MedFlowBench evaluates VLM agents on full radiology and pathology studies by requiring both task answers and verifiable evidence like key slices and regions of interest, revealing that answer-only scores overestimate performance.
hub Mixed citations
MONAI: An open-source framework for deep learning in healthcare
Mixed citation behavior. Most common role is background (50%).
53
Pith papers citing it
Background
50% of classified citations
abstract
Artificial Intelligence (AI) is having a tremendous impact across most areas of science. Applications of AI in healthcare have the potential to improve our ability to detect, diagnose, prognose, and intervene on human disease. For AI models to be used clinically, they need to be made safe, reproducible and robust, and the underlying software framework must be aware of the particularities (e.g. geometry, physiology, physics) of medical data being processed. This work introduces MONAI, a freely available, community-supported, and consortium-led PyTorch-based framework for deep learning in healthcare. MONAI extends PyTorch to support medical data, with a particular focus on imaging, and provide purpose-specific AI model architectures, transformations and utilities that streamline the development and deployment of medical AI models. MONAI follows best practices for software-development, providing an easy-to-use, robust, well-documented, and well-tested software framework. MONAI preserves the simple, additive, and compositional approach of its underlying PyTorch libraries. MONAI is being used by and receiving contributions from research, clinical and industrial teams from around the world, who are pursuing applications spanning nearly every aspect of healthcare.
hub tools
citation-role summary
background 3
method 2
other 1
citation-polarity summary
representative citing papers
DiffBCP combines a cumulative shrinkage prior and pre-trained diffusion model prior in Bayesian CP decomposition via a split Gibbs sampler with noise-adaptive coupling, yielding gains on image inpainting and denoising.
A sequential diffusion framework generates controllable abdominal anatomies with a Volume Control Scalar that decouples organ size from body habitus, achieving Dice scores around 0.83 and reducing distributional mismatch by 73.6% in a hepatomegaly example.
Camyla autonomously generates research proposals, experiments, and manuscripts in medical image segmentation, outperforming baselines on 24 of 31 recent datasets while producing 40 human-reviewed papers.
MP-ViT uses dual transformers and cross-attention on axial and sagittal MRI to classify hemorrhages, reporting 5.5% higher AUC than standard ViT and 1.8% higher than CNNs on a dataset of 12,869 subjects.
Tabular clinical data guides contrastive learning on cardiac MR images to build better visual representations by identifying patient similarities, outperforming image-only augmentation on downstream disease prediction tasks.
Proposes a cyclic 2.5D perceptual loss with manufacturer SUVR standardization for T1w MRI to tau PET synthesis, reporting improved regional agreement on ADNI and SCAN cohorts across U-Net, UNETR, SwinUNETR, CycleGAN, and Pix2Pix.
BenchX supplies an 85k-scan benchmark that exposes poor performance of 12 tumor-detection models on underrepresented demographic and protocol subgroups.
CSCS selects initial annotation samples for 3D medical segmentation by combining self-supervised typicality and reconstruction uncertainty through a closed-form pacing rule based on the Difficulty-Coverage Ratio.
SurgVista mitigates spatial interaction incoherence and temporal fidelity collapse in surgical world models through trajectory-based contrastive regularization and drift-perturbed training, outperforming prior methods on a new long-horizon benchmark.
Acquisition parameters like kernel and noise produce axis-specific effects on lung-nodule AI (measurement vs detection) that a 4-feature pixel fingerprint can recover even when DICOM metadata cannot.
GLINT introduces sparsely gated alignment and dense feature regularization on top of DINOv3 and V-JEPA encoders to enable query-specific zero-shot grounding and segmentation in 2D CXR and 3D CT.
Entropy minimization amplifies prediction bias from merged feature clusters under distribution shifts, and DSBR mitigates collapse by equalizing predicted class contributions to the unsupervised loss.
A foundation VAE pretrained on natural images and videos serves as a frozen interface for CT reconstruction, augmentation, and generation, yielding 3.9% NSD gains in segmentation and improved generation metrics across 18 diseases.
A histogram-regularized latent diffusion model synthesizes controllable pulmonary nodules in full 3D CT volumes with improved lesion-level intensity distributions.
