Abstract
Accurate segmentation of enhancing tumor (ET), whole tumor (WT), and tumor core (TC) from multisequence magnetic resonance imaging (MRI) remains challenging because tumor morphology, signal characteristics, and spatial extent vary substantially across cases. We present SlimInterPAS, a focused extension of SlimUNETRV2 that modifies only the deepest encoder stage while retaining the original decoder, segmentation head, Dice–cross-entropy objective, and inference strategy. A three-dimensional axis adapter (AxisAdapter3D) exposes depth-, height-, and width-oriented responses, InterBottleneck models complementary cross-plane interactions, a position-adaptive spatial recalibration block (PASBlock) combines local and dilated contextual responses, and Gated Residual Fusion controls the incremental correction written back to the baseline representation. Evaluation is organized around controlled component ablation on the labeled training/reference set of the 2026 Brain Tumor Segmentation (BraTS) Generalizability Across Tumors (GoAT) challenge, paired per-case statistical analysis, computational-efficiency profiling, and a separate official hidden validation evaluation. The complete configuration provides the most favorable overall balance among the evaluated ablation variants while adding only a small parameter and patch-runtime overhead relative to the direct baseline. Training/reference-set evidence and hidden-set generalization are reported separately.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_GoAT_003.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=qiMpwflEI
BibTex
@InProceedings{DonJin_SlimInterPAS_MICCAISAT2026,
author = { Dong, Jinlin AND Peng, Yanjun},
title = { { SlimInterPAS for Brain Tumor Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17253},
month = {pending},
page = {pending}
}