Abstract
Cranial implant design for cranioplasty remains a time-intensive task requiring specialized expertise, often leading to delays in treating traumatic brain injuries (TBI). While automating the implant design shows promise, there is a scarcity of diverse, annotated surgical datasets and models that generalize across different clinical domains (different acquitision machines and image processing techniques). We propose a pipeline to bridge this gap by enabling the generation of a low computational cost and large-scale, surgical training data from standard, nondefective CT scans. We introduce CranioImpInp, an end-to-end frame work for AI-assisted implant inpainting in surgical planning. The pipeline starts with raw CT scans, which are segmented using the open-weights TotalSegmentator model. To address data scarcity, we developed a defect generation algorithm that injects realistic, diverse defects into binarized skulls, serving as a reproducible research resource for dataset centric community building. Augmented datasets were used to train a 3D Conditional Wavelet Diffusion Model (cWDM), which leverages wavelet transforms for computationally efficient 3D inpainting without the need for intensive GPU-based latent space training. Our model was trained using the augmented CQ500 dataset and evaluated against the Skull Break benchmark. The model SBCQ10 achieved a Dice Similarity Coefficient (DSC) of 90.3% and a 95th percentile Hausdorff Distance (HD95) of 2.15mm on the SkullBreak test set. Crucially, the data augmenta tion pipeline enabled superior generalization to out-of-distribution data, achieving 92.7% DSC and 6.31mm HD95 on the CQ500 test set, where the non-augmented baseline performed worse. Qualitative evaluation on three real clinical TBI cases demonstrated high-quality surgical planning proposals. This work provides a scalable framework for the usability of non-defective head CT data. By providing a method to transform public CT datasets into robust training resources, we move toward generalizable AI software that deceases the heavy workload in cranioplasty surgery planning.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COLAS_012.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=yQgDTVCC7V
BibTex
@InProceedings{BriRic_From_MICCAISAT2026,
author = { Brioso, Ricardo Coimbra AND Bieder, Florentin AND Cattin, Philippe},
title = { { From CT to Clinical Cranioplasty Inpainting using a Conditional Wavelet Diffusion Model } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS pending},
month = {pending},
page = {pending}
}