80 papers
This paper introduces DBT-2026, a freely available, de-identified dataset comprising 558 digital breast tomosynthesis exams with expert annotations and clinical reports from patients with BI-RADS scores of 0, 1, or 2, designed to facilitate non-commercial research in breast cancer imaging.
This study introduces and validates BUDAPEST, a novel Bayesian adaptive algorithm with an open-source GUI that enables rapid, reliable, and user-controlled motor threshold estimation for transcranial magnetic stimulation using significantly fewer pulses than conventional methods.
This study demonstrates that 3D-DXA cortical and trabecular parameters measured using GE Healthcare Prodigy and iDXA densitometers show excellent agreement and comparable short-term precision, indicating that no adjustments are necessary when using 3D-Shaper software across these different device models.
The Holder-Optimized Elastography (HOE) method enhances the non-invasive detection of breast cancer-related lymphedema by stabilizing ultrasound probes to visualize fluid-filled lymphatic obstructions as High-Velocity Areas, offering a promising adjunct for monitoring treatment response despite current limitations in sensitivity and specificity.
This paper identifies and resolves the "NLP-to-Expert Gap" in chest X-ray AI by demonstrating that models optimized on automated report labels overfit to labeling errors, whereas superior diagnostic performance is achieved by using expert labels as a validation compass, employing early stopping to prevent memorization, and relying on frozen ImageNet features with regularization rather than direct metric optimization.
This study introduces a scalable automated deep-learning framework for segmenting 20 individual hip and thigh muscles on MRI, revealing distinct sex-specific patterns of anatomical heterogeneity, longitudinal decline, and metabolic risk in the UK Biobank that are obscured by traditional compartment-level measures.
This paper presents a deep learning-based segmentation model and associated open-source tools for the non-invasive, automatic quantification of 19 metabolically relevant adipose tissue compartments and ectopic fat deposits from whole-body Dixon MRI.
This study reveals that chronic sports-related mild traumatic brain injury in young adults leads to sex-differentiated structural brain alterations, where increased cortical thickness in the superior parietal lobule correlates with inattentive symptoms in males, while higher white matter integrity in specific tracts correlates with reduced hyperactive/impulsive symptoms in females.
This study demonstrates that a hierarchical 3D deep neural network can accurately predict renal compartment volumes from routine non-contrast MRI by transferring knowledge from contrast-enhanced arterial-phase scans, although it exhibits systematic biases in cortical and medullary segmentation and struggles with surface area estimation.
This study demonstrates that a fully automated CT-based segmentation method using convolutional neural networks can reliably quantify renal subcompartment volumes, providing reproducible biomarkers that correlate with both global and split renal function in living kidney donor candidates.
This pilot study demonstrates that an LLM-orchestrated AI agent can successfully automate the end-to-end workflow of PET/CT interpretation from raw DICOM data to structured reporting in 170 lung cancer patients, achieving perfect primary tumor detection while revealing systematic limitations in nodal and metastatic assessment that necessitate continued expert oversight.
This paper benchmarks transfer learning strategies for dense breast tissue segmentation on small datasets, demonstrating that CNNs with full fine-tuning, multi-view self-supervised pre-training, and hybrid loss functions outperform transformer-based models and parameter-efficient updates to achieve optimal accuracy and efficiency for annotation-limited mammography workflows.
This study introduces a robust, data-driven framework that identifies five distinct white matter hyperintensity spatial subtypes across large cohorts, revealing their unique associations with vascular risk factors and demonstrating that regional lesion patterns offer superior predictive value for future disease progression compared to total lesion burden alone.
This study demonstrates that a simplified, muscle-normalized dynamic contrast-enhanced MRI method reliably distinguishes carotid plaque cores from remote vessel walls with high reproducibility over six months, although the metric remained unaffected by low-dose colchicine treatment.
This study demonstrates that while highly curated, smaller datasets can achieve equivalent 3D segmentation performance to much larger mixed-curation datasets, the latter offers superior generalizability and local improvements, indicating that the optimal balance between data quality and quantity depends on specific training goals.
This study demonstrates that optimized 7T heavily T2-weighted MRI enables the detection and quantification of perivascular spaces at the gray-white matter interface in healthy individuals, revealing that while cortical PVS density is lower than in white matter, leukocortical segments constitute a significant portion of total PVS volume.
This study demonstrates that O-17-labeled saline eye drops can non-invasively visualize the distribution of aqueous humor into the cerebrospinal fluid surrounding the optic nerve in living humans, supporting the existence of an extra-trabecular outflow pathway potentially linked to the ocular glymphatic system.
This study evaluates and compares radiomics, contrastive learning, and convolutional deep learning architectures for predicting EGFR mutation status in non-small cell lung cancer using CT images, finding that a hybrid model integrating radiomic and clinical features achieved the highest performance (AUC 0.790) while also discussing the clinical translation challenges and potential utility of radiogenomics.
This study compares three assessment strategies for deep-learning-based super-resolution on small human brain MRI datasets and concludes that k-fold cross-validation offers the most favorable balance between accuracy, stability, and computational feasibility.
This study presents an informed machine learning approach using contrastive principal component analysis (cPCA) to successfully filter artifacts and identify statistically significant, interpretable neurometabolic states from complex MRSI data in multiple sclerosis patients, thereby transforming sparse spectral information into testable representations for clinical and fundamental research.