跳转至

λSplit：用于荧光显微成像的自监督内容感知光谱解混模型

文章背景与核心概要

在荧光显微成像中，光谱解混旨在从混合发射的光谱图像中恢复单个荧光团的浓度。传统方法通常基于逐像素的最小二乘法拟合，但在面对高度重叠的发射光谱或高噪声环境时，其性能会显著下降。为了解决这一问题，研究人员提出了 \(\lambda\)Split，这是一种结合物理先验的深度生成模型。

\(\lambda\)Split 利用分层变分自编码器（VAE）学习浓度图的条件分布，并通过完全可微的光谱混合器（Spectral Mixer）确保与图像形成过程的一致性。该方法不仅实现了最先进的解混效果，还具备隐式去噪能力。由于其兼容标准共聚焦显微镜且无需硬件改造，该技术在处理高噪声、光谱重叠严重及低光谱维度数据时表现出极强的鲁棒性。

概述与总结

\(\lambda\)Split 是一款物理信息驱动的深度生成模型，旨在解决荧光显微成像中的光谱解混难题。

\(\lambda\)Split is a physics-informed deep generative model designed to tackle the challenge of spectral unmixing in fluorescence microscopy.

  • 问题所在： 经典的光谱解混方法采用逐像素最小二乘法拟合，当发射光谱高度重叠或噪声水平较高时，这些方法往往失效。现有的基于学习的成像方法要么不适用于显微镜数据，要么过于狭窄，缺乏通用性。

    • The Problem: Classical spectral unmixing methods operate pixel-wise using least-squares fitting, which fails when emission spectra overlap heavily or when noise levels are high. Existing learning-based imaging approaches are either unsuited for microscopy data or too narrowly specialized.

  • 解决方案： \(\lambda\)Split 利用分层变分自编码器 (VAE) 来学习浓度图的条件分布。它集成了一个完全可微的光谱混合器，以符合真实的图像形成过程，从而提供最先进的解混效果和隐式噪声去除。

    • The Solution: \(\lambda\)Split utilizes a hierarchical Variational Autoencoder (VAE) to learn a conditional distribution over concentration maps. It integrates a fully differentiable Spectral Mixer to align with the true image formation process, providing state-of-the-art unmixing and implicit noise removal.

  • 核心优势： \(\lambda\)Split 兼容标准共聚焦显微镜（无需专门的硬件改造），在处理高噪声、显著光谱重叠和降低光谱维度的情况下，均表现出稳健的性能。

    • Key Advantages: Compatible with standard confocal microscopes (requiring no specialized hardware modifications), \(\lambda\)Split demonstrates robust performance across high-noise regimes, significant spectral overlap, and reduced spectral dimensionality.

论文元数据

  • arXiv ID: arXiv:2603.23647 [cs.CV]
  • 主要学科： 计算机视觉与模式识别 (cs.CV)
  • 次要学科： 人工智能 (cs.AI)，机器学习 (cs.LG)
  • 作者： Federico Carrara, Talley Lambert, Mehdi Seifi, Florian Jug
  • 会议/状态： 已被 ECCV 2026 接收（正文 14 页，补充材料 25 页，16 张图，14 张表）
  • 提交时间线：
  • 提交日期：2026 年 3 月 24 日 (v1)
  • 最后修订：2026 年 8 月 6 日 (v3)

Paper Metadata

  • arXiv ID: arXiv:2603.23647 [cs.CV]
  • Primary Subject: Computer Vision and Pattern Recognition (cs.CV)
  • Secondary Subjects: Artificial Intelligence (cs.AI), Machine Learning (cs.LG)
  • Authors: Federico Carrara, Talley Lambert, Mehdi Seifi, Florian Jug
  • Venue / Status: Accepted at ECCV 2026 (14 pages main, 25 pages supplement, 16 figures, 14 tables)
  • Submission Timeline:
  • Submitted: March 24, 2026 (v1)
  • Last Revised: August 6, 2026 (v3)

摘要

在荧光显微成像中，光谱解混旨在从捕获混合荧光团发射的光谱图像中恢复单个荧光团的浓度。由于传统方法是逐像素操作并依赖于最小二乘法拟合，其性能会随着发射光谱重叠的增加和噪声水平的提高而下降，这表明能够学习并利用结构先验的数据驱动方法可能会带来更好的结果。

In fluorescence microscopy, spectral unmixing aims to recover individual fluorophore concentrations from spectral images that capture mixed fluorophore emissions. Since classical methods operate pixel-wise and rely on least-squares fitting, their performance degrades with increasingly overlapping emission spectra and higher levels of noise, suggesting that a data-driven approach that can learn and utilize a structural prior might lead to improved results.

现有的光谱成像学习方法虽然存在，但它们要么未针对显微镜数据进行优化，要么是针对不适用于荧光显微成像环境的特定情况开发的。为了解决这个问题，我们提出了 \(\lambda\)Split，这是一种结合物理先验的深度生成模型，它使用分层变分自编码器学习浓度图的条件分布。一个完全可微的光谱混合器强制执行与图像形成过程的一致性，而学习到的结构先验则实现了最先进的解混和隐式去噪。

Learning-based approaches for spectral imaging do exist, but they are either not optimized for microscopy data or are developed for very specific cases that are not applicable to fluorescence microscopy settings. To address this, we propose \(\lambda\)Split, a physics-informed deep generative model that learns a conditional distribution over concentration maps using a hierarchical Variational Autoencoder. A fully differentiable Spectral Mixer enforces consistency with the image formation process, while the learned structural priors enable state-of-the-art unmixing and implicit noise removal.

我们在 3 个真实世界数据集上演示了 \(\lambda\)Split，这些数据集被合成为总共 66 个具有挑战性的光谱解混基准。我们将我们的结果与包括经典方法和一系列基于学习的方法在内的总共 10 种基准方法进行了比较。结果一致表明，在处理高噪声、光谱重叠严重或光谱维度降低的情况下，该方法具有竞争力的性能和更高的鲁棒性，使 \(\lambda\)Split 成为荧光显微成像数据光谱解混的最先进技术。重要的是，\(\lambda\)Split 与标准共聚焦显微镜产生的光谱数据兼容，无需专门的硬件修改即可立即采用。

We demonstrate \(\lambda\)Split on 3 real-world datasets that we synthetically cast into a total of 66 challenging spectral unmixing benchmarks. We compare our results against a total of 10 baseline methods, including classical methods and a range of learning-based methods. Our results consistently show competitive performance and improved robustness in high noise regimes, when spectra overlap considerably, or when the spectral dimensionality is lowered, making \(\lambda\)Split a new state-of-the-art for spectral unmixing of fluorescent microscopy data. Importantly, \(\lambda\)Split is compatible with spectral data produced by standard confocal microscopes, enabling immediate adoption without specialized hardware modifications.

全文与资源

Full-Text & Resources

外部参考与工具

External References & Tools