AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
hub
nnU-Net: A Self-Configuring Method for Deep Learning-Based Biomedical Image Segmentation
27 Pith papers cite this work, alongside 8,561 external citations. Polarity classification is still indexing.
27
Pith papers citing it
8,561
external citations · OpenAlex
hub tools
citation-role summary
method 3
citation-polarity summary
roles
method 3polarities
use method 3representative citing papers
On the public ReMIND dataset, a systematic benchmark of six synthesis models across 48 experiments finds LPIPS correlates with downstream segmentation utility while SSIM does not, with SynDiff-2.5D performing best.
TriALS introduces a 150-case four-phase CT dataset and challenge showing top segmentation methods reach 0.754 Dice on venous phase but only 0.57 on non-contrast CT, with external validation gains up to 28%.
A differentiable Max-Min reachability loss redirects segmentation gradients to connectivity bottleneck pixels, improving topological correctness across nine architectures and six datasets.
Cross-sectional patches with near-identical intensity but inconsistent masks are flagged as annotation noise, revealing systematic orientation-dependent bias in single-rater vascular CT labels.
TRACE is a RANO 2.0-aligned concept bottleneck model for 4-class longitudinal glioblastoma response classification on 3D MRI that reports 0.4769 macro F1 on the LUMIERE dataset via 5-fold patient-wise cross-validation.
Neuro-JEPA, a JEPA-and-MoE transformer pretrained on 1.55M brain MRI scans, outperformed prior neuroimaging foundation models and was the only one to beat a simple CNN baseline on average.
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.
AtlasGS uses shared subject-specific Gaussian geometry learned from isotropic scans to achieve through-plane super-resolution and multi-modal harmonization in brain MRI with reported state-of-the-art fidelity on UK Biobank, GBM, and ABCD datasets.
A Jacobian sensitivity curve computed at initialization identifies the narrowest U-Net configuration that avoids performance collapse, matching nnU-Net accuracy with 400-1600x fewer parameters on six medical datasets.
LARGO uses a low-rank hypernetwork with CP decomposition to unify 2^N-1 missing-modality models into one, ranking first in 47 of 52 configurations on BraTS and ISLES with small Dice gains over baselines.
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.
MonoUNet is a tiny segmentation network that achieves 92-95% Dice scores on multi-device knee cartilage ultrasound while using 10-700x fewer parameters than prior lightweight models by injecting trainable local phase features.
Patient identity and clinical features predict brain tumor segmentation accuracy more strongly than model choice, with localized spatial biases consistent across models and no formal fairness guarantees in any.
In a private dataset of 353 patients, medical records and cardiac biomarkers outperform vascular biomarkers and GNNs on vascular graphs for PE risk stratification, suggesting vascular graphs hold no discriminative information.
SemanticVessel is a new CTA vessel segmentation dataset using intensity-guided region growing, expert labeling of 20 arterial classes, and multi-phase label reuse, with reported gains from including a generic minor-artery class.
ML pipeline on fetal MRI features predicts gestational age at birth (R²=0.13, MAE=2.74 weeks) and preterm status (accuracy 0.77) from 426 cases.
CheXanatomy trains VLMs to generate 2D anatomical masks via next-token prediction on synthetic CXRs from CT, matching U-Net performance with better domain-shift robustness and sample efficiency.
Episodic sampling for class-balanced batches in low-data CT segmentation delays overfitting compared to random or weighted sampling, revealing training iteration budget as a key evaluation confound.
TwinTrack applies post-hoc calibration to align deep learning segmentation probabilities with the empirical mean human response for explicit modeling of inter-rater disagreement.
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.
SegSTRONG-C provides a new benchmark where top models reach 0.9394 DSC and 0.9301 NSD on corrupted surgical tool segmentation tests, showing conventional techniques help but calling for more innovative robustness methods.
Automated reference-region normalization of optical density in myelin histology yields substantially stronger voxel-wise correlation with 7T ex vivo MRI than unnormalized measurements, including inside white matter hyperintensities.
MAML with auxiliary cavity tasks and boundary loss improves 5-shot LA wall segmentation over standard fine-tuning (DSC 0.54 vs 0.48) and nears fully supervised performance at 20 shots.
citing papers explorer
-
AutoMedBench: Towards Medical AutoResearch with Agentic AI Models
AutoMedBench evaluates AI agents on long-horizon medical workflows across five stages and finds validation and submission as dominant failure points based on thousands of runs.
-
A Systematic Benchmark of Intraoperative Ultrasound-to-MR Synthesis for Brain Tumour Surgery
On the public ReMIND dataset, a systematic benchmark of six synthesis models across 48 experiments finds LPIPS correlates with downstream segmentation utility while SSIM does not, with SynDiff-2.5D performing best.
-
TriALS: Triphasic-Aided Liver Lesion Segmentation Benchmark in Non-Contrast CT
TriALS introduces a 150-case four-phase CT dataset and challenge showing top segmentation methods reach 0.754 Dice on venous phase but only 0.57 on non-contrast CT, with external validation gains up to 28%.
-
Widest-Path Reachability Fields for Connectivity-Preserving Slender Structure Segmentation
A differentiable Max-Min reachability loss redirects segmentation gradients to connectivity bottleneck pixels, improving topological correctness across nine architectures and six datasets.
-
Decoupled Single-Mask Annotation Noise Detection via Cross-Sectional Patch Self-Consistency
Cross-sectional patches with near-identical intensity but inconsistent masks are flagged as annotation noise, revealing systematic orientation-dependent bias in single-rater vascular CT labels.
