Transformer Dil-DenseUNet for Stroke Segmentation
Author Information
Author(s): Jazzar Nesrine, Mabrouk Besma, Douik Ali
Primary Institution: National Engineering School of Sousse, University of Sousse, Tunisia
Hypothesis
Can a novel architecture combining DenseNet, dilated convolutions, and Transformer blocks improve stroke lesion segmentation accuracy in MRI images?
Conclusion
The Transformer Dil-DenseUNet outperforms existing models in accurately segmenting stroke lesions in MRI images.
Supporting Evidence
- The model achieved a Dice coefficient of 0.80 ± 0.30 on the SISS 2015 dataset.
- The model achieved a Dice coefficient of 0.81 ± 0.33 on the ISLES 2022 dataset.
- Ablation studies confirmed the effectiveness of each architectural component.
- The model surpassed the current state-of-the-art results on both datasets.
Takeaway
This study created a new model to help doctors better see and understand strokes in brain scans, making it easier to treat patients.
Methodology
The model was evaluated on the ISLES 2015 and ISLES 2022 datasets using a combination of DenseNet, dilated convolutions, and Transformer blocks.
Limitations
The model's performance may be influenced by dataset size, image quality, and variability in stroke lesion characteristics.
Participant Demographics
Patients with ischemic stroke, with a total of 64 cases in the ISLES 2015 dataset.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website