Abstract
Accelerated MRI acquisition reduces scan time but risks degrading image quality in clinically relevant regions. Existing adaptive sampling methods optimize k-space line selection within individual slices, ignoring the volumetric structure of clinical acquisitions where pathology spans only a subset of slices. We propose a volume-level Reinforcement Learning (RL) framework that learns to allocate acquisition budget across slices of a 3D MRI volume, prioritizing slices containing abnormalities. Our approach formulates the acquisition as a sequential decision process: a reinforcement learning agent observes per-slice lesion confidence scores and sampling fractions, then selects which slice should receive additional k-space lines at each step. We evaluate on the fastMRI knee dataset with various knee pathology annotations at 2×–8× acceleration, comparing against random uniform sampling and methods that optimize sampling based on reconstruction quality, detection rate, or disease severity. Our agent attains the highest lesion-region SSIM at the clinically common 2×–4× factors and remains competitive at 8×, improving over random uniform sampling by +0.041 to +0.078 SSIM without any access to lesion annotations at test time.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_061.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=7np1k2QHJJ
BibTex
@InProceedings{HimMev_InterSlice_MICCAISAT2026,
author = { Himmetoglu, Meva AND Konukoglu, Ender},
title = { { Inter-Slice Acquisition Budget Allocation for Pathology Aware Accelerated MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}