Neuro-TransUNet: Segmentation of stroke lesion in MRI using transformers

TL;DR

Ischemic stroke lesion segmentation in 3D MRI is challenged by lesion heterogeneity and brain complexity. The authors introduce Neuro-TransUNet, a hybrid architecture that marries U-Net's local feature extraction with SwinUNETR's global contextual reasoning, enhanced by a robust preprocessing pipeline and a fusion module to produce accurate 3D segmentation. On the ATLAS v2.0 benchmark, Neuro-TransUNet achieves a Dice score of $0.730$, outperforming several state-of-the-art approaches, with favorable boundary metrics and strong ablation support for the integrated design. The work demonstrates that volumetric transformer-CNN fusion can improve lesion delineation and generalization, offering potential clinical value for rapid and reliable stroke assessment.

Abstract

Accurate segmentation of the stroke lesions using magnetic resonance imaging (MRI) is associated with difficulties due to the complicated anatomy of the brain and the different properties of the lesions. This study introduces the Neuro-TransUNet framework, which synergizes the U-Net's spatial feature extraction with SwinUNETR's global contextual processing ability, further enhanced by advanced feature fusion and segmentation synthesis techniques. The comprehensive data pre-processing pipeline improves the framework's efficiency, which involves resampling, bias correction, and data standardization, enhancing data quality and consistency. Ablation studies confirm the significant impact of the advanced integration of U-Net with SwinUNETR and data pre-processing pipelines on performance and demonstrate the model's effectiveness. The proposed Neuro-TransUNet model, trained with the ATLAS v2.0 \emph{training} dataset, outperforms existing deep learning algorithms and establishes a new benchmark in stroke lesion segmentation.

Figures (6)

• Figure 1: Different types of brain strokes (ischemic and hemorrhagic).
• Figure 2: Diverse representations of stroke (in red) in MRI scans.
• Figure 3: Stroke distribution in the ATLAS v2.0 training dataset. The bar plot shows the number of subjects with either single or multiple lesions, subdivided into specific hemisphere locations. The pie chart shows the percentage of lesions identified in different regions.
• Figure 4: A semantic visualization of the Neuro-TransUNet architecture, emphasizing the synergistic combination of the spatial and contextual information processing components. It effectively illustrates the model’s three pivotal components: the U-Net spatial encoder-decoder (U-NetSED), the SwinUNETR global context encoder (Swin-GCE), and the feature fusion and segmentation synthesis mechanism.
• Figure 5: The model performance across 100 epochs. The training loss that is gradually reduced shows the model is learning, while the test loss trend is also reducing, indicating that the model has generalization capability. The DSC demonstrates increasing accuracy in segmenting the desired regions. The Test HD95 shows a downward trend, indicating better precision in boundary segmentation