---
title: "Self-Supervised Implicit CEST Reconstruction via Physics-Informed Lorentz Encoding"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.06132v1"
permalink: "https://feed.astralor.com/items/itm_f61ee42a7ab52ca40839e356"
published: "2026-07-07T10:49:07.000Z"
score: 78
channel: "finance"
tags: ["CEST MRI","Lorentz Encoding","Medical Imaging","Physics-Informed Neural Networks","Self-Supervised Learning","Implicit Neural Representations"]
---

# Self-Supervised Implicit CEST Reconstruction via Physics-Informed Lorentz Encoding

## AI commentary

> This paper stands out by elegantly blending self-supervised learning with rigorous physical models rather than simply stacking neural networks. For researchers in medical imaging AI, LE offers an elegant paradigm, particularly suited for resource-constrained clinical scenarios. CEST MRI is a powerful tool for metabolic imaging, and LE removes its acquisition speed bottleneck — great news for radiologists and patients alike.

## At a glance

Chemical Exchange Saturation Transfer (CEST) MRI provides valuable metabolic information but is clinically limited by long acquisition times. Although sparse sampling reduces scanning time, reconstructing high-resolution Z-spectra from limited data remains an ill-posed inverse problem. This paper proposes Lorentz Encoding (LE), a physics-informed framework that formulates CEST reconstruction as a self-supervised task via implicit continuous coordinate learning. LE regularizes spectral mapping by projecting sparse coordinates into a physically constrained space governed by parametric Lorentzian profiles with learnable basis functions, effectively reducing noise and enforcing physical consistency. Experiments on in vivo human brain data demonstrate significant improvement: under 39-point sampling, LE achieves a PSNR of 57.58 dB and an SSIM of 0.9994, outperforming state-of-the-art methods and enabling accurate quantitative metabolite mapping (APT, NOE, MT).

## Key points

- Propose Lorentz Encoding (LE), modeling CEST reconstruction as self-supervised implicit neural representation learning with physics-informed constraints.
- LE projects sparse sampling coordinates into a physically constrained space using learnable Lorentzian basis functions, suppressing artifacts and invalid signals.
- Achieves PSNR of 57.58 dB and SSIM of 0.9994 on in vivo human brain data with 39-point sampling, significantly outperforming existing methods.
- Learned encodings form a continuous, geometrically ordered latent space trajectory, enabling accurate quantitative metabolite mapping (APT, NOE, MT).

## Full text

Multi-Pool Chemical Exchange Saturation Transfer (CEST) MRI provides valuable metabolic information but is clinically limited by long acquisition times. Although sparse sampling reduces scanning time, reconstructing high-resolution Z-spectra from limited data remains an ill-posed inverse problem. Conventional interpolation and generic Implicit Neural Rep-resentations (INRs) often lack physical constraints, leading to spectral artifacts and physically invalid signals. To address this, we propose Lorentz Encoding (LE), a physics-informed framework that formulates CEST reconstruction as a self-supervised reconstruction task via implicit continuous coordinate learning. Unlike generic positional encodings, LE regularizes the continuous spectral mapping by projecting sparse coordinates into a physically constrained space governed by a combination of parametric Lorentzian profiles with learnable basis functions. This mechanism effectively reduces noise and enforces consistency with physical models. Experiments on in vivo human brain data demonstrate that LE significantly outperforms state-of-the-art methods. Specifically, under a 39-point sampling strategy, LE achieves a PSNR of 57.58 dB and an SSIM of 0.9994. Furthermore, the learned physics-informed encodings form a continuous, geometrically ordered trajectory in the latent space, ensuring accurate quantitative metabo-lite mapping (APT, NOE, MT).

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_f61ee42a7ab52ca40839e356