15 DAILY MICCAI Wednesday Luyi Han module to generate residual attentional features. Then, we add the basic features with the attentional residual features together, and we can force it to reconstruct the target sequence. With these residual attentional features or without these attentional features, we can get two reconstructed images.” A task-specific enhanced map (TSEM) is proposed by subtracting these two reconstructed images. This map works like an attention map to identify which regions do not synthesize well in the image but at the pixel level. Unlike uncertainty, which requires multi-time outputs, it can be obtained in one inference. Looking to the future, Luyi tells us that the implications of this research extend beyond multisequence MRI synthesis. The model’s versatility allows it to be adapted for other tasks, such as segmentation or classification. “We also want to use this model on multi-sequence breast MRI,” Luyi reveals. “My PhD project is on using multi-sequence breast MRI to predict the response to neoadjuvant therapy, so I think this is the next step for this paper. My last paper was about longitudinal registration on breast MRI and is still under review. Another work is on 3D/4D multi-sequence MRI synthesis – not only on the breast but also on the brain. That’s also under review and is the basis of this paper.” To learn more about Luyi’s work, visit Poster 6 this afternoon at 14:30-16:00 in the Poster Hall.
RkJQdWJsaXNoZXIy NTc3NzU=