Abstract
Magnetic Resonance Imaging (MRI) acquisition routinely requires technologists to adapt volumetric coverage to each patient’s anatomy. These changes can alter SNR (signal-to-noise ratio), contrast, and scan time. Expert MR technologists adapt MR parameters to bring these quantities within desired limits; as, patient- and hardware-dependent limits do not allow simple closed-form adaptations, limiting wider adoption of MRI.
We present CORTEX, Continuous Optimization via Reinforcement for Targeted Examination eXcellence, a deep reinforcement learning (DRL) framework for constrained MRI protocol adaptation. An actor–critic agent observes the protocol context and updates controllable parameters to preserve scan time, contrast, and SNR.
During training, a vendor-exposed API is called at each transition to compute patient and hardware limit compliant MR parameter values. Targeted ablations study network depth and residual connections under the selected reward function.
On common brain MR protocols using 2D Fast Spin Echo (2D FSE), CORTEX achieved 97.0\% convergence (The ratio between the reference and adapted protocols shows SNR > $0.80$ and scan time < $1.20$.) with 100\% of parameters compliant with patient and hardware limits. The MAE/RMSE of SNR ratio and scan-time ratio were 0.079/0.095 and 0.137/0.170, respectively suggesting clinically usability of adapted protocol.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_020.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_020_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=79tmmw4BJ4
BibTex
@InProceedings{SheNee_CORTEX_MICCAISAT2026,
author = { Sheth, Neem AND Agarwal, Harsh AND Venkatesan, Ramesh},
title = { { CORTEX: Deep Reinforcement Learning with Pulse-Sequence Computation for Resource-Efficient MRI Protocol Adaptation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}