97 papers
Radiology and imaging serve as the eyes of modern medicine, allowing doctors to peer inside the human body without making a single incision. This rapidly evolving field uses technologies like X-rays, MRI scans, and ultrasound to detect diseases, guide treatments, and monitor patient recovery. As new research emerges, these visual tools become increasingly sophisticated, offering deeper insights into conditions ranging from broken bones to complex neurological disorders.
At Gist.Science, we bridge the gap between raw scientific data and public understanding by processing every new preprint in this category from medRxiv. Our team translates these complex studies into both plain-language overviews and detailed technical summaries, ensuring that breakthroughs in medical imaging are accessible to everyone, from students to specialists. Below are the latest papers in radiology and imaging, ready for you to explore.
This paper demonstrates that the common "slopes-first" approach for estimating correlations between subject-specific rates of change in longitudinal studies with dropout is biased when cross-covariance varies over time, and proposes an unbiased alternative that estimates the joint covariance matrix prior to applying contrasts.
This multi-centre study demonstrates that while the MELD-graph tool maintains comparable sensitivity for detecting focal cortical dysplasia at 7T MRI as it does at 3T, its specificity is significantly lower, necessitating cautious interpretation and highlighting the benefits of harmonisation despite limited gains from 7T-specific pipeline adaptations.
This study addresses the lack of independent comparisons in the field by benchmarking eight open-source automated thigh muscle MRI segmentation algorithms across diverse datasets, revealing that while domain-specific models generally outperform foundation models, significant performance variations and reduced accuracy on pathological cases highlight critical trade-offs between accuracy, generalizability, and clinical usability.
Although the study developed a highly accurate deep learning model for vertebral artery segmentation on CTA, a reader study revealed that its use actually decreased diagnostic accuracy for vertebral artery dissection, indicating a need for further validation before clinical implementation.
This paper presents a sparsely supervised AI framework that leverages an ultrasound foundation model with Gaussian soft temporal scoring to automatically select optimal frames for ONSD measurement in ultrasound videos, significantly outperforming existing baselines and enabling reliable intracranial pressure assessment in low-resource settings.
In cognitively unimpaired individuals from the Baltimore Longitudinal Study of Aging, plasma p-tau217 demonstrated the strongest association with both subsequent Alzheimer's disease-related dementia and longitudinal changes in AD-specific brain atrophy patterns compared to other plasma biomarkers.
This study demonstrates that visible/near-infrared hyperspectral imaging, particularly when analyzed using full-spectrum machine learning algorithms, can accurately and rapidly classify cellulitis, non-cellulitis edema, and healthy lower-leg tissue in emergency department patients across diverse skin pigmentation types.
This proof-of-concept study identified a set of reproducible and repeatable radiomics features from gadoxetic acid-enhanced MRI that significantly correlate with quantitative liver function indices, establishing a foundation for future liver function-related radiomics research.
This phantom study evaluates the accuracy of CTA, 3DRA, and 2D DSA in measuring intracranial vessel diameters against physical ground truth, revealing that while all modalities generally measure within 0.5 mm of true dimensions, they exhibit consistent biases that vary by vessel size, contrast concentration, and detection algorithm, with maximum-gradient edge detection proving superior to FWHM and fixed-HU methods.
This paper introduces Residual Endpoint Flow Matching (REFM), a single-step deep learning method that leverages the inherent anatomical similarity between low-dose and standard-dose CT images to achieve real-time, high-quality denoising comparable to iterative diffusion models while drastically reducing computational costs.