Brief

last 24h

[9/9] 221 sources

Multi-source AI news clustered, deduplicated, and scored 0–100 across authority, cluster strength, headline signal, and time decay.

TOOL · arXiv cs.AI English(EN) · 2h

Catching MRI outliers: unsupervised detection and localization of MRI artefacts and clinical anomalies using deep learning

Researchers have developed an unsupervised deep learning framework to detect and localize anomalies in MRI scans, aiming to improve radiotherapy workflows. The two-stage system first tokenizes MRI slices and then models the distribution of normal tokens to identify deviations. This approach demonstrated high accuracy, with AUCs of 0.97 for pelvic MRI and 0.81 for brain MRI, and showed strong spatial agreement for anomaly localization. AI

IMPACT Enhances AI reliability in medical imaging by providing a quality control layer for radiotherapy workflows.
- deep learning
- radiotherapy
TOOL · arXiv cs.CV English(EN) · 1d

A European Multi-Center Breast Cancer MRI Dataset

Researchers have introduced a new, publicly accessible dataset of European breast cancer MRI scans to advance AI development in medical imaging. The dataset includes 741 examinations from six institutions across five countries, encompassing a range of lesion types and reflecting real-world variations in acquisition protocols and scanner hardware. Baseline experiments using a transformer-based model are also provided to establish performance benchmarks for future research. AI

IMPACT Enables development of AI tools for earlier and more accurate breast cancer detection through standardized imaging data.
TOOL · arXiv cs.CV English(EN) · 4d

SO-Mamba: State-Ownership Mamba for Unrolled MRI Reconstruction

Researchers have developed SO-Mamba, a novel state-space model designed for accelerated MRI reconstruction. This model improves upon existing methods by differentiating between persistent reconstruction evidence and update-dependent information within its processing stages. SO-Mamba utilizes a State-Ownership Router to manage this evidence, leading to enhanced accuracy and anatomical coherence in MRI scans. Experiments on multiple public benchmarks demonstrate SO-Mamba's superior performance compared to CNN, Transformer, and standard Mamba-based approaches, while maintaining efficient computation. AI

IMPACT Introduces a new model architecture that improves MRI reconstruction accuracy and efficiency.
- CNN
- Transformer
- Mamba
- SO-Mamba
RESEARCH · arXiv cs.CV English(EN) · 1w · [2 sources]

SIREM: Speech-Informed MRI Reconstruction with Learned Sampling

Researchers have developed new methods for real-time MRI (rtMRI) of speech production by integrating acoustic information with visual data. One approach, Speech-Guided Multimodal Learning, uses phonological representations derived from speech to guide articulator localization and fuses visual and acoustic encoders for precise segmentation. Another method, SIREM, reconstructs rtMRI by combining an audio-driven component with MRI data, allowing for faster acquisition and reconstruction while maintaining anatomical accuracy. These techniques aim to improve the visualization of vocal tract motion for speech science and clinical applications. AI

IMPACT Advances in multimodal AI for medical imaging could lead to faster, more accurate diagnostic tools for speech and vocal tract disorders.
RESEARCH · arXiv cs.CV English(EN) · 4d · [2 sources]

Robustness of breast lesion segmentation under MRI undersampling improves with k-space-aware deep learning

Researchers have developed a k-space-aware deep learning approach that enhances the accuracy of breast lesion segmentation in MRI scans, particularly when data is undersampled or noisy. This novel method, tested on public DCE-MRI datasets, demonstrated superior performance compared to traditional image-space baselines under accelerated sampling conditions. The study suggests that integrating frequency-domain filtering with image-domain localization improves segmentation robustness without sacrificing accuracy in fully sampled scenarios. AI

IMPACT Enhances diagnostic accuracy in medical imaging by improving segmentation robustness under challenging data conditions.
TOOL · arXiv cs.CV English(EN) · 1w

Benchmarking transferability of SSL pretraining to same and different modality segmentation tasks

Researchers have benchmarked nine self-supervised learning (SSL) methods for their transferability in medical image segmentation tasks. The study found that the Self-Distilled Masked Image Transformer (SMIT) method, which combines masked image modeling with self-distillation, achieved the highest accuracy and fastest convergence. SMIT also demonstrated superior data efficiency, particularly in few-shot learning scenarios, outperforming contrastive learning and rotation prediction methods. AI

IMPACT Highlights SMIT as a highly data-efficient method for medical image segmentation, crucial for scenarios with limited annotations.
- CT
- Swin Transformer
- SwinUNETR
- SMIT
RESEARCH · arXiv cs.CV English(EN) · 4d · [2 sources]

D3Seg: Dependency-Aware Diffusion for Brain Tumor Segmentation with Missing Modalities

Researchers have developed a new model called D3Seg to improve brain tumor segmentation from MRI scans, particularly when some imaging modalities are missing. The model uses a novel Multi-hop Modality Graph Fusion technique to understand relationships between different MRI sequences and a diffusion-based imputation method to fill in gaps. Evaluations on the BraTS 2023 dataset show D3Seg achieves significant improvements in accuracy, outperforming current state-of-the-art methods by 1-2% in Dice scores for tumor subregions while remaining computationally efficient. AI

IMPACT Enhances medical imaging analysis by providing a more robust segmentation model for scenarios with incomplete data.
- D3Seg
- BraTS 2023
TOOL · arXiv cs.CV English(EN) · 6d

Decentralized Direct Volume Rendering: A Browser-Native GPU Architecture for MRI Digital Twins in Resource-Constrained Settings

Researchers have developed a new browser-native GPU architecture for creating interactive MRI digital twins. This decentralized approach bypasses traditional server-side rendering, executing complex 3D simulations directly on low-cost edge GPUs. The system achieves rapid rendering times and stable interactivity, making high-fidelity anatomical models accessible even in resource-constrained environments without relying on deep learning. AI

IMPACT Enables more accessible and interactive medical simulations, potentially improving surgical planning and personalized medicine.
RESEARCH · arXiv cs.AI English(EN) · 6d · [9 sources]

WikiVQABench: A Knowledge-Grounded Visual Question Answering Benchmark from Wikipedia and Wikidata

Researchers have introduced several new benchmarks and methods for Visual Question Answering (VQA) systems. HyLoVQA proposes a dynamic hypernetwork-generated low-rank adaptation technique for continual VQA, improving adaptation to new tasks and objects. WikiVQABench offers a knowledge-grounded VQA benchmark using Wikipedia and Wikidata, designed to test models requiring external knowledge. Additionally, UCSF-PDGM-VQA focuses on brain tumor MRI interpretation, highlighting current VLM limitations in clinical settings, while RoboSurg-VQA addresses surgical segmentation-aware VQA, and VISTAQA benchmarks joint answer correctness and pixel-level evidence grounding. AI

IMPACT These new benchmarks and adaptation techniques aim to improve the reliability and capabilities of Vision-Language Models in complex, real-world scenarios.