MOTION COMPENSATION FOR OPTICAL COHERENCE TOMOGRAPHY ANGIOGRAPHY BASED ON MULTI-DIMENSIONAL MATRIX PENCIL METHOD AND ADAPTIVE EN-FACE IMAGE ENHANCEMENT

Abstract

This paper proposes a two-stage algorithm combining the multi-dimensional matrix pencil method (MDMP) with En-face image enhancement to address the problem of eye motion artifacts severely degrading image quality in optical coherence tomography angiography (OCTA). In the first stage, a two-dimensional Hankel matrix pencil in the depth-time domain is constructed to jointly estimate Doppler frequency shifts and motion parameters of blood flow signals. The Centro-Hermitian property is exploited via unitary transformation to reduce computational complexity by approximately 75%, and an automatic parameter pairing mechanism is introduced to avoid mispairing problems inherent in conventional approaches. In the second stage, En-face image enhancement is applied to the MDMP-compensated retinal volume data, sequentially performing FFT difference calculation, weighted layered projection, CLAHE adaptive enhancement, column-wise median normalization for stripe removal, and FAZ region cosine gradient darkening to generate high-quality En-face vascular images. Experimental validation on a retinal OCTA dataset demonstrates that compared with multiple existing methods, the proposed method achieves a 65.1% improvement in signal-to-noise ratio (SNR), an 80.2% improvement in contrast-to-noise ratio (CNR), a structural similarity index (SSIM) of 0.94, and a 34.0% reduction in motion artifact index (MAI). The results confirm that the proposed two-stage algorithm effectively suppresses eye motion artifacts and significantly enhances OCTA En-face vascular imaging quality.