# Why Registration Quality Matters: Enhancing sCT Synthesis with IMPACT-Based Registration

arXiv: https://arxiv.org/abs/2510.21358 · HTML: https://mlsanity.com/paper/2510.21358/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 13 references that are in the map.
- First posted on arXiv 24 Oct 2025.

## Builds on (most cited first, 7 of 7)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [UNet++: A Nested U-Net Architecture for Medical Image Segmentation](https://mlsanity.com/paper/1807.10165/) (2018)
- [Deep learning-based synthetic-CT generation in radiotherapy and PET: a review](https://mlsanity.com/paper/2102.02734/) (2021)
- [SynthRAD2025 Grand Challenge dataset: generating synthetic CTs for radiotherapy](https://mlsanity.com/paper/2502.17609/) (2025)
- [IMPACT: A Generic Semantic Loss for Multimodal Medical Image Registration](https://mlsanity.com/paper/2503.24121/) (2025)
- [KonfAI: A Modular and Fully Configurable Framework for Deep Learning in Medical Imaging](https://mlsanity.com/paper/2508.09823/) (2025)

## Built on by (most cited first, 4 of 4)

- [S1-Omni-Image: A Unified Model for Scientific Image Understanding, Generation, and Editing](https://mlsanity.com/paper/2606.24441/) (2026)
- [Eliminating Registration Bias in Synthetic CT Generation: A Physics-Based Simulation Framework](https://mlsanity.com/paper/2602.02130/) (2026)
- [S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation](https://mlsanity.com/paper/2607.15686/) (2026)
- [When Misalignment Becomes Supervision: Structured Label Noise in Supervised Synthetic CT Generation](https://mlsanity.com/paper/2609.29387/) (2026)

## Abstract (arXiv)

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.