VISTA is a test-time adaptation framework for multi-sequence MRI that uses inter-sequence intervention probes and cross-view disagreement variance to gate self-training, yielding Dice gains of +1.89% on low-field African data and +2.82% on pediatric data over the source model.
A teacher–student semi-supervised framework with alignment-preserving patch mixing, position-aware text augmentation, and positional contrastive learning improves medical referring segmentation at low label ratios.
A uniform benchmark across 77 experiments finds SRGAN superior to latent diffusion models for 3D medical image translation, with synthetic volumes indistinguishable from real ones in a 17-physician Turing test.
SIAM achieves state-of-the-art whole-head MRI segmentation of 16 structures including extra-cerebral tissues by training on synthetic data from just six manual templates, matching or exceeding prior methods on 301 scans across eight heterogeneous datasets.
GeoSAE extracts a compact, interpretable feature set from frozen brain MRI foundation models that predicts MCI-to-AD conversion (AUC 0.746) with age-deconfounded annotations and replicates across cohorts.
ESICA delivers state-of-the-art accuracy on a five-modality 3D medical segmentation benchmark while offering a compact variant with far fewer parameters.
A 4D diffusion generative model learns topology-preserving spatiotemporal deformations to synthesize realistic longitudinal brain anatomy trajectories in neurodegenerative diseases from sparse follow-up scans.
Post-hoc rank-1 probe heads convert feature perturbation energy into separate calibration and ranking uncertainty maps that beat other single-pass methods on ACDC, BraTS2024, and LiTS.
DAGMaN uses co-distilled attention-guided masked image modeling with a noisy teacher to enable effective self-supervised pretraining on medical images by selective masking of co-occurring patches and maintenance of attention head diversity, with demonstrations on nodule classification, immunotherapy
citing papers explorer
-
MedOpenClaw and MedFlowBench: Auditing Medical Agents in Full-Study Workflows
MedFlowBench evaluates VLM agents on full radiology and pathology studies by requiring both task answers and verifiable evidence like key slices and regions of interest, revealing that answer-only scores overestimate performance.
-
Bayesian Tensor Decomposition with Diffusion Model Prior
DiffBCP combines a cumulative shrinkage prior and pre-trained diffusion model prior in Bayesian CP decomposition via a split Gibbs sampler with noise-adaptive coupling, yielding gains on image inpainting and denoising.
-
AbdomenGen: Sequential Volume-Conditioned Diffusion Framework for Abdominal Anatomy Generation
A sequential diffusion framework generates controllable abdominal anatomies with a Volume Control Scalar that decouples organ size from body habitus, achieving Dice scores around 0.83 and reducing distributional mismatch by 73.6% in a hepatomegaly example.
-
Camyla: Scaling Autonomous Research in Medical Image Segmentation
Camyla autonomously generates research proposals, experiments, and manuscripts in medical image segmentation, outperforming baselines on 24 of 31 recent datasets while producing 40 human-reviewed papers.
-
Multi-Plane Vision Transformer for Hemorrhage Classification Using Axial and Sagittal MRI Data
MP-ViT uses dual transformers and cross-attention on axial and sagittal MRI to classify hemorrhages, reporting 5.5% higher AUC than standard ViT and 1.8% higher than CNNs on a dataset of 12,869 subjects.
-
Tables Guide Vision: Learning to See the Heart through Tabular Data
Tabular clinical data guides contrastive learning on cardiac MR images to build better visual representations by identifying patient similarities, outperforming image-only augmentation on downstream disease prediction tasks.
-
Cyclic 2.5D Perceptual Loss for Cross-Modal 3D Medical Image Synthesis: T1w MRI to Tau PET
Proposes a cyclic 2.5D perceptual loss with manufacturer SUVR standardization for T1w MRI to tau PET synthesis, reporting improved regional agreement on ADNI and SCAN cohorts across U-Net, UNETR, SwinUNETR, CycleGAN, and Pix2Pix.
-
BenchX: Benchmarking AI Models for Cancer Detection and Localization with Demographic and Protocol Biases
BenchX supplies an 85k-scan benchmark that exposes poor performance of 12 tumor-detection models on underrepresented demographic and protocol subgroups.
