AI-driven cardiovascular imaging
Abstract
Artificial intelligence (AI) is ushering in a new era of medical imaging, transforming how clinicians acquire, interpret, and utilize image-based information. Traditionally, imaging modalities such as MRI, CT, and ultrasound have provided invaluable visual insights into anatomy and pathology. Yet the human eye perceives only a fraction of the complex data encoded within these images. AI changes this dynamic by greatly expanding the information that can be extracted from a single examination and by enabling high-throughput, automated analysis of quantitative features often invisible to conventional interpretation. These features can then be integrated into diagnostic, predictive, and prognostic models that support more precise clinical decision-making.
This shift mirrors developments in other biomedical fields, such as genomics, proteomics, and metabolomics, where large-scale data and computational advances have unlocked new biological understanding and therapeutic potential. In medical imaging, this data-driven approach is sometimes known as radiomics. Radiomics moves imaging beyond simple visual assessment, offering detailed, regional, and even whole-body information about tissue structure, function, and pathology. Such advances promise to improve patient outcomes through earlier disease detection, fewer diagnostic errors, refined risk stratification, personalized treatment planning, and more efficient clinical workflows.
Comprehensive assessment of the function of the heart and blood vessels is particularly challenging, as these are complex three-dimensional structures that move both actively and passively throughout the cardiac cycle. In cardiovascular imaging, not only the anatomy but also the dynamic function of the heart and vessels must be evaluated to obtain a complete picture of cardiovascular health. This requires capturing and analyzing large, multidimensional datasets that reflect motion, blood flow, and tissue characteristics over time. Consequently, cardiovascular imaging generates vast amounts of complex data, which places significant demands on image acquisition, post-processing, and interpretation workflows. Efficiently managing and analyzing this information remains a major challenge, underscoring the need for advanced computational methods and automation to support accurate and timely diagnosis.
In this project, we aim to harness the power of AI to accelerate and enhance cardiovascular imaging.
Specifically, our objectives are threefold: first, to develop advanced AI-driven undersampling strategies that allow faster acquisition of high-quality imaging data; second, to enhance image fidelity by optimizing artifact correction and employing super-resolution techniques; and third, to improve overall workflow efficiency through automatic segmentation and data analysis algorithms.
For this purpose, we will train 3D and potentially 4D neural networks using synthetic and/or anonymized imaging data derived from clinical research studies conducted under appropriate ethical approvals. The trained algorithms will be evaluated on local GPU systems capable of securely handling sensitive data. No sensitive or identifiable data will be processed or stored on the Berzelius supercomputing infrastructure.
By fostering interdisciplinary collaboration, developing robust algorithms, and integrating them into clinical practice, this area has the potential to transform modern healthcare and set new standards for diagnostic excellence.
The PI and co-PI are both SciLifeLab Group Leaders.
