Publications by authors named "S R Arridge"
Denoising diffusion models have emerged as the go-to generative framework for solving inverse problems in imaging. A critical concern regarding these models is their performance on out-of-distribution tasks, which remains an under-explored challenge. Using a diffusion model on an out-of-distribution dataset, realistic reconstructions can be generated, but with hallucinating image features that are uniquely present in the training dataset.
View Article and Find Full Text PDFComplex biological systems undergo sudden transitions in their state, which are often preceded by a critical slowing down of dynamics. This results in longer recovery times as systems approach transitions, quantified as an increase in measures such as the autocorrelation and variance. In this study, we analysed paediatric patients in intensive care for whom mechanical ventilation was discontinued through removal of the endotracheal tube (extubation).
View Article and Find Full Text PDFA fast all-optical 3D photoacoustic scanner for clinical vascular imaging.
Nat Biomed Eng
September 2024
Article Synopsis
- The study discusses advancements in photoacoustic tomography (PAT) for visualizing microvascular structures, essential for assessing conditions like diabetes and inflammatory skin diseases.
- By improving the scanner's speed from minutes to mere seconds or milliseconds, the authors enable better visualization with less motion-related interference and detailed 3D imaging.
- These enhancements allow for the exploration of microvascular changes in various medical conditions, suggesting potential applications in fields such as cardiovascular medicine, oncology, dermatology, and rheumatology.
To develop and assess a deep learning (DL) pipeline to learn dynamic MR image reconstruction from publicly available natural videos (Inter4K). Learning was performed for a range of DL architectures (VarNet, 3D UNet, FastDVDNet) and corresponding sampling patterns (Cartesian, radial, spiral) either from true multi-coil cardiac MR data (N = 692) or from synthetic MR data simulated from Inter4K natural videos (N = 588). Real-time undersampled dynamic MR images were reconstructed using DL networks trained with cardiac data and natural videos, and compressed sensing (CS).
View Article and Find Full Text PDFArticle Synopsis
- The interaction between the left and right ventricles in the cardiovascular system is complicated and influenced by structures like the septum and pericardium, leading to complex nonlinear equations in modeling.
- Existing computational models often oversimplify these interactions or ignore them altogether, prompting a search for better methods.
- The proposed approach utilizes a hybrid neural ordinary differential equation model that incorporates neural networks to simulate ventricular interactions, demonstrating strong predictive abilities even with added noise in the data.