Raman microscopy， a label-free detection technique that identifies targets through the intrinsic vibrational spectra of samples， is highly significant in biomedicine， especially for early disease diagnosis. Its ability to avoid exogenous labels offers distinct advantages when studying biological tissues. However， biological tissues and complex materials have strong optical scattering properties， which disrupt light propagation by randomizing their paths. This severely limits the detection depth of traditional Raman techniques—often restricting imaging to superficial layers， rarely exceeding 100 μm in most biological tissues—and impairs signal contrast， as scattered light mix with target signals， resulting in blurred or noisy outputs. To tackle these critical limitations， this study presents an innovative wavefront shaping method based on spectrally resolved variance optimization.​

The core of this method is establishing a sophisticated wavefront feedback mechanism that combines Raman spectroscopy with speckle variance analysis. The detailed process is as follows： First， the experimental sample is carefully designed to simulate in vivo conditions， consisting of Raman-active particles embedded in a scattering medium. This medium is engineered to replicate the forward scattering characteristics of biological tissues， such as those in epithelial layers or soft tissues， ensuring the results are relevant to real-world applications. The Raman particles， chosen for their biological significance， show a unique and distinct Raman peak at 1 200 cm⁻¹， a wavelength range less prone to autofluorescence interference in biological samples. Using an imaging spectrometer with high spatial and spectral resolution， the system simultaneously captures both the spatial distribution （2D speckle patterns） and spectral signatures of Raman signals. A custom-developed algorithm then filters these data to isolate speckle signals specifically at the 1 200 cm⁻¹ peak. This targeted selection minimizes interference from background noise in non-target wavebands—like autofluorescence from cellular components or stray light—thus extracting speckle contrast parameters suitable for quantitative analysis.​

Next， an iterative optimization algorithm is used to build a closed-loop feedback system， a key innovation of this approach. In each iteration， the Spatial Light Modulator （SLM）， a device with high phase modulation precision， dynamically adjusts the wavefront of the input light. This adjustment causes changes in the excited speckles， rearranging their distribution within the scattering medium. At the same time， the system monitors Raman speckle variance in real time， with maximizing variance as the main objective. By continuously refining the phase modulation depth of the SLM—adjusting individual pixels to counteract wavefront distortions from the scattering medium—the input light is gradually guided to focus on a single Raman particle in the medium. The optimization process goes through 3 000 iterations， a number determined by preliminary tests to balance convergence speed and accuracy， ensuring stable and reliable focusing of excitation energy on the target particle. A critical point for this enhancement is the strict linear relationship between the intensity of spontaneous Raman signals and the excitation light intensity： when the focused excitation light reaches maximum intensity， the Raman signal is proportionally amplified to its peak level. Additionally， to address potential focal deformation caused by light absorption by Raman particles—a common issue in high-intensity focusing—a pair of galvanometers rotate the incident light at small angles （within the optical memory effect range）. Using the memory effect principle， where speckles translate as a whole without changing their distribution under small-angle rotations， the system obtains clear， undistorted focused spots from the translated speckle patterns.​

Experimental results confirm the effectiveness of this method　it achieves a 6-fold enhancement of Raman signals in scattering media， outperforming many existing techniques. The spatial resolution reaches the micrometer scale， allowing precise targeting of individual particles—essential for distinguishing between cellular or subcellular structures in biological samples. Furthermore， the method enables non-invasive focusing on single Raman particles with high specificity： background noise outside the 1200 cm⁻¹ peak is enhanced by only~30%， showing effective suppression of non-target signals. This specificity comes from the spectral filtering step， which limits optimization to the target Raman peak.​

In summary， this method provides a new technical solution for enhancing weak Raman signals in scattering environments， addressing long-standing challenges in deep-tissue imaging. Its ability to support label-free imaging in deep biological tissues， along with high signal enhancement and precise spatial focusing， makes it a valuable tool in biomedicine. Potential applications include early diagnosis of diseases like cancer—where subtle Raman spectral changes precede morphological alterations—and sensitive detection of biomarkers， such as proteins or metabolites， in their native tissue environments. By overcoming limitations in depth and signal strength， this approach advances the practical use of Raman microscopy in clinical and research settings.