We participated in the SynthRAD2025 challenge (Tasks 1 and 2) with a unified pipeline for synthetic CT (sCT) generation from MRI and CBCT, implemented using the KonfAI framework. Our model is a 2.5D U-Net++ with a ResNet-34 encoder, trained jointly across anatomical regions and fine-tuned per region. The loss function combined pixel-wise L1 loss with IMPACT-Synth, a perceptual loss derived from SAM and TotalSegmentator to enhance structural fidelity. Training was performed using AdamW (initial learning rate = 0.001, halved every 25k steps) on patch-based, normalized, body-masked inputs (320x320 for MRI, 256x256 for CBCT), with random flipping as the only augmentation. No post-processing was applied. Final predictions leveraged test-time augmentation and five-fold ensembling. The best model was selected based on validation MAE. Two registration strategies were evaluated: (i) Elastix with mutual information, consistent with the challenge pipeline, and (ii) IMPACT, a feature-based similarity metric leveraging pretrained segmentation networks. On the local test sets, IMPACT-based registration achieved more accurate and anatomically consistent alignments than mutual-information-based registration, resulting in improved sCT synthesis with lower MAE and more realistic anatomical structures. On the public validation set, however, models trained with Elastix-aligned data achieved higher scores, reflecting a registration bias favoring alignment strategies consistent with the evaluation pipeline. This highlights how registration errors can propagate into supervised learning, influencing both training and evaluation, and potentially inflating performance metrics at the expense of anatomical fidelity. By promoting anatomically consistent alignment, IMPACT helps mitigate this bias and supports the development of more robust and generalizable sCT synthesis models.

翻译：我们使用KonfAI框架实现了一个从MRI和CBCT生成合成CT（sCT）的统一流程，参与了SynthRAD2025挑战赛（任务1和2）。我们的模型是一个采用ResNet-34编码器的2.5D U-Net++，在解剖区域上进行联合训练，并针对每个区域进行微调。损失函数结合了逐像素L1损失与IMPACT-Synth——一种源自SAM和TotalSegmentator的感知损失，以增强结构保真度。训练使用AdamW优化器（初始学习率=0.001，每25k步减半），在基于图像块、经过归一化、身体掩码处理的输入（MRI为320x320，CBCT为256x256）上进行，随机翻转是唯一的增强方式。未应用任何后处理。最终预测利用了测试时增强和五折集成。最佳模型根据验证集MAE选择。我们评估了两种配准策略：（i）使用互信息的Elastix，与挑战赛流程一致；（ii）IMPACT，一种利用预训练分割网络的特征相似性度量。在本地测试集上，基于IMPACT的配准比基于互信息的配准实现了更准确且解剖结构更一致的配准，从而改善了sCT合成效果，获得了更低的MAE和更真实的解剖结构。然而，在公共验证集上，使用Elastix配准数据训练的模型获得了更高的分数，这反映出一种配准偏差，即偏向与评估流程一致的配准策略。这突显了配准误差如何传播到监督学习中，影响训练和评估，并可能以牺牲解剖保真度为代价来夸大性能指标。通过促进解剖结构一致的配准，IMPACT有助于减轻这种偏差，并支持开发更鲁棒、更可泛化的sCT合成模型。