Abstract
Magnetic Resonance Imaging (MRI) provides excellent soft-tissue contrast, but long acquisition times remain a major limitation in clinical practice. Accelerated MRI relies on undersampled k-space data, resulting in ill-posed reconstruction problems that can degrade fine anatomical structures and high-frequency details. Although deep learning-based reconstruction methods have shown promising performance, commonly used losses, including the Structural Similarity Index (SSIM), often under-emphasize high-frequency components and may lead to over-smoothed reconstructions. In this work, we propose a Patch-wise Wavelet Loss (PWL) for accelerated MRI reconstruction to improve local detail preservation. The proposed loss decomposes image patches using a multi-level wavelet transform and applies a weighted L1 penalty to high-frequency sub-bands, encouraging the reconstruction of edges, textures, and small anatomical structures. Experiments on the fastMRI datasets show consistent improvements across PromptMR and HUMUS-Net. On the knee dataset, the proposed method achieves SSIM values of 0.9332 and 0.9301, respectively. Qualitative results further demonstrate sharper reconstructions and improved structural fidelity. The proposed loss is architecture-agnostic and can be integrated into existing deep learning-based MRI reconstruction frameworks without modifying the reconstruction architecture or introducing additional inference-time parameters. Accelerated MRI reconstruction Wavelet loss Patch-wise supervision High-frequency detail preservation Deep learning
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/RIME_009.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=1FJxoLkKvl
BibTex
@InProceedings{RazNar_AHierarchical_MICCAISAT2026,
author = { Razizadeh, Narges AND Hareendranathan, Abhilash},
title = { { A Hierarchical Patch-wise Wavelet Loss for High-Frequency Detail Preservation in Accelerated MRI Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}