-
Dataset-Aware Cold-Start Active Learning for Annotation-Efficient 3D Medical Image Segmentation
CSCS selects initial annotation samples for 3D medical segmentation by combining self-supervised typicality and reconstruction uncertainty through a closed-form pacing rule based on the Difficulty-Coverage Ratio.
-
SurgVista: Long-Horizon Surgical World Modeling with Plausible Instrument-Tissue Dynamics
SurgVista mitigates spatial interaction incoherence and temporal fidelity collapse in surgical world models through trajectory-based contrastive regularization and drift-perturbed training, outperforming prior methods on a new long-horizon benchmark.
-
Acquisition state behaves as a structured, measurable variable governing lung-nodule AI: kernel-driven measurement instability and noise-driven detection fragility, invisible to DICOM metadata
Acquisition parameters like kernel and noise produce axis-specific effects on lung-nodule AI (measurement vs detection) that a 4-feature pixel fingerprint can recover even when DICOM metadata cannot.
-
GLINT: Sparsely Gated Vision-Language Alignment for Fine-Grained Radiology Representations
GLINT introduces sparsely gated alignment and dense feature regularization on top of DINOv3 and V-JEPA encoders to enable query-specific zero-shot grounding and segmentation in 2D CXR and 3D CT.
-
Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging
Entropy minimization amplifies prediction bias from merged feature clusters under distribution shifts, and DSBR mitigates collapse by equalizing predicted class contributions to the unsupervised loss.
-
Foundation VAEs for 3D CT Reconstruction, Augmentation, and Generation
A foundation VAE pretrained on natural images and videos serves as a frozen interface for CT reconstruction, augmentation, and generation, yielding 3.9% NSD gains in segmentation and improved generation metrics across 18 diseases.
-
Controllable Lung Nodule Synthesis via Histogram-Regularized Latent Diffusion Models
A histogram-regularized latent diffusion model synthesizes controllable pulmonary nodules in full 3D CT volumes with improved lesion-level intensity distributions.
-
VISTA: Variance-Gated Inter-Sequence Test-Time Adaptation for Multi-Sequence MRI Segmentation
VISTA is a test-time adaptation framework for multi-sequence MRI that uses inter-sequence intervention probes and cross-view disagreement variance to gate self-training, yielding Dice gains of +1.89% on low-field African data and +2.82% on pediatric data over the source model.
-
Semi-MedRef: Semi-Supervised Medical Referring Image Segmentation with Cross-Modal Alignment
A teacher–student semi-supervised framework with alignment-preserving patch mixing, position-aware text augmentation, and positional contrastive learning improves medical referring segmentation at low label ratios.
-
Cross Modality Image Translation In Medical Imaging Using Generative Frameworks
A uniform benchmark across 77 experiments finds SRGAN superior to latent diffusion models for 3D medical image translation, with synthetic volumes indistinguishable from real ones in a 17-physician Turing test.
-
SIAM: Head and Brain MRI Segmentation from Few High-Quality Templates via Synthetic Training
SIAM achieves state-of-the-art whole-head MRI segmentation of 16 structures including extra-cerebral tissues by training on synthetic data from just six manual templates, matching or exceeding prior methods on 301 scans across eight heterogeneous datasets.
-
GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation Models
GeoSAE extracts a compact, interpretable feature set from frozen brain MRI foundation models that predicts MCI-to-AD conversion (AUC 0.746) with age-deconfounded annotations and replicates across cohorts.
-
ESICA: A Scalable Framework for Text-Guided 3D Medical Image Segmentation
ESICA delivers state-of-the-art accuracy on a five-modality 3D medical segmentation benchmark while offering a compact variant with far fewer parameters.
-
Generative Modeling of Neurodegenerative Brain Anatomy with 4D Longitudinal Diffusion Model
A 4D diffusion generative model learns topology-preserving spatiotemporal deformations to synthesize realistic longitudinal brain anatomy trajectories in neurodegenerative diseases from sparse follow-up scans.
-
SegWithU: Uncertainty as Perturbation Energy for Single-Forward-Pass Risk-Aware Medical Image Segmentation
Post-hoc rank-1 probe heads convert feature perturbation energy into separate calibration and ranking uncertainty maps that beat other single-pass methods on ACDC, BraTS2024, and LiTS.
