Synthesizing healthy brain scans from diseased brain scans offers a potential solution to address the limitations of general-purpose algorithms, such as tissue segmentation and brain extraction algorithms, which may not effectively handle diseased images. We consider this a 3D inpainting task and investigate the adaptation of 2D inpainting methods to meet the requirements of 3D magnetic resonance imaging(MRI) data. Our contributions encompass potential modifications tailored to MRI-specific needs, and we conducted evaluations of multiple inpainting techniques using the BraTS2023 Inpainting datasets to assess their efficacy and limitations.

翻译：从病变脑部扫描图像合成健康脑部扫描图像，为解决通用算法（如组织分割和脑提取算法）在处理病变图像时可能存在的局限性提供了潜在方案。我们将此视为三维修复任务，并研究了二维修复方法如何适应三维磁共振成像（MRI）数据的需求。我们的贡献包括针对MRI特定需求所进行的潜在改进，同时利用BraTS2023图像修复数据集对多种修复技术进行了评估，以衡量其有效性与局限性。