics (e.g., accuracy, sensitivity, specificity).Conclusions. Machine learning is a powerful tool that complements traditional neuroradiology and enables rapid, high-precision processing of large volumes of imaging data. Having evolved from simple decision-support systems to complex neural networks, ML algorithms have reached expert-level performance in diagnosing stroke, analyzing tumors, and detecting neurodegenerative diseases. A transition from research to clinical deployment is underway, as evidenced by the certification of the first AI tools. The key paradigm of the future is not replacement but synergy between clinician and algorithm: automation of routine tasks will allow neuroradiologists to focus on complex cases and patient-centered care. For full-scale integration, issues of algorithmic reliability and ethics must be addressed. However, it is already clear that ML is a transformational technology that will become a foundation of neuroradiological practice in the coming years.

Ähnliche Arbeiten