Abstract
Accurate brain tumor segmentation typically relies on four MRI modalities, which are not always available in clinical practice. Although recent methods for brain tumor segmentation with missing MRI modalities have achieved strong performance on pre-operative MRI benchmark datasets, their generalization under intra-operative MRI conditions remains largely unexplored. Intra-operative MRI differs substantially from pre-operative imaging due to surgically induced anatomical deformation, resection-cavity artifacts, and frequently incomplete modality acquisitions. In this study, we present a systematic approach for evaluating four representative models that cover diverse architectural paradigms widely explored in missing-modality segmentation: mmFormer, M3AE, IM-Fuse, and D3Seg. These models were firstly trained and tested on two pre-operative datasets (BraTS and ReMIND) and then tested on intraoeprative MRI dataset (ReMIND). Quantitative results show substantial performance degradation for all models when applied to intra-operative MRI, particularly for tumor core segmentation. A complementary qualitative analysis further reveals model-specific failure patterns, including edema over-segmentation, misclassification of ventricular regions, and consistently challenging resection cavity delineation across all four models. Among the evaluated models, D3Seg and mmFormer show comparatively better performance under intra-operative conditions. The code and model weights are available at https://github.com/danishali6421/Pre-Op-to-Intra-Op.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BrainWorks_013.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/profile?id=~Danish_Ali1
BibTex
@InProceedings{AliDan_PreOperative_MICCAISAT2026,
author = { Ali, Danish AND Azimi, Amirali AND Galvin, Colin Patrick AND Kapur, Tina AND Golby, Alexandra AND Mian, Ajmal AND Akhtar, Naveed AND Hassan, Ghulam Mubashar},
title = { { Pre-Operative to Intra-Operative: A Brain Tumor Segmentation Generalization Study } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17254},
month = {pending},
page = {pending}
}