Article

Transformers Achieve 4× Better Cross-Scanner Generalization Than CNNs: Multi-Center Validation of 3D Swin Transformers Across 1,066 Subjects

2026-07-02

Abstract excerpt

<title>Abstract</title> <p>Purpose: Cross-scanner generalization remains the critical barrier to clinical deployment of deep learning for medical image segmentation [1], [2]. While convolutional neural networks (CNNs) show 8-15% performance degradation when applied to data from different scanners [3]–[5], the ability of transformer architectures [6], [7] to learn scanner-invariant features is unknown. This work p...

Topics

Open a Topic to create a Post that cites this publication.

Identifiers and source

Literature Corpus work: 59a27830-a2c0-51c8-ae04-a56e4fe086b1
DOI: 10.21203/rs.3.rs-9402599/v1

Open publication

Related research

Semantic proximity does not establish scientific evidence.

Click a neighbor to travelStep 1 · 12 closest

Transformers Achieve 4× Better Cross-Scanner Generalization Than CNNs: Multi-Center Validation of 3D Swin Transformers Across 1,066 SubjectsDOI 10.21203/rs.3.rs-9402599/v1Read

Do Hybrid CNN–Transformer Architectures Really Generalize? A Systematic Review for Medical ImagingSemantic proximity Read

Deep Learning in Brain MRI: Effect of Data Leakage Due to Slice-level Split Using 2D Convolutional Neural NetworksSemantic proximity Read

Domain Shift-Robust Tumor Segmentation Across HospitalsSemantic proximity Read

Few-Shot Deployment of Pretrained MRI Transformers in Brain Imaging TasksSemantic proximity Read

A unified deep learning framework for cross-platform harmonization of multi-tracer PET quantificationSemantic proximity Read

Advances in Deep Learning for Medical Imaging: Foundations, Evolution, and ImpactSemantic proximity Read

A deep learning framework for comprehensive segmentation of deep grey nucleiSemantic proximity Read

MAESTRO: A Public, Generalizable Model for Stroke Lesion Segmentation from T1 MRI Across the Recovery ContinuumSemantic proximity Read

MRICombo: a deep-learning-based framework for universal volume segmentation, grading-staging, and malignancy detection across heterogeneous MRISemantic proximity Read

UltimateSynth: MRI Physics for Pan-Contrast AISemantic proximity Read

Synthetic MRI Pretraining for Medical Imaging TasksSemantic proximity Read

ADG-Net: A Lightweight Adaptive Deformable Convolution Network with Gradient-Optimized Pooling for Medical Image SegmentationSemantic proximity Read