Uploaded April 2026 | Updated November 2026, 3 days ago
Qian Tao (TU Delft): "Deformable Image Registration: The Old and New Paradigms."
Abstract: Deformable image registration is a cornerstone of quantitative biomedical imaging, enabling the integration of anatomical and functional information across subjects, modalities, and timepoints. Traditionally, registration has relied on iterative optimization techniques that seek the best spatial alignment. These classical methods are robust and mathematically grounded, but they can be computationally intensive and challenging to scale to large problems.
In recent years, a new paradigm has emerged: deep learning–based registration, which leverages neural network parameterization to learn an implicit optimization strategy from data. By training neural networks to predict deformation fields, these approaches achieve significant speed-ups and open new possibilities for real-time and high-throughput applications.
Deformable image registration using deep learning remains an evolving research frontier, with open questions and ongoing debates. In this talk, I will introduce the classical and modern paradigms of deformable image registration, compare their strengths and limitations, and discuss their real-world applications in quantitative biomedical imaging.
Qian Tao (TU Delft): "Deformable Image Registration: The Old and New Paradigms."
Abstract: Deformable image registration is a cornerstone of quantitative biomedical imaging, enabling the integration of anatomical and functional information across subjects, modalities, and timepoints. Traditionally, registration has relied on iterative optimization techniques that seek the best spatial alignment. These classical methods are robust and mathematically grounded, but they can be computationally intensive and challenging to scale to large problems.
In recent years, a new paradigm has emerged: deep learning–based registration, which leverages neural network parameterization to learn an implicit optimization strategy from data. By training neural networks to predict deformation fields, these approaches achieve significant speed-ups and open new possibilities for real-time and high-throughput applications.
Deformable image registration using deep learning remains an evolving research frontier, with open questions and ongoing debates. In this talk, I will introduce the classical and modern paradigms of deformable image registration, compare their strengths and limitations, and discuss their real-world applications in quantitative biomedical imaging.

