🎯 Authors propose a distillation method to reduce shortcut learning in medical imaging models by guiding training with a task-relevant teacher network. The approach improves robustness across datasets, even with limited bias annotations.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors present a model for CT-based body composition analysis, offering accurate muscle and fat segmentation with key metric extraction. It supports 2D and 3D assessments and performs well across diverse datasets.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors present an interpretable model for multi-label classification that uses counterfactual attribution maps to align predictions with clinical reasoning, offering local and global explanations without sacrificing performance.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors explore calibration as a fairness metric for skin cancer detection, showing that models often overdiagnose and are poorly calibrated across sex, skin tone, and age, highlighting the need for better model auditing in clinical AI.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors introduce an instance segmentation framework for ball-shaped medical objects and achieve robust, rotation-invariant segmentation, outperforming benchmarks on glomeruli and nucleus detection.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose a deep learning method for rapid voxel-wise estimation of tissue composition parameters from chemical shift-encoded MRI, enabling accurate, anatomy-independent estimation with up to 2800× speed-up compared to traditional fitting methods.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose a hybrid 3D CNN with cross-attention for glaucoma classification from OCT scans, designed to capture asymmetries across hemiretinas and integrate optic nerve head and macula features.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose a domain-adaptive framework for brain vessel segmentation that uses image-to-image translation and disentanglement to handle varied imaging modalities without domain-specific design.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors examine sex bias in deep learning models for ECG classification and find that performance varies across conditions and model types. Even with balanced training data, disparities persist, emphasizing the need for fairness in clinical AI.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose a distance map–based cell segmentation method that supports partially annotated objects, addressing limitations of fully supervised approaches. It enables effective transfer learning while maintaining segmentation quality.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose a lightweight debiasing method that fine-tunes model parameters based on their contributions to bias and prediction. With minimal data and training, it improves fairness and generalization without compromising accuracy.
🔎 Free to read: doi.org/10.59275/j.m...

🎯 Authors propose Neural CRF (NCRF), an end-to-end model for prostate MRI segmentation using learnable deep feature-based potentials. NCRF improves consistency and outperforms traditional CRFs in zonal segmentation accuracy.
🔎 Free to read: doi.org/10.59275/j.m...