VITAL: Value-Invariant Transformation and Alignment Learning for quantitative photoacoustic microscopy.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Photoacoustic Microscopy

Background:

• Fast optical-resolution photoacoustic microscopy (OR-PAM) is crucial for dynamic microvascular imaging.
• Conventional registration methods struggle with spatial misalignment and biased functional quantification in OR-PAM.
• Existing techniques inadequately address signal amplitude preservation during image registration.

Purpose of the Study:

• To introduce VITAL (Value-Invariant Transformation and Alignment Learning), a novel registration framework for OR-PAM.
• To address spatial misalignment and preserve photoacoustic (PA) signal amplitude during OR-PAM image registration.
• To improve the accuracy and reliability of functional quantification in dynamic microvascular imaging.

Main Methods:

• Developed a cascaded PA signal amplitude-preserving (PSAP) registration framework named VITAL.
• Integrated interpretable keypoint-based global alignment (SuperPoint, LightGlue, thin-plate spline) with topology-preserving dense refinement.
• Employed nearest-neighbor warping to prevent interpolation-induced PA signal amplitude distortion.

Main Results:

• VITAL achieved an 89.1% reduction in mean squared error (MSE) and improved normalized cross-correlation (NCC) from 0.178 to 0.821.
• The framework maintained a high PA signal-amplitude distribution fidelity rate (PFr) of 97.83%.
• Validated VITAL on *in vivo* mouse brain imaging and synthetic deformation data.

Conclusions:

• VITAL effectively separates geometric alignment from PA signal amplitude preservation for improved OR-PAM registration.
• The proposed method significantly enhances spatial accuracy and quantitative fidelity in dynamic microvascular imaging.
• VITAL supports robust downstream functional OR-PAM analysis, overcoming limitations of conventional registration techniques.