Abstract
Segmentation models trained with supervised methods on a single brain tumor type often fail to generalize to unseen tumor types, scanners, and patient populations. In this study, we present a systematic analysis of four model variants for the BraTS-GoAT 2026 challenge, obtained by combining two representative 3D backbones (3D U-NET and Swin UNETR). The models were trained using supervised learning on labeled data and self-supervised learning (SSL), with pretraining on unlabeled data followed by fine-tuning on labeled data. We also applied a prediction-time pipeline that includes test-time augmentation and post-processing operations. On the held-out validation set, the SSL-based models consistently demonstrated improved generalization across both backbones. Among them, the SSL-based Swin UNETR model achieved the best performance, with Dice Similarity Coefficient and Normalized Surface Distance values of 0.830 and 0.420 for the whole tumor, 0.765 and 0.438 for the tumor core, and 0.742 and 0.492 for the enhancing tumor, respectively. This study demonstrates that self-supervised learning yields more robust and generalized multi-tumor brain MRI segmentation. Code for reproducing our results is publicly available at https://github.com/sanskriti21mnnit/BraTS-GoAT-2026
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_GoAT_011.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=0pGWxs5Zf3
BibTex
@InProceedings{SriSan_SelfSupervised_MICCAISAT2026,
author = { Srivastava, Sanskriti AND Maurya, Satyajit AND Raghuwanshi, Kuldeep AND Singh, Anup},
title = { { Self-Supervised Pretraining for Generalizable Brain Tumor Segmentation in the BraTS-GoAT Challenge } },
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}
}
