| Summary: | Multi-modal Magnetic Resonance Imaging sequences along with 4D Perfusion Weighted Imaging scans provide important information for stroke lesion outcome prediction. However, the proposed methodologies until now were not able to discriminate correctly the most informative features from the less useful ones. In this work, we propose an enhanced version of a data fusion method for stroke tissue outcome prediction by employing attention models. We compare our proposal with two other recent attention mechanisms for image segmentation, showing that all of them improved over the baseline in most metrics. However, our proposal also improved all distance metrics, which indicates a reduction in false positive detections far from the lesion.
|
|---|