{
"display_name": "Synthesis: Finetune BICMAC",
"short_description": "Supervised whole-body CBCT-to-sCT model for SynthRAD Task 2.
Training data:
929 paired CT-CBCT cases from SynthRAD 2023+2025.
How to cite:
V. Boussot et al., Why Registration Quality Matters: Enhancing sCT Synthesis with IMPACT-Based Registration, arXiv preprint arXiv:2510.21358, https://doi.org/10.48550/arXiv.2510.21358.",
"description": "Description:
Supervised whole-body CBCT-to-sCT model trained using SynthRAD 2023 and SynthRAD 2025 Task 2 datasets, distributed as five validation-fold checkpoints (CV_0.pt to CV_4.pt). Training pairs are aligned with IMPACTReg. The available Task 2 data include 291 patients from SynthRAD 2023 and 638 patients from SynthRAD 2025, for 929 patients in total.
Architecture and loss:
2.5D UNet++ with a ResNet34 encoder, trained with KonfAI and optimized using IMPACT-Synth, a perceptual loss leveraging semantic priors from SAM 2.1-s.
Training data:
Whole-body paired CT-CBCT volumes come from SynthRAD 2023 Task 2 (brain, pelvis) and SynthRAD 2025 Task 2 (AB, HN, TH). Additional OOD evaluation cohorts are available for AB_ood, HN_ood, and TH_ood. Pairs are aligned with IMPACTReg, and the associated transform files are available at synthrad2023-impact-registration and synthrad2025-impact-registration. The files CrossValidation_0.txt to CrossValidation_4.txt correspond to validation folds for the distributed checkpoints, and Validation.txt contains 150 release-validation patients.
Release performance (CV overall, n=150):
Dice 0.758, MAE 53.55 HU, SSIM 0.939, PSNR 32.09 dB, SAM 13.70, Reg 1.04, Uncertainty 784.24.
Inference note:
The prediction pipeline supports test-time augmentation and writes an averaged InferenceStack output that can also be used for uncertainty estimation.
How to cite:
V. Boussot et al., Why Registration Quality Matters: Enhancing sCT Synthesis with IMPACT-Based Registration, arXiv preprint arXiv:2510.21358, https://doi.org/10.48550/arXiv.2510.21358.",
"tta": 2,
"mc_dropout": false,
"models": [
"CV_0.pt",
"CV_1.pt",
"CV_2.pt",
"CV_3.pt",
"CV_4.pt"
],
"inputs": {
"CBCT": {
"display_name": "CBCT",
"volume_type": "VOLUME",
"required": true
}
},
"outputs": {
"sCT": {
"display_name": "sCT",
"volume_type": "VOLUME",
"required": true
}
},
"inputs_evaluations": {
"Image": {
"Evaluation.yml": {
"sCT": {
"display_name": "sCT",
"volume_type": "VOLUME",
"required": true
},
"CT": {
"display_name": "CT",
"volume_type": "VOLUME",
"required": true
}
}
}
},
"vram_plan": {
"8": {
"patch_size": [
1,
512,
512
],
"batch_size": 16
},
"16": {
"patch_size": [
1,
512,
512
],
"batch_size": 28
},
"24": {
"patch_size": [
1,
512,
512
],
"batch_size": 48
}
}
}