Related Experiment Video
Updated: May 15, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.7K
Multi-plane vision transformer for hemorrhage classification using axial and sagittal MRI data
Badhan Kumar Das1,2, Gengyan Zhao3, Boris Mailhe3
1Digital Technology and Innovation, Siemens Healthineers, Erlangen, Germany. badhankumar.das@siemens-healthineers.com.
Scientific Reports
|March 19, 2026
Summary
A new 3D multi-plane vision transformer (MP-ViT) effectively identifies brain hemorrhages in MRI scans with varying orientations. This approach improves detection accuracy by integrating information across different image planes, outperforming existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Identifying brain hemorrhages from magnetic resonance imaging (MRI) is crucial for clinical diagnosis.
- Variations in MRI acquisition contrasts and orientations pose challenges for automated detection using neural networks.
- Traditional methods like resampling can lead to information loss when handling varied image orientations.
Purpose of the Study:
- To propose a novel 3D multi-plane vision transformer (MP-ViT) model for robust brain hemorrhage classification.
- To address the challenge of varying MRI acquisition orientations without information loss.
- To improve the accuracy of hemorrhage detection in diverse clinical MRI datasets.
Main Methods:
- Developed a 3D multi-plane vision transformer (MP-ViT) incorporating separate encoders for axial and sagittal contrasts.
- Utilized cross-attention mechanisms to integrate information across different image orientations.
- Introduced a modality indication vector to supply missing contrast information to the model.
- Validated the model on a large-scale clinical dataset (10,084 training, 1,289 validation, 1,496 test subjects).
Main Results:
- MP-ViT demonstrated substantial improvements in the area under the curve (AUC) for hemorrhage classification.
- The proposed MP-ViT outperformed standard vision transformer (ViT) by 5.5% in AUC.
- MP-ViT achieved a 1.8% higher AUC compared to convolutional neural network (CNN)-based architectures.
- The model showed significant potential in enhancing hemorrhage detection performance with varied orientation contrasts.
Conclusions:
- The 3D multi-plane vision transformer (MP-ViT) offers a promising solution for brain hemorrhage detection in diverse MRI orientations.
- Integrating multi-plane information and contrast data improves classification accuracy.
- MP-ViT represents a significant advancement in automated medical image analysis for neurological conditions.