-
TRACE: A Concept Bottleneck Model for Longitudinal 3D Glioblastoma Response Assessment
TRACE is a RANO 2.0-aligned concept bottleneck model for 4-class longitudinal glioblastoma response classification on 3D MRI that reports 0.4769 macro F1 on the LUMIERE dataset via 5-fold patient-wise cross-validation.
-
Learning Sparse Latent Predictive Foundation Model for Multimodal Neuroimaging
Neuro-JEPA, a JEPA-and-MoE transformer pretrained on 1.55M brain MRI scans, outperformed prior neuroimaging foundation models and was the only one to beat a simple CNN baseline on average.
-
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.
-
AtlasGS: Brain MRI Spatial Resolution Harmonization With Shared Gaussian Geometry
AtlasGS uses shared subject-specific Gaussian geometry learned from isotropic scans to achieve through-plane super-resolution and multi-modal harmonization in brain MRI with reported state-of-the-art fidelity on UK Biobank, GBM, and ABCD datasets.
-
XTinyU-Net: Training-Free U-Net Scaling via Initialization-Time Sensitivity
A Jacobian sensitivity curve computed at initialization identifies the narrowest U-Net configuration that avoids performance collapse, matching nnU-Net accuracy with 400-1600x fewer parameters on six medical datasets.
-
LARGO: Low-Rank Hypernetwork for Handling Missing Modalities
LARGO uses a low-rank hypernetwork with CP decomposition to unify 2^N-1 missing-modality models into one, ranking first in 47 of 52 configurations on BraTS and ISLES with small Dice gains over baselines.
-
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.
-
MonoUNet: A Robust Tiny Neural Network for Automated Knee Cartilage Segmentation on Point-of-Care Ultrasound Devices
MonoUNet is a tiny segmentation network that achieves 92-95% Dice scores on multi-device knee cartilage ultrasound while using 10-700x fewer parameters than prior lightweight models by injecting trainable local phase features.
-
Fairboard: a quantitative framework for equity assessment of healthcare models
Patient identity and clinical features predict brain tumor segmentation accuracy more strongly than model choice, with localized spatial biases consistent across models and no formal fairness guarantees in any.
-
Pulmonary Embolism Risk Stratification from CTPA and Medical Records: Vascular Graphs Are Not All You Need
In a private dataset of 353 patients, medical records and cardiac biomarkers outperform vascular biomarkers and GNNs on vascular graphs for PE risk stratification, suggesting vascular graphs hold no discriminative information.
-
Scaling up fine-grained intracranial vessel annotations in computed tomography angiography
SemanticVessel is a new CTA vessel segmentation dataset using intensity-guided region growing, expert labeling of 20 arterial classes, and multi-phase label reuse, with reported gains from including a generic minor-artery class.
-
Predicting gestational age at birth in the context of preterm birth from multi-modal fetal MRI
ML pipeline on fetal MRI features predicts gestational age at birth (R²=0.13, MAE=2.74 weeks) and preterm status (accuracy 0.77) from 426 cases.
-
CheXanatomy: Anatomy-Aware Vision-Language Modeling for Chest Radiographs
CheXanatomy trains VLMs to generate 2D anatomical masks via next-token prediction on synthetic CXRs from CT, matching U-Net performance with better domain-shift robustness and sample efficiency.
-
Disentangling Sampling from Training Budget in Class-Imbalanced CT Body Composition Segmentation
Episodic sampling for class-balanced batches in low-data CT segmentation delays overfitting compared to random or weighted sampling, revealing training iteration budget as a key evaluation confound.
-
TwinTrack: Post-hoc Multi-Rater Calibration for Medical Image Segmentation
TwinTrack applies post-hoc calibration to align deep learning segmentation probabilities with the empirical mean human response for explicit modeling of inter-rater disagreement.
-
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.
-
SegSTRONG-C: Segmenting Surgical Tools Robustly On Non-adversarial Generated Corruptions -- An EndoVis'24 Challenge
SegSTRONG-C provides a new benchmark where top models reach 0.9394 DSC and 0.9301 NSD on corrupted surgical tool segmentation tests, showing conventional techniques help but calling for more innovative robustness methods.
-
Automated Optical Density Normalization for Myelin Quantification: Cross-Modal Validation with 7T Ex Vivo MRI
Automated reference-region normalization of optical density in myelin histology yields substantially stronger voxel-wise correlation with 7T ex vivo MRI than unnormalized measurements, including inside white matter hyperintensities.
-
Few-Shot Left Atrial Wall Segmentation in 3D LGE MRI via Meta-Learning
MAML with auxiliary cavity tasks and boundary loss improves 5-shot LA wall segmentation over standard fine-tuning (DSC 0.54 vs 0.48) and nears fully supervised performance at 20 shots.
-
Clinical utility of foundation models in musculoskeletal MRI for biomarker fidelity and predictive outcomes
Fine-tuned foundation models produce reliable MSK MRI biomarkers that support workload-reducing triage and calibrated 48-month prediction of knee replacement and incident OA.
-
HEad and neCK TumOR (HECKTOR) 2025: Benchmark of Segmentation, Diagnosis, and Prognosis in Multimodal PET/CT
The HECKTOR 2025 challenge establishes performance benchmarks for multimodal PET/CT analysis of head and neck cancer, with top methods reaching Dice 0.75 on tumor segmentation, C-index 0.66 on recurrence-free survival, and balanced accuracy 0.56 on HPV classification.
-
DALight-3D: A Lightweight 3D U-Net for Brain Tumor Segmentation from Multi-Modal MRI
DALight-3D achieves a mean Dice of 0.727 with 2.22M parameters on the Medical Segmentation Decathlon BrainTumour benchmark, slightly above the 0.710 Dice of Residual 3D U-Net with 3.20M parameters.