-
Co-distilled attention guided masked image modeling with noisy teacher for self-supervised learning on medical images
DAGMaN uses co-distilled attention-guided masked image modeling with a noisy teacher to enable effective self-supervised pretraining on medical images by selective masking of co-occurring patches and maintenance of attention head diversity, with demonstrations on nodule classification, immunotherapy
-
Backbone-Conditional Behavior of Modality Gating in Multi-Modal Prostate MRI Segmentation: A 5-Fold Cross-Validation and Gate Mechanism Analysis
Modality gating is backbone-conditional: nnU-Net gates become static priors that degrade ranking score, while Mamba gates retain per-sample routing and improve it; only modality dropout helps both.
-
Neuro-Oracle: A Trajectory-Aware Agentic RAG Framework for Interpretable Epilepsy Surgical Prognosis
Neuro-Oracle distills longitudinal MRI changes into trajectory vectors via a 3D Siamese encoder, retrieves similar cases, and generates LLM-based prognoses, achieving AUC 0.834-0.905 on a resection-type proxy task versus 0.793 for single-timepoint baseline.
-
Distilling Photon-Counting CT into Routine Chest CT through Clinically Validated Degradation Modeling
SUMI distills photon-counting CT quality into routine chest CT by learning to reverse clinically validated acquisition degradations, yielding 15-20% gains in image metrics, better radiologist utility, and up to 15% higher lesion detection sensitivity.
-
SMILE-UHURA Challenge -- Small Vessel Segmentation at Mesoscopic Scale from Ultra-High Resolution 7T Magnetic Resonance Angiograms
The paper reports a new annotated 7T ToF MRA dataset for small vessel segmentation and shows that top deep learning methods reach Dice scores of 0.838 on internal test data and 0.716 on an external secret dataset.
-
CenSynCMB: Centre Maps and Physics-Guided Synthesis for Microbleed Detection
A centre-guided Attention U-Net with physics-guided synthetic mimics achieves best local-comparison lesion-level F1 for CMB detection on VALDO and external AIBL SWI.
-
Unsupervised Susceptibility Distortion Correction of EPI without Calibration Scans via Image Translation-Based Registration
SACRED performs unsupervised susceptibility distortion correction of EPI fMRI via image translation-based registration between T1w and unidirectional BOLD images, with test-time adaptation for robustness.
-
When Average Calibration Fails: Site-Conditional Federated Conformal Risk Control
Pooled conformal calibration in federated medical segmentation can miss its target at 40% of sites; blending site-level and global risk curves with a tuned prior restores per-site coverage at 2.0x prediction-set stretch.
-
vesselFM-CT: Segmenting All Blood Vessels in CT Images for System-Level Cardiovascular Analysis
vesselFM-CT is presented as the first model to segment the entire vascular network in 3D CT images via iterative multi-step training and a custom TubeLoss function.
-
Universal CT Representations from Anatomy to Disease Phenotype through Agglomerative Pretraining
FlexiCT provides CT foundation models via agglomerative pretraining on 266227 volumes from 56 datasets that match or exceed task-specific models on five task families while organizing embeddings along tumor-stage gradients.
-
NeuroAgent: LLM Agents for Multimodal Neuroimaging Analysis and Research
NeuroAgent uses a hierarchical LLM agent framework with Generate-Execute-Validate loops to automate neuroimaging preprocessing, reaching 84.8% end-to-end correctness and 0.9518 AUC for Alzheimer's classification on 1470 ADNI subjects using four modalities.
-
The autoPET3 Challenge: Automated Lesion Segmentation in Whole-Body PET/CT $\unicode{x2013}$ Multitracer Multicenter Generalization
The autoPET3 challenge finds that leading AI models reach a mean Dice score of 0.66 for multitracer PET/CT lesion segmentation, with compositional generalization to unseen tracer-center pairs remaining an open problem driven by volume overestimation and case heterogeneity.
-
Tumor-aware augmentation with task-guided attention analysis improves rectal cancer segmentation from magnetic resonance images
Tumor-aware augmentation and anisotropic cropping mitigate token inefficiency and feature adaptation issues in CT-to-MRI transfer for rectal cancer segmentation, achieving 88.7-91.1% detection rates.
