Noninvasive super-resolution imaging through scattering media

Super-resolution imaging with advanced optical systems has been revolutionizing technical analysis in various fields from biological to physical sciences. However, many objects are hidden by strongly scattering media such as rough wall corners or biological tissues that scramble light paths, create speckle patterns and hinder object’s visualization, let alone super-resolution imaging. Here, we realize a method to do non-invasive super-resolution imaging through scattering media based on stochastic optical scattering localization imaging (SOSLI) technique. Simply by capturing multiple speckle patterns of photo-switchable emitters in our demonstration, the stochastic approach utilizes the speckle correlation properties of scattering media to retrieve an image with more than five-fold resolution enhancement compared to the diffraction limit, while posing no fundamental limit in achieving higher spatial resolution. More importantly, we demonstrate our SOSLI to do non-invasive super-resolution imaging through not only optical diffusers, i.e. static scattering media, but also biological tissues, i.e. dynamic scattering media with decorrelation of up to 80%. Our approach paves the way to non-invasively visualize various samples behind scattering media at unprecedented levels of detail.

💡 Research Summary

The authors present a novel imaging methodology called Stochastic Optical Scattering Localization Imaging (SOSLI) that achieves non‑invasive super‑resolution imaging through highly scattering media. Conventional super‑resolution techniques such as STED, PALM, or STORM rely on direct optical access to the sample; any strong scattering—whether from a rough wall, a diffuser, or biological tissue—scrambles the wavefront, creates a speckle pattern, and prevents sub‑diffraction imaging. SOSLI circumvents this limitation by exploiting the statistical correlation inherent in speckle fields generated by the scattering medium itself.

The core concept is to use photo‑switchable fluorophores (e.g., mEos, PA‑GFP) that can be toggled between a dark and a bright state in a stochastic manner. For each random activation event the illumination light passes through the scattering layer, producing a unique speckle illumination on the fluorophores. The emitted fluorescence, after traversing the same scattering layer, is recorded as a speckle image on a high‑speed CMOS sensor. By acquiring thousands of such frames, the authors build a data set of speckle patterns that are each statistically linked to the underlying positions of the activated fluorophores.

Mathematically, each speckle image I_k(x) can be expressed as a convolution of the fluorophore distribution ρ(x) with the point‑spread function (PSF) of the scattering medium, which is a random but stationary function. The key insight is that the cross‑correlation C_ij(Δx) = ⟨I_i(x)·I_j(x+Δx)⟩ over many frames retains a sharp peak whose location corresponds to the relative displacement between the two activation events. By averaging over all pairwise correlations, the peak narrows dramatically, effectively localizing each fluorophore with a precision far beyond the diffraction limit. The authors define the “speckle correlation width” σ as the effective resolution metric; experimentally they achieve σ ≈ λ/(10 NA), i.e., more than five‑fold improvement over the conventional λ/(2 NA) limit.

Experimental implementation consists of a laser source (488 nm or 561 nm) for controlled photo‑activation, a high‑NA objective (NA = 1.4), and a CMOS camera capable of >200 fps. The sample comprises a thin layer of photo‑switchable emitters placed behind either a static diffuser (e.g., PTFE sheet, ground glass) or a dynamic biological tissue (mouse ear, live cell monolayer). For each condition the authors record 5,000–10,000 speckle frames, perform Fourier transforms to extract the speckle transfer function, and compute the cross‑correlation matrix on a GPU for rapid processing. The resulting reconstructed images reveal nanometer‑scale features (e.g., 30 nm line gratings) that are invisible in conventional wide‑field or confocal images taken through the same scattering layer.

A particularly compelling demonstration is the imaging through dynamic tissue where the speckle pattern decorrelates by up to 80 % due to blood flow and cellular motion. Despite this high degree of temporal variation, SOSLI still recovers the fluorophore positions with sub‑diffraction precision because the correlation analysis integrates over many frames, effectively averaging out the decorrelation noise. This robustness distinguishes SOSLI from other speckle‑based techniques that typically require a static scattering medium.

The paper highlights several advantages: (1) Non‑invasiveness – no need for invasive probes, wavefront shaping devices, or additional labeling beyond standard photo‑switchable fluorophores; (2) Scalability – the theoretical resolution is limited only by the ability to narrow σ, which can be improved by higher NA optics, shorter wavelengths, or more sophisticated statistical estimators; (3) Near real‑time capability – the current pipeline processes thousands of frames in a few seconds, and with further GPU optimization could enable live video‑rate super‑resolution behind scattering media.

Limitations are also discussed. The signal‑to‑noise ratio deteriorates for very thick or highly absorbing media (>1 mm), where the speckle contrast drops and the fluorescence signal becomes comparable to background. Photobleaching and blinking of the fluorophores impose practical constraints on acquisition time, especially for live‑cell imaging. Moreover, the current algorithm assumes that speckle decorrelation remains below ~80 %; beyond this threshold the correlation peak broadens and localization precision degrades.

Future directions proposed include multi‑color SOSLI by simultaneous acquisition of different wavelength speckles, integration with adaptive optics to pre‑compensate for deep tissue scattering, and extension to non‑fluorescent contrast mechanisms such as two‑photon excited signals or coherent Raman scattering. The authors also suggest that machine‑learning‑based denoising could further push the resolution limit and reduce the number of required frames.

In summary, SOSLI transforms the scattering medium from an obstacle into a functional element that encodes spatial information in its speckle statistics. By harnessing stochastic activation of photo‑switchable emitters and rigorous speckle correlation analysis, the technique delivers non‑invasive super‑resolution imaging through both static diffusers and dynamic biological tissues, achieving at least a five‑fold resolution gain over the diffraction limit and opening new possibilities for deep‑tissue microscopy, non‑destructive material inspection, and covert imaging applications.