研究人员开发了一种新颖的框架，用于生成可控的4D心脏MRI序列，解决了标注数据不足和领域偏移的限制。该系统利用半监督变分自编码器学习解剖表示，并利用级联潜在扩散模型将解剖结构与运动解耦。这种方法可以精确控制静态解剖结构并确保时间连贯性，从而在扩充训练数据时提高下游分割性能。 AI
影响 该框架通过实现更强大的数据增强和跨不同设备的更好泛化能力，有望显著改进4D医学成像AI模型的开发。
排序理由 该集群描述了一篇关于用于医学图像合成的新型AI框架的新研究论文。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2606.26764v1 Announce Type: cross Abstract: Developing robust artificial intelligence models for 4D (3D + time) medical imaging is constrained by limited annotated data, inter-device domain shifts, and privacy restrictions. To address this, we propose a 4D controllable gene…
Developing robust artificial intelligence models for 4D (3D + time) medical imaging is constrained by limited annotated data, inter-device domain shifts, and privacy restrictions. To address this, we propose a 4D controllable generative framework for anatomically consistent data …
Developing robust artificial intelligence models for 4D (3D + time) medical imaging is constrained by limited annotated data, inter-device domain shifts, and privacy restrictions. To address this, we propose a 4D controllable generative framework for anatomically consistent data …