-
Multimodal synthesis of MRI and tabular data with diffusion in a joint latent space via cross-attention
A latent diffusion model jointly synthesizes MRI volumes and mixed-type tabular clinical data in a shared space via cross-attention and separate decoders after VAE fusion.
-
Why Invariance is Not Enough for Biomedical Domain Generalization and How to Fix It
MaskGen improves domain generalization for biomedical image segmentation by using source intensities plus domain-stable foundation model representations with minimal added complexity.
-
Submanifold Sparse Convolutional Networks for Automated 3D Segmentation of Kidneys and Kidney Tumours in Computed Tomography
A two-stage sparse convolutional network pipeline for native high-resolution 3D kidney and tumor segmentation in CT that matches top Dice scores while reducing VRAM and runtime versus nnU-Net and SegVol.
-
The Large Cancer Assistant (LCA): A Model-Agnostic Orchestration Framework for Scalable Clinical Decision Support in Oncology
A formal orchestration framework for oncology AI pipelines is proposed, demonstrating that routing logic and output schema remain invariant under model substitution via a proof-of-concept with synthetic stubs.
-
APRIL-MedSeg: A Modular Medical Image Segmentation Toolbox Embracing Modern Paradigms
Presents APRIL-MedSeg, a modular YAML-configurable toolbox for 2D medical image segmentation integrating semi-supervised, domain adaptation, distillation, weakly supervised, text-guided, and foundation model paradigms with unified dataset and deployment interfaces.
-
Confidence is Not Reliability: Rethinking MC Dropout in Brain Tumour Segmentation
MC Dropout yields strong global uncertainty-error alignment in brain tumor segmentation yet reveals region-specific miscalibration on enhancing tumor that standard metrics miss.
-
Pre-Deployment Robustness Stress Testing for CT Segmentation Systems Using Clinically Motivated Multi-Corruption Augmentation
RAMP combines anatomically constrained spatial perturbations, CT intensity transforms, and stochastic multi-corruption during training, raising mean corrupted Dice from 0.610 to 0.753 and shrinking the clean-to-corrupted gap from 0.264 to 0.064 on a five-organ benchmark versus nnU-Net.
-
One Sequence to Segment Them All: Efficient Data Augmentation for CT and MRI Cross-Domain 3D Spine Segmentation
Targeted data augmentations let single-sequence 3D spine segmentation models generalize to seven unseen CT and MRI datasets with 155% average Dice gain and almost no in-domain loss.
-
AMO-ENE: Attention-based Multi-Omics Fusion Model for Outcome Prediction in Extra Nodal Extension and HPV-associated Oropharyngeal Cancer
An attention-based fusion model combining semi-supervised CT segmentation, radiomics, and clinical features predicts metastatic recurrence, overall survival, and disease-free survival in HPV+ oropharyngeal cancer with AUCs of 88.2%, 79.2%, and 78.1% on an internal cohort of 397 patients.
-
GPAFormer: Graph-guided Patch Aggregation Transformer for Efficient 3D Medical Image Segmentation
GPAFormer with 1.81M parameters reports top Dice scores on BTCV (75.70%), Synapse (81.20%), ACDC (89.32%), and BraTS (82.74%) while running inference in under one second on consumer GPUs.
-
PR3DICTR: A modular AI framework for medical 3D image-based detection and outcome prediction
PR3DICTR is a new open-access modular framework for 3D medical image classification and outcome prediction that works with as little as two lines of code.
-
In search of truth: Evaluating concordance of AI-based anatomy segmentation models
A harmonization framework enables comparison of six AI segmentation models on 31 structures in NLST CT scans, revealing strong agreement for lungs but invalid outputs for some vertebrae and ribs.
-
Improving Prostate Gland Segmentation Using Transformer based Architectures
SwinUNETR outperforms 3D UNet with Dice scores up to 0.902 on larger gland subsets using mixed-cohort five-fold training, while UNETR performs poorly on the same subsets.
-
Patient-Level Elbow Abnormality Detection: Leakage-Aware Evaluation of Learned Preprocessing, Calibration, and Triage-Oriented Operating Points
Learned preprocessing yields modest and configuration-dependent gains in discrimination and calibration for elbow abnormality detection, with the raw DenseNet121 baseline remaining